Open Access Article
This Open Access Article is licensed under a
Creative Commons Attribution 3.0 Unported Licence

Science and technology roadmap for graphene, related two-dimensional crystals, and hybrid systems

Andrea C. Ferrari *a, Francesco Bonaccorso ab, Vladimir Fal'ko c, Konstantin S. Novoselov d, Stephan Roche ef, Peter Bøggild g, Stefano Borini h, Frank H. L. Koppens i, Vincenzo Palermo j, Nicola Pugno klm, José A. Garrido n, Roman Sordan o, Alberto Bianco p, Laura Ballerini q, Maurizio Prato r, Elefterios Lidorikis s, Jani Kivioja h, Claudio Marinelli t, Tapani Ryhänen h, Alberto Morpurgo u, Jonathan N. Coleman vw, Valeria Nicolosi vwx, Luigi Colombo y, Albert Fert zaa, Mar Garcia-Hernandez ab, Adrian Bachtold i, Grégory F. Schneider ac, Francisco Guinea ab, Cees Dekker ad, Matteo Barbone a, Zhipei Sun a, Costas Galiotis aeaf, Alexander N. Grigorenko d, Gerasimos Konstantatos i, Andras Kis ag, Mikhail Katsnelson ah, Lieven Vandersypen ad, Annick Loiseau ai, Vittorio Morandi aj, Daniel Neumaier ak, Emanuele Treossi j, Vittorio Pellegrini bal, Marco Polini al, Alessandro Tredicucci al, Gareth M. Williams am, Byung Hee Hong an, Jong-Hyun Ahn ao, Jong Min Kim ap, Herbert Zirath aq, Bart J. van Wees ar, Herre van der Zant ad, Luigi Occhipinti as, Andrea Di Matteo as, Ian A. Kinloch at, Thomas Seyller au, Etienne Quesnel av, Xinliang Feng aw, Ken Teo ax, Nalin Rupesinghe ax, Pertti Hakonen ay, Simon R. T. Neil az, Quentin Tannock az, Tomas Löfwander aq and Jari Kinaret ba
aCambridge Graphene Centre, University of Cambridge, Cambridge, CB3 0FA, UK. E-mail:
bIstituto Italiano di Tecnologia, Graphene Labs, Genova, 16163, Italy
cDepartment of Physics, Lancaster University, Lancaster, LA1 4YB, UK
dSchool of Physics and Astronomy, University of Manchester, Manchester, M13 9PL, UK
eICN2-Institut Catala de Nanociencia i Nanotecnologia, Campus UAB, 08193 Bellaterra (Barcelona), Spain
fInstitució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, 08070, Spain
gCentre for Nanostructured Graphene (CNG), Department of Micro- and Nanotechnology, Technical University of Denmark, Lyngby, 2800, Denmark
hNokia Technologies, Broers Building, Cambridge, CB3 0FA, UK
iInstitut de Ciències Fotòniques (ICFO), Castelldefels (Barcelona), 08860, Spain
jCNR-Istituto per la Sintesi Organica e la Fotoreattività, Bologna, 40129, Italy
kDipartimento di Ingegneria Civile, Ambientale e Meccanica, Università di Trento, Trento, 38123, Italy
lFondazione Bruno Kessler, Trento, 38122, Italy
mSchool of Engineering and Materials Science, Queen Mary University of London, London, E1 4NS, UK
nWalter Schottky Institut, Technische Universität München, Garching, 85748, Germany
oL-NESS, Dipartimento di Fisica, Politecnico di Milano, Como, 22100, Italy
pCNRS, Institut de Biologie Moléculaire et Cellulaire, Immunopathologie et Chimie Thérapeutique, 67000 Strasbourg, France
qCentre for Neuroscience (BRAIN) and Dipartimento di Scienze della Vita, Università di Trieste, Trieste, 34127, Italy
rDipartimento di Scienze Farmaceutiche, Università di Trieste, Trieste, 34127, Italy
sDepartment of Materials Science and Engineering, University of Ioannina, Ioannina, 45110, Greece
tApplied Graphene Materials, The Wilton Centre, Redcar, Cleveland, TS10 4RF, UK
uDépartement de Physique de la Matière Condensée, Université de Genève, Geneva, 1205, Switzerland
vSchool of Physics, Trinity College, Dublin, D2 Dublin, Ireland
wCentre for Research on Adaptive Nanostructures and Nanodevices (CRANN), Trinity College, Dublin, D2 Dublin, Ireland
xSchool of Chemistry, Trinity College, Dublin, D2 Dublin, Ireland
yTexas Instruments Incorporated, Dallas, TX, USA
zUnité Mixte de Physique CNRS/Thales, Palaiseau, 91767, France
aaUniversité de Paris-Sud, Orsay, 91405, France
abInstituto de Ciencia de Materiales de Madrid, CSIC, Madrid, 28049, Spain
acLeiden University, Faculty of Science, Leiden Institute of Chemistry, Leiden 2333 CC, The Netherlands
adKavli Institute of Nanoscience, Delft University of Technology, Delft, 2628 CJ, The Netherlands
aeInstitute of Chemical Engineering Sciences (ICE-HT/FORTH), Rio, 26504, Greece
afDepartment of Chemical Engineering, University of Patras, Rio, 26504, Greece
agElectrical Engineering Institute, Ecole Polytechique Fédérale de Lausanne, Lausanne, 1015, Switzerland
ahInstitute for Molecules and Materials, Radboud University Nijmegen, Nijmegen, 6525 AJ, The Netherlands
aiLaboratoire d'Etude des Microstructures (LEM), ONERA-CNRS, Chatillon, 92322, France
ajCNR-Istituto per la Microelettronica e i Microsistemi, Bologna, 40129, Italy
akAdvanced Microelectronic Centre Aachen, AMO GmbH, Aachen, 52074, Germany
alNEST, Istituto Nanoscienze-CNR and Scuola Normale Superiore, Pisa, 56127, Italy
amAirbus UK Ltd, Broughton, CH4 0DR, UK
anDepartment of Chemistry, Seoul National University, Seoul, 151-747, South Korea
aoSchool of Electrical & Electronic Engineering, Yonsei University, Seoul, 120-749, South Korea
apDepartment of Engineering Science, University of Oxford, Oxford, OX1 3PJ, UK
aqDepartment of Microtechnology and Nanoscience, Chalmers University of Technology, Gothenburg, 412 96, Sweden
arPhysics of Nanodevices, Zernike Institute for Advanced Materials, University of Groningen, Groningen, 9747 AG, The Netherlands
asSTMicroelectronics, Arzano (Naples), 80022, Italy
atMaterials Science Centre, School of Materials, University of Manchester, Manchester, M13 9PL, UK
auInstitut für Physik, Technische Universität Chemnitz, Chemnitz, 09126, Germany
avInstitut LITEN, CEA LITEN, Grenoble Cedex 9, 38054 9, France
awMax-Planck-Institut für Polymerforschung, Mainz, 55128, Germany
axAixtron Ltd., Cambridge, UK
ayAalto University, FI-00076, Finland
azCambridgeIP, Cambridge, CB2 1SJ, UK
baDepartment of Applied Physics, Chalmers University of Technology, Gothenburg, 412 96, Sweden

Received 24th March 2014 , Accepted 12th September 2014

First published on 22nd September 2014

We present the science and technology roadmap for graphene, related two-dimensional crystals, and hybrid systems, targeting an evolution in technology, that might lead to impacts and benefits reaching into most areas of society. This roadmap was developed within the framework of the European Graphene Flagship and outlines the main targets and research areas as best understood at the start of this ambitious project. We provide an overview of the key aspects of graphene and related materials (GRMs), ranging from fundamental research challenges to a variety of applications in a large number of sectors, highlighting the steps necessary to take GRMs from a state of raw potential to a point where they might revolutionize multiple industries. We also define an extensive list of acronyms in an effort to standardize the nomenclature in this emerging field.

List of acronyms

0dZero dimensional
1LGSingle layer graphene
1L-MoS2Single layer molybdenum disulfide
2DOvertone of Raman D peak
2DEGTwo-dimensional electron gas
2LGTwo-layer graphene
2L-MoS2Two-layer molybdenum disulfide
3LGTri-layer graphene
3L-MoS2Tri-layer molybdenum disulfide
α Absorption coefficient
AabsAbsorbed photon fraction
a-CAmorphous carbon
a-C:HHydrogenated amorphous carbon
A/DAnalog to digital
AFMAtomic force microscopy
AGArtificial graphene
ALDAtomic layer deposition
ALEAtomic Layer Epitaxy
Al2O3Aluminium oxide
APDAvalanche photodiode
ARPESAngle-resolved photoemission spectroscopy
AVVoltage Gain
AVAlternating voltage
BCBlock copolymer
BGIBroken Galilean invariance
BISFETBilayer pseudospin field effect transistor
BLGBi-layer graphene
BMIMPF61-Butyl-3-methylimidazolium hexafluorophosphate
Bi2Se3Bismuth selenide
Bi2Te3Bismuth telluride
BNBoron nitride
c Concentration
CBEChemical beam epitaxy
c–BNCubic–boron nitride
CdSCadmium sulfide
CdSeCadmium selenide
CECounter electrode
CERNEuropean organization for nuclear research
C-faceCarbon face
CFRCCeramic fiber reinforced concrete
CINColloidal inorganic nanocrystal
CMGChemically modified graphene
CMOSComplementary metal oxide semiconductor
CNPCarbon nanoparticle
CNTCarbon nanotube
CNWCarbon nanowall
CSGCovalently bound stretched graphene
CVDChemical vapour deposition
CVFFConsistent valence force field
γ Surface tension
DRaman D peak
D*Specific detectivity
DCDirect current
DDADiscrete dipole approximation
DFPTDensity functional perturbation theory
DFTDensity functional theory
DGMDensity gradient medium
DGUDensity gradient ultracentrifugation
DLCDiamond-like carbon
DNADeoxyribonucleic acid
DoEDepartment of Energy-USA
DSSCDye-sensitized solar cell
D-WDMDense wavelength division multiplexer
η abs Light absorption efficiency
EELSElectron energy loss spectroscopy
EDFAErbium-doped fibre amplifier
EDFLErbium-doped fiber laser
EDLCElectrochemical double layer capacitor
EGFPEnhanced green fluorescence protein
EHFExtremely high frequency
EMIElectromagnetic interference
EphPhoton energy
EPOEuropean patent office
EPRElectric paramagnetic resonance
EPREEnhanced permeability and retention effect
EQEExternal quantum efficiency
ERMDEmerging research memory device
η trans charge transfer efficiency
FDTDFinite-difference time-domain
FEField emission
FETField effect transistor
FFFill factor
FGFunctionalized graphene
FIRFar infrared
FLGFew-layer graphene
FPAFocal plane array
FQHEFractional quantum hall effect
FTOFluorine-doped tin oxide
Z Field enhancement factor
GRaman G peak
GaAsGallium arsenide
GaNGallium nitride
GBGrain boundary
g d Output conductance
GDGravimetric density
GFETGraphene field-effect transistor
GFRCGlass fiber reinforced concrete
GHzGiga Hertz
GICGraphite intercalation compound
g m Transconductance
GNRGraphene nanoribbon
GOGraphene oxide
GOQDGraphene oxide quantum dot
GOTCFGraphene oxide transparent conductive film
GPDGraphene-based photodetector
GphPhotocondutive gain
GQDGraphene quantum dot
GRMsGraphene and related materials
GROGraphene ring oscillator
GSAGraphene saturable absorber
GSAMGraphene saturable absorber mirror
GTCEGraphene transparent conductive electrode
GTCFGraphene transparent conductive film
GWCGraphene-enabled wireless communications
h-BNHexagonal boron nitride
HCIHuman computer interaction
HClHydrochloric acid
hcpHexagonal closed packed
HeLaHenrietta Lacks
HEMTHigh-electron mobility transistor
HFHigh frequency
HfO2Hafnium oxide
HMIH1-Hexyl-3-methylimidazolium hexafluorophosphate
HOMOHighest occupied molecular orbital
HPCHigh performance computing
HRTEMHigh resolution transmission electron microscope
HSCHybrid supercapacitor
κ Thermal conductivity
[scr K, script letter K] Dielectric constant
KOHPotassium hydroxide
ICPInductively coupled plasma
ICTInformation and communications technology
ILIonic liquid
In2O3Indium oxide
ION/IOFFOn/Off current ratio
InPIndium phosphide
InSbIndium antimonide
IPIntellectual Property
IQEInternal quantum efficiency
ITOIndium tin oxide
ITRSInternational technology roadmap for semiconductors
LaB6Lanthanum hexaboride
LCLiquid crystal
LEDLight emitting diode
LEEDLow-energy electron diffraction
LIBLithium ion battery
LMLayered material
LNALow-noise amplifier
LOLocal oscillator
LPCVDLow pressure chemical vapour deposition
LPELiquid phase exfoliation
LSPRLocalized surface plasmon resonance
LUMOLowest unoccupied molecular orbital
μ Carrier mobility
m Staging index
MACMedium access control
MBEMolecular beam epitaxy
MCMicromechanical cleavage
MCCMesoporous carbon capsule
MC-SLGMechanically cleaved-single layer graphene
MDMolecular dynamics
MEGrOMicrowave expanded graphite oxide
MEMSMicro electro-mechanical systems
MESFETMetal-semiconductor field effect transistor
MgOMagnesium oxide
MLGMultilayer graphene
MnO2Manganese dioxide
MNPMetallic nanoparticle
MOCVDMetal–organic chemical vapour deposition
MoS2Molybdenum disulfide
MoSe2Molybdenum diselenide
MOSFETMetal–oxide-semiconductor field-effect transistor
MoTe2Molybdenum ditelluride
MRAMMagnetoresistive random-access memory
MSSCMeso-super-structured solar cell
m-SWNTMetallic single wall carbon nanotube
MTJMagnetic tunnel junction
MWCVDMicro wave chemical vapour deposition
ν Viscosity
n carrier density
N Number of layers
NaOHSodium hydroxide
NbSe2Niobium diselenide
NEMNano electromechanical
NEMSNano electromechanical systems
NEPNoise equivalent power
NIRNear infrared
NiTe2Nickel ditelluride
NLGN-layer graphene
NMP N-Methylpyrrolidone
NOEMSNano optoelectromechanical systems
OASOptical absorption spectroscopy
OLEDOrganic light-emitting diode
O(N)Computations scaling linearly with the number of atoms
O(N3)Computations scaling with the cube of the number of atoms
OPVOrganic photo-voltaic
OTOptical tweezers
ϕ abs absorbed photon flux
ϕ in incoming photon flux
P3HTPoly-3-hexyl thiophene
PA6Polyamide6 (also known as Nylon6)
PAHsPoly-aromatic hydrocarbons
PbSLead sulphide
PCBMPhenyl-C61-butyric acid methyl ester
PCFPhotonic crustal fiber
PCTPatent cooperation treaty
PECVDPlasma enhanced chemical vapour deposition
PEDOTpoly(3,4 ethylenedioxythiophene)
PEGPolyethylene glycol
PEOPolyethylene glycol
PENPolyethylene naphthalate
PETPolyethylene terephthalate
PDLCPolymer dispersed liquid crystal
pDNAPlasmid deoxyribonucleic acid
PinIncident power
PIL:RGOpoly(ionic liquid)-modified reduced graphene oxide
PMFPolarization-maintaining fiber
PMTPhotomultiplier tube
PNFPolyaniline Nanofiber
PPCPoly-propylene carbonate
PRACEPartnership for advanced computing in Europe
PSSPolystyrene sulphonate
PTCDAPerylene-3,4,9,10-tetracarboxylic dianhydride
PVCPolyvinyl chloride
PVDPhysical vapour deposition
PVDFPolyvinyl difluoride
qElectron charge
QDQuantum dot
QEQuantum efficiency
QHEQuantum Hall effect
QMCQuantum Monte Carlo
QM/MMQuantum mechanics/molecular mechanics
QPCQuantum point contact
QSHQuantum spin Hall
QSHEQuantum spin Hall effect
ρ Density
RCARadio corporation of America
R&DResearch and development
RESReticuloendothelial system
RFRadio frequency
RGOReduced graphene oxide
RIXSResonant inelastic X-ray scattering
RNARibonucleic acid
RORing oscillator
ROICRead-out integrated circuits
ROSReactive oxygen species
RPARandom phase approximation
R2RRoll to roll
R s Sheet resistance
RTRoom temperature
RZSRate zonal separation
ς Surface energy
σ Electrical conductivity
S Seebeck coefficient
SASaturable absorber
SAMSelf-assembled monolayer
SbF5Antimony pentafluoride
SBSSedimentation based-separation
SCSodium cholate
SCMScanning catalyst microscope
SDBSSodium dodecyl benzene sulfonate
SDCSodium deoxycholate
SDSSodium dodecyl sulphate
SEISolid electrolyte interphase
SEIRASurface-enhanced infrared Raman absorption
SESAMSemiconductor saturable absorber mirror
SERSSurface enhanced Raman spectroscopy
SETSingle electron transistor
SEYSecondary electron yield
SHESpin Hall effect
SiCSilicon carbide
SiO2Silicon dioxide
SiRNASmall interfering ribonucleic acid
SLGSingle layer graphene
SMMAStyrene methyl methacrylate
SnO2Tin oxide
SOSpin orbit
SPPSurface plasmon polariton
SPRSurface plasmon resonance
SQDSemiconductor quantum dot
SQUIDSuperconducting quantum interference device
SSASpecific surface area
ssDNASingle-stranded DNA
s-SNOMScattering-type near-field microscopy
STScience and technology
STEMScanning transmission electron microscopy
STMScanning tunnelling microscopy
STRScience and technology roadmap
STSScanning tunnelling spectroscopy
STTSpin transfer torque
SWIRShort wavelength infrared
SWNTSingle wall carbon nanotube
s-SWNTSemiconducting single wall carbon nanotube
T Temperature
ta-CTetrahedral amorphous carbon
ta-C:HHydrogenated tetrahedral amorphous carbon
ta-C:NNitrogenated tetrahedral amorphous carbon
TaSe2Tantalum selenide
TcCritical temperature
TCTransparent conductor
TCETransparent conductor electrode
TCFTransparent conductor film
TDDFTTime-dependent density functional theory
TeElectron temperature
TEMTransmission electron microscope
TGAThermo-gravimetric analysis
TITopological insulator
TiO2Titanium dioxide
TLGTrilayer graphene
TMTransverse magnetic
TMDTransition metal dichalcogenide
TMOTransition metal oxide
TPUThermoplastic polyurethane
TRLTechnology readiness level
τ tr Charge lifetime
τ transit Drift transit time
UHVUltra-high vacuum
UIUser interface
ULGAUltralight and highly compressible graphene aerogels
US PTOUnited States patent and trademark office
VCDVibrational circular dichroism
VDVolumetric density
vdWvan der Waals
vHvan Hove singularity
VRHVariable range hopping
WDMWavelength division multiplexer
WIPOWorld intellectual property organization
WNSNWireless nanosensor network
WO2Tungsten dioxide
WSTungsten sulfide
WS2Tungsten disulfide
XASX-ray absorption spectroscopy
XMCDX-ray magnetic circular dichroism
XPSX-ray photoelectron spectroscopy
Y M Yield by SLG percentage
Y W Yield by weight
Y WM Yield by SLG weight
ζ Field enhancement factor
ZnOZinc oxide
ZnSZinc sulfide
ZnSeZinc selenide

1. Introduction

The primary objective of this roadmap is to guide the community towards the development of products based on graphene, related two dimensional (2d) crystals and hybrid systems. For simplicity we will refer to this new materials platform as graphene and related materials and use the acronym GRM. These have a combination of properties that could make them key enablers for many applications, generating new products that cannot (or may be difficult to) be obtained with current technologies or materials. The creation of new disruptive technologies based on GRMs is conditional to reaching a variety of objectives and overcoming several challenges throughout the value chain, ranging from materials to components and systems.

The main scientific and technological objectives are:

A) Material technologies

 ○ Identification of new layered materials (LMs) and assessment of their potential.

 ○ Reliable, reproducible, sustainable and safe, large scale production of GRMs, satisfying the specific needs of different application areas.

B) Component technologies

 ○ Identification of new device concepts enabled by GRMs.

 ○ Identification of component technologies that utilize GRMs.

 ○ Electronic technologies, comprising high frequency electronics, optoelectronics, spintronics and sensors.

C) Systems integration

 ○ Route to bring components and structures based on GRMs to systems capable of providing new functionalities and open new application areas.

 ○ New concepts for integrating GRMs in existing technology platforms.

 ○ Integration routes for nanocomposites, flexible electronics and energy applications.

Our science and technology roadmap (STR) outlines the principal routes to develop the GRM knowledge base and the means of production and development of new devices, with the final aim of integrating GRMs into systems. In the Information and Communications Technology (ICT) area, the STR focuses on technology that will enable new applications, such as the Morph concept1 (Fig. 1a), which exploits the electrical, optical and mechanical properties of GRMs to realize new types of personal communicators. In the domain of physical communication, the STR targets several key technologies in energy production and storage, as well as new functional light-weight composites. These are to be integrated in transportation systems, such as new airplanes, buses, cars (as illustrated by the SmartForVision concept electric car,2Fig. 1b). The STR also considers areas such as Health and Energy. By exploiting the GRM's unique electrical and optical properties, the STR will highlight the directions towards the development of novel systems for information processing and communications.

image file: c4nr01600a-f1.tif
Fig. 1 Morph1 (left) and SmartForVision2 (right) are examples of visionary applications where GRMs’ unique properties might be combined to enable new products.

The STR is divided in 11 thematic chapters, summarized in Fig. 2. Each of them comprises a dedicated timeline. A final chapter presents two overall summary roadmaps.

image file: c4nr01600a-f2.tif
Fig. 2 Symbols associated with each theme. In the document, the symbol image file: c4nr01600a-u1.tif is associated/replaced by the symbol image file: c4nr01600a-u2.tif when we refer to industrial/large scale production.

The present STR may not be fully complete, leaving out some of the most recent and rapidly evolving areas. We plan to present regular updates over the next 10 years to keep abreast with the latest developments in GRM science and technology. These include charge-based high speed electronic devices, as well as non-charge-based devices (e.g. spintronic devices) with novel functionalities. A key area is advanced methods to produce GRMs, combining structural functions with embedded electronics in an environmentally sustainable manner. The STR extends beyond mainstream ICT to incorporate novel sensor applications and composites that take advantage of the GRMs chemical, biological and mechanical properties. Beyond ICT, the STR reaches out to several related areas. Graphene's high electrical conductivity, σ, and large surface area per unit mass make it an interesting material for energy storage, e.g. in advanced batteries and supercapacitors. These could have a large impact on portable electronics and other key areas, such as electric cars. The prospect of rapidly chargeable lightweight batteries would give environmentally friendly transportation a push and advance the large scale implementation of electric cars as a key component in urban and suburban transport. Strong and lightweight composites would also allow us to build new cars, airplanes and other structures using less material and energy, and contribute directly to a more sustainable world, see Fig. 3.

image file: c4nr01600a-f3.tif
Fig. 3 Overview of Applications of Graphene in different sectors ranging from conductive ink to chemical sensors, light emitting devices, composites, energy, touch panels and high frequency electronics.

1.1. Graphene-based disruptive technologies: overview

Technologies, and our economy in general, usually advance either by incremental developments (e.g. scaling the size and number of transistors on a chip) or by quantum leaps (transition from vacuum tubes to semiconductor technologies). Disruptive technologies, behind such revolutions, are usually characterised by universal, versatile applications, which change many aspects of our lives simultaneously, penetrating every corner of our existence. In order to become disruptive, a new technology needs to offer not incremental, but orders of magnitude improvements. Moreover, the more universal the technology, the better chances it has for broad base success. This can be summarized by the “Lemma of New Technology”, proposed by Herbert Kroemer, who received the Nobel Prize in Physics in 2000 for basic work in ICT: “The principal applications of any sufficiently new and innovative technology always have been – and will continue to be – applications created by that technology”.3 Graphene is no exception to this lemma. Does graphene have a chance to become the next disruptive technology? Can graphene be the material of the 21st century?

In terms of its properties, it certainly has potential. The 2010 Nobel Prize in Physics already acknowledged the profound novelty of the physical properties that can be observed in graphene: different physics applies, compared with other electronic materials, such as common semiconductors. Consequently, a plethora of outstanding properties have arisen from this material. Many are unique and superior to those of other materials. More importantly, such combination of properties cannot be found in any other material or material system. So, it is not a question of if, but a question of how many applications will graphene be used for, and how pervasive will it become. There are indeed many examples of “wonder” materials that have not yet lived up to expectations, nor delivered the promised revolution, while more “ordinary” ones are now pervasively used. Are the properties of graphene so unique to overshadow the unavoidable inconveniences of switching to a new technology, a process usually accompanied by large research and development (R&D) and capital investments? The advancing R&D activity on GRMs has already shown a significant development aimed at making GRMs suitable for industrial applications.

The production of graphene is one striking example of rapid development, with progress from random generation of micro-flakes in the laboratory5 to large-scale,6 roll-to-roll (R2R) processing of graphene sheets of sizes approaching the metre-scale7 (Fig. 4).

image file: c4nr01600a-f4.tif
Fig. 4 Rapid evolution of graphene production: from microscale flakes4 to roll-to-roll processing.7

It is reasonable to expect a rapid clearing of further technological hurdles towards the development of a GRM-based industry in the coming years (Fig. 5).

image file: c4nr01600a-f5.tif
Fig. 5 Towards GRM-based products.

Therefore, in spite of the inherent novelty associated with GRMs and the lack of maturity of GRM technology, an initial roadmap can be envisaged, including short-term milestones, and some medium- to long-term targets, less detailed, but potentially more disruptive. This should guide the expected transition towards a technological platform underpinned by GRMs, with opportunities in many fields and benefits to society as a whole.

1.1.1. Opportunities. GRMs are expected to have a major impact in several technological fields (see Table 1), due to the new applications enabled by their properties. E.g., potential electronic applications include high-frequency devices, touch screens, flexible and wearable devices, as well as ultrasensitive sensors, nano- electromechanical systems (NEMS), super-dense data storage, photonic devices, etc. In the energy field, applications include batteries and supercapacitors to store and transport electrical power, and solar cells. However, in the medium term, some of graphene's most appealing potential lies in its ability to transmit light as well as electricity, offering improved performance for light emitting diodes (LEDs), flexible touch screens, photodetectors, and ultrafast lasers.
image file: c4nr01600a-u11.tif

The upscaling of GRM production is steadily progressing, and challenges remain when it comes to maintaining the properties and performance upon up-scaling, which includes mass production for material/energy-oriented applications and wafer-scale integration for device/ICTs-oriented applications. Nevertheless, GRMs technology is expected to provide opportunities for the development of a novel platform, contributing to key technological fields with important social and economic impacts. The definition of “quality” of a GRM cannot be given in absolute terms, but strictly depends on the applications. E.g. the “quality” of graphene needed for high performance electronics is “the opposite” of that required for batteries or supercapacitors, in that the latter work better with materials having defects, voids and cavities, while the former require defect free, and flat material. This will be a challenge for standardization, since the materials properties will have to be defined in relation to a variety of possible applications. New opportunities for electronics. The introduction of more functions in integrated electronic systems will enable applications in domotics (i.e. home automation by means of distributed sensors, actuators and controllers), environmental control, and office automation to meet the social request for better safety, health and comfort. An increase in automation should also consider the aging population and people at work, and the need of adequate facilities. Sensors or metrological devices based on GRMs can further extend functionalities of hybrid circuits. Three dimensional (3d) integration of GRMs-based devices may be conceivable in a Si flow, and could be the solution for low cost chips with extended functionalities.

Graphene has many record properties, see Fig. 6. It is transparent like (or better than) plastic, but conducts heat and electricity better than any metal, it is an elastic film, behaves as an impermeable membrane, and it is chemically inert and stable. Thus it seems ideal as the next generation transparent conductor. There is a real need to find a substitute for indium tin oxide (ITO) in the manufacturing of various types of displays and touch screens, due to the brittleness of indium that makes it difficult to use them when flexibility is a requirement.8 Graphene is an ideal candidate for such a task.9 Thus, coupled with carbon's abundance, this presents a more sustainable alternative to ITO. Prototypes of graphene-based displays have been produced7 and commercial products seem imminent.10

image file: c4nr01600a-f6.tif
Fig. 6 Graphene properties and application areas.

In 2010, the first R2R production of 30-inch graphene transparent conductors (TC), with low sheet resistance (Rs) and 90% transmittance (Tr), competitive with commercial transparent electrodes, such as ITO, was reported.7 Graphene electrodes have been incorporated into fully functional touch-screens capable of withstanding high strain.10 Thus, one can envision the development of flexible, portable and reconfigurable electronics, such as the MORPH concept1 (Fig. 1 and 7).

image file: c4nr01600a-f7.tif
Fig. 7 NOKIA Morph:1 the future mobile device will act as a gateway. It will connect users to local environment, as well as the global internet. It is an attentive device that shapes according to the context. It can change its form from rigid to flexible and stretchable.1

New horizons have opened with the demonstration of high-speed graphene circuits11 offering high-bandwidth, which might impact future low-cost smart phones and displays.

Complementary metal oxide semiconductor (CMOS) technology, as currently used in integrated circuits, is rapidly approaching the limits of downsizing transistors,12 and graphene is considered a possible candidate for post-Si electronics by the International Technology Roadmap for Semiconductors (ITRS).12 However, a graphene-based low power device meeting all of the requirements of CMOS technology has not been demonstrated yet. The technology needed to produce graphene circuits is still in its infancy, and growth of large area films with good electrical properties on flat dielectric surfaces has not yet been demonstrated. Novel architectures,13,14 not necessarily based on graphene ribbons,15 need to be developed.

In 2011 ref. 11 reported the first wafer-scale graphene circuit (broadband frequency mixer) in which all components, including graphene field-effect transistors (GFETs) and inductors, were integrated on a single SiC wafer. The circuit operated as a broadband Radio Frequency (RF) mixer at frequencies up to 10 GHz, with thermal stability and little reduction in performance (less than one decibel) in the temperature (T) range 300–400 K. This suggests that graphene devices with complex functionality and performance may be achieved.

Being just one atom thick, graphene appears as a suitable candidate to eventually realize a new generation of flexible electronic devices.14 Electronics on plastics or paper is low cost.16,17 It will offer the possibility to introduce more information on goods used on a daily basis, e.g. on food for safety and health, as well as on many other products. Bar codes may not be able to store all the required information. Magnetic strips or stand-alone memories do not offer the same opportunities as active electronics interacting in a wireless network. The possibility to develop passive components in GRMs (resistors, capacitors, antennas) as well as diodes (Schottky) or simple FETs, and the rapid growth of technology in this direction may enable RF flexible circuits in a wireless networked environment.

Thin and flexible GRMs-based electronic components might be obtained and modularly integrated, and thin portable devices might be assembled and distributed. Graphene can withstand mechanical deformation18 and can be folded without breaking.18 Such a feature provides a way to tune the electronic properties, through so-called “strain engineering”19 of the electronic band structure. Foldable devices can be imagined, together with a wealth of new device form factors, which could enable innovative concepts of integration and distribution.

By enabling flexible electronics, GRMs will allow the use of the existing knowledge base and infrastructures of various organizations working on organic electronics (organic LEDS as used in displays, conductive polymers, plastics, printable electronics), providing a synergistic framework for collecting and underpinning many distributed technical competences. New energy solutions. GRMs could bring new solutions to the current challenges related to energy generation and storage, first in nano-enhanced products, then in new nano-enabled products. GRMs-based systems for energy production (photovoltaics, PV, fuel cells), energy storage (supercapacitors, batteries, and hydrogen storage) may be developed via relevant proof of concept demonstrators that will progress towards the targeted technology readiness levels (TRLs) required for industrial adoption. TRLs are used to assess the maturity of technologies during their development. The commonly used NASA scale,20,21 is shown in Fig. 8: 1. Basic principles observed and reported; 2. Technology concept and/or application formulated; 3. Analytical and experimental critical function and/or characteristic proof of concept; 4. Component and/or breadboard validation in laboratory environment; 5. Component and/or breadboard validation in relevant environment; 6. System/subsystem model or prototype demonstration in a relevant environment; 7. System prototype demonstration in an operational environment; 8. Actual system completed and qualified through test and demonstration. 9. Actual system proven through successful operations.
image file: c4nr01600a-f8.tif
Fig. 8 TRL definitions, adapted from ref. 21.

Furthermore, graphene technology may provide new power management solutions, key to allow efficient and safe use of energy. To date in Europe nearly the 60% of the energy is electrical (lighting, electronics, telecommunications, motor control).22 Of the remaining 40%, nearly all is used for transportation.22 New technologies and materials: towards a novel technological platform. GRMs may favour not only an improvement of existing technologies, such as electronics and optoelectronics, but may also enable the emergence of new technologies, currently hampered by intrinsic limitations. The GRMs' properties, with a qualitatively different physics with respect to the other commonly used materials, may enable technological concepts, thus far only theoretically possible, but not practically developed.

One example is that of spintronics,23 an emerging technology that exploits the spin rather than the charge of electrons as the degree of freedom for carrying information,24 with the primary advantage of consuming less power per computation.25 Although one spintronic effect – namely, giant magnetoresistance26 – is already a fundamental working principle in hard disk technology,27 the use of spintronic devices as a replacement for CMOS has not been realized yet. Scientific papers have highlighted graphene properties that are suitable for the development of spintronic devices,28–30 and many groups are now pursuing this.

Radically new technologies could be enabled by graphene, such as the so-called “valleytronics”,31 which exploits the peculiar “isospin”31 of charge carriers in graphene as a degree of freedom for carrying information. Further, there are some still not experimentally proven theoretical predictions, such as a “chiral superconductivity”,32 which may lead to completely new applications.

Taking just these few examples into account, we expect that the development of some new applications based on the salient properties of GRMs might happen in the coming years.

Graphene is also an ideal candidate for engineering new materials, and many examples have already been realised.33–36 The “all-surface” nature of graphene offers the opportunity to tailor its properties by surface treatments (e.g. by chemical functionalization33). E.g., graphene has been converted into a band-gap semiconductor (hydrogenated graphene, or “graphane”33) or into an insulator (fluorinated graphene, or “fluorographene”34). In addition, graphene flakes can be placed in dispersions.35 These retain many of its outstanding properties, and can be used for the realisation of composite materials (e.g. by embedding in a polymeric matrix36,37) with improved performance.35–37

Graphene is not only important for its own properties, but also because it is the paradigm for a new class of materials, which is likely to grow following the rise of graphene technology. Some examples have already been reported, such as hexagonal boron nitride (h-BN)5,38 and molybdenite monolayers5,38,39 The crystal structure of the latter was studied since 1923 by Dickinson and Pauling,40 with studies extended to a few layers in the sixties (a possible observation of monolayer MoS2 reported in the pioneering work of Frindt in Cambridge in 1963)41,42 and a definite identification of monolayer MoS2 in 1986.39 The assembly of such 2d crystals, i.e. by stacking different atomic planes (heterostructures43), or by varying the stacking order of homogeneous atomic planes,44 provides a rich toolset for new, customised materials. We expect that the lessons learnt developing graphene science and technology will drive the manufacturing of many other innovative materials.

At present, the realisation of an electronic device (such as, e.g., a mobile phone) requires the assembly of a variety of components obtained by many different technologies. GRMs, by including many properties, may offer the opportunity to build a comprehensive technological platform for different device components, including transistors, batteries, optoelectronic components, detectors, photovoltaic cells, photodetectors, ultrafast lasers, bio- and physicochemical sensors, etc. Such a change in the paradigm of device manufacturing may open big opportunities for the development of a new industry.

1.2. Scientific output

GRM research is an example of an emerging translational nanotechnology, where discoveries in laboratories are transferred to applications. This is evidenced, in part, by the rise in patenting activity since 2007 by corporations around the world.45 The concept of translational technology is typically associated with biomedicine,46 where it is a well-established link between basic research and clinical studies, but the principle can be applied more generally. A striking example is giant magnetoresistance,47 that moved from an academic discovery to a dominant information storage technology in a few years.48 Similarly, GRMs have the potential to make a profound impact: Integrating GRMs components with Si-based electronics, and gradually replacing Si in some applications, allows not only substantial performance improvements but, more importantly, new applications.

Carbon has been the driving force behind several technological revolutions: in the 19th century, energy production by burning carbon was integral to the industrial revolution;49 in the 20th century, carbon-based plastics revolutionized the manufacturing industry;50 in the 21st century, graphitic carbon might be a key component in a third technological revolution.

The growth of publications on GRMs is shown in Fig. 9, with no sign of slowing down. The reasons for the growth of research on GRMs are manifold. First, graphene is a material with a unique set properties. Either separately or in combinations, these can be exploited in many areas of research and applications; new possibilities are being recognized all the time as the science of GRMs progresses. Second, graphene Science and Technology (ST) relies on one of the most abundant materials on earth,51 carbon. It is an inherently sustainable and economical technology. Thirdly, graphene is a planar material and, as such, compatible with the established production technologies in ICT, and integrable with conventional materials such as Si. Combined, these premises give realistic promise of creating a new, more powerful and versatile, sustainable and economically viable technology platform. As a result, graphene research has already emerged as the top research front in materials science.52 However, due to the unique structure of graphene, many of the possibilities it offers are still poorly understood, and their analysis requires highly sophisticated methods; To quote the Nobel Laureate Frank Wilczek: «graphene is probably the only system where ideas from quantum field theory can lead to patentable innovations».46

image file: c4nr01600a-f9.tif
Fig. 9 Publications on graphene from 2000 to Aug. 2014 (thus, well over 18000 are expected by end 2014). Source ISI Web of Science (search: Topic = Graphene). Publications on graphene prior to 2000 are not plotted.
1.2.1. Intellectual property landscape analysis. In the graphene area, there has been a particularly rapid increase in patent activity from around 2007.45 Much of this is driven by patent applications made by major corporations and universities in South Korea and USA.53 Additionally, a high level of graphene patent activity in China is also observed.54 These features have led some commentators to conclude that graphene innovations arising in Europe are being mainly exploited elsewhere.55 Nonetheless, an analysis of the Intellectual Property (IP) provides evidence that Europe already has a significant foothold in the graphene patent landscape and significant opportunities to secure future value. As the underlying graphene technology space develops, and the GRM patent landscape matures, re-distribution of the patent landscape seems inevitable and Europe is well positioned to benefit from patent-based commercialisation of GRM research.

Overall, the graphene patent landscape is growing rapidly and already resembles that of sub-segments of the semiconductor and biotechnology industries,56 which experience high levels of patent activity. The patent strategies of the businesses active in such sub-sectors frequently include ‘portfolio maximization’56 and ‘portfolio optimization’56 strategies, and the sub-sectors experience the development of what commentators term ‘patent thickets’56, or multiple overlapping granted patent rights.56 A range of policies, regulatory and business strategies have been developed to limit such patent practices.57 In such circumstances, accurate patent landscaping may provide critical information to policy-makers, investors and individual industry participants, underpinning the development of sound policies, business strategies and research commercialisation plans.

The analysis of the top graphene patent owners (patent assignees) and their patent applications, illustrates the broad relevance of graphene to diverse industry sectors, such as automotive, computing and industrial chemicals.58 The uses of patents between and within these industry sectors and over time can vary widely, adding to the navigational challenges that face readers of even the most accurate graphene IP maps.

Understanding and correctly navigating the rapidly growing patent landscape will be crucial to those who seek to secure future value from GRM research. Patents may be particularly important to the realisation of future commercial value, as patents are a form of IP important to the business models and business practices observed in many of the technology sectors in which GRM research is and will be deployed.56

The IP analysis and discussion in section 1.2.2 highlights the disparity between graphene-related scientific production (represented by publications), see Fig. 9, 10, and graphene-related patent applications (associated with technical exploitation), providing additional evidence of the need for a large scale, concentrated action to bring together leading players in academia (who are, broadly, responsible for scientific production) and industrial leaders (who are, broadly, responsible for patent applications).

image file: c4nr01600a-f10.tif
Fig. 10 Geographical distribution of scientific papers on graphene as of December 2013.
1.2.2. Graphene IP landscape analysis. Fig. 11 indicates that the global IP activity around graphene has surged since 2007, mimicking the trend in research described in section 1.2 and evidence perhaps that research investment worldwide is fuelling rapid growth in graphene technology. Interestingly, IP activity around graphene predates 2004, and patent filings can be found around processes which would have resulted in graphene production from as early as 1896: see, e.g.ref. 59.
image file: c4nr01600a-f11.tif
Fig. 11 Patent applications on graphene as a function of application year. Note: patents remain unpublished for up to 18 months from their filing. Accordingly, 2013 and 2014 are under-represented. Data updated as of July 2014.

The patent space prior to 2006 is dominated by US research institutions and start ups, with a significant volume of filings starting after 2006. The surge in filings from 2007 has been driven heavily by innovations from South Korean multinationals, especially Samsung, as well as research institutes with Samsung connections.

A detailed review of the patent dataset reveals that patents have been filed for a very diverse range of applications including characterization,60 polymer composites,61 transparent displays,62 transistors,63 capacitors,64 solar cells,65 biosensors,66 conductive inks,67–69 windows,70 saturable absorbers,71 photodetectors,72 tennis rackets.73 However, overall, the graphene patent space comprises patent filings in two main sectors: synthesis (e.g. production of graphene by chemical vapour deposition – CVD, exfoliation, etc.,) and electronics (e.g. use of graphene for displays, transistors and computer chips), each ∼30% of the total space, as for Fig. 12, although there is some overlap between sectors. Such overlapping filings can be the result of cross-disciplinary research and can provide evidence of ‘transformational’ and ‘disruptive’ technologies.

image file: c4nr01600a-f12.tif
Fig. 12 Proportion of overall graphene patents, by sector as of July 2014.

Considering the wide range of potential graphene applications, indicative of crossing vertical technology ‘silos’ (with applications in sectors as diverse as electronics, ICT, energy, consumer goods, polymers, automotive industry, medicine, and industrial chemicals/catalysis), the dominance of synthesis and electronics alone suggests this is an early stage space with plenty of scope for development.

Additionally, given the relatively young age of this space and the demands for mass-production, the strong drive toward synthesis observed in the patent data is unsurprising.74 As the underlying graphene technology space develops and the patent space matures, re-distribution seems inevitable, probably away from synthesis and towards the currently less well-established (or not yet conceived) end-use applications.

Our analysis of filing geography gives an indication of the key innovation locations and potential markets. This interpretation is further supported by noticing that the patenting trend closely follows the standard technology evolution pattern as discussed in ref. 75.

Fig. 13 plots the geographical breakdown of graphene patent filings by filing jurisdiction. Companies tend to file first in their home jurisdiction. The second filing location (other than in the case of an international Patent Cooperation Treaty – PCT – application) is likely to be a key market or a key manufacturing location.

image file: c4nr01600a-f13.tif
Fig. 13 Graphene patent filing authorities. EPO, European patents office; WIPO, World Intellectual Property Organization; US PTO United States Patent and Trademark Office.

Fig. 13 provides evidence of a relative increase in graphene patent filings in South Korea from 2007 to 2009 compared to 2004–2006. This could indicate increased commercial interest in graphene technology from around 2007. The period 2010 to 2012 shows a marked relative increase in graphene patent filings in China. It should be noted that a general increase in Chinese patent filings across many ST domains in this period is observed.76 Notwithstanding this general increase in Chinese patent activity, there does appear to be increased commercial interest in graphene in China. It is notable that the European Patent Office contribution as a percentage of all graphene patent filings globally falls from a 8% in the period 2007 to 2009 to 4% in the period 2010 to 2012.

The importance of the US, China and South Korea is emphasised by the top assignees, shown in Fig. 14. The corporation with most graphene patent applications is the Korean multinational Samsung, with over three times as many filings as its nearest rival. It has also patented an unrivalled range of graphene-technology applications, including synthesis procedures,77 transparent display devices,78 composite materials,79 transistors,80 batteries and solar cells.81 Samsung's patent applications indicate a sustained and heavy investment in graphene R&D, as well as collaboration (co-assignment of patents) with a wide range of academic institutions.82,83

image file: c4nr01600a-f14.tif
Fig. 14 Top 10 graphene patent assignees by number and cumulative over all time as of end-July 2014. Number of patents are indicated in the red histograms referred to the left Y axis, while the cumulative percentage is the blue line, referred to the right Y axis.

It is also interesting to note that patent filings by universities and research institutions make up a significant proportion (∼50%) of total patent filings: the other half comprises contributions from small and medium-sized enterprises (SMEs) and multinationals.

Europe's position is shown in Fig. 10, 12 and 14. While Europe makes a good showing in the geographical distribution of publications, it lags behind in patent applications, with only 7% of patent filings as compared to 30% in the US, 25% in China, and 13% in South Korea (Fig. 13) and only 9% of filings by academic institutions assigned in Europe (Fig. 15).

image file: c4nr01600a-f15.tif
Fig. 15 Geographical breakdown of academic patent holders as of July 2014.

While Europe is trailing other regions in terms of number of patent filings, it nevertheless has a significant foothold in the patent landscape. Currently, the top European patent holder is Finland's Nokia, primarily around incorporation of graphene into electrical devices, including resonators and electrodes.72,84,85

European Universities also show promise in the graphene patent landscape. We also find evidence of corporate-academic collaborations in Europe, including e.g. co-assignments filed with European research institutions and Germany's AMO GmbH,86 and chemical giant BASF.87,88 Finally, Europe sees significant patent filings from a number of international corporate and university players including Samsung,77 Vorbeck Materials,89 Princeton University,90–92 and Rice University,93–95 perhaps reflecting the quality of the European ST base around graphene, and its importance as a market for graphene technologies.

There are a number of features in the graphene patent landscape which may lead to a risk of patent thickets96 or ‘multiple overlapping granted patents’ existing around aspects of graphene technology systems. There is a relatively high volume of patent activity around graphene, which is an early stage technology space, with applications in patent intensive industry sectors. Often patents claim carbon nano structures other than graphene in graphene patent landscapes, illustrating difficulties around defining ‘graphene’ and mapping the graphene patent landscape. Additionally, the graphene patent nomenclature is not entirely settled. Different patent examiners might grant patents over the same components which the different experts and industry players call by different names. Use of a variety of names for the same components could be a deliberate attempt at obfuscation. There is some support for this view in the relevant academic literature. E.g., ref. 97 suggested that where patent assessment quality is low (e.g. due to inadequate expertise by patent examiners of a particular technology space), leading players might engage in high-volume patenting to deliberately create a ‘patent thicket’, with a range of possible negative effects on innovation.98

Despite the challenges described above, there are a number of important opportunities of which academics, SMEs and multinationals should take advantage, including increased occurrences of academia-industry collaboration (following the lead of South Korea and the US); preparing for the inevitable re-distribution of the graphene patent space as it matures and, most likely, moves away from synthesis, towards the currently less well-established (or not yet conceived) end-use applications.

2. Fundamental research

One of the reasons for the fast progress of graphene research is the wealth of its unique properties. However, what makes it really special, and gives it a disruptive value, is that all those properties are combined in a single material. Transparency–conductivity–elasticity can find use in flexible electronics, high mobility (μ)-ultimate thinness in efficient transistors for RF applications, while transparency–impermeability–conductivity can be exploited for transparent protective coatings. The list of such combinations is ever growing. The most important are probably those not yet explored, as they might lead to new applications.

Currently, several record high characteristics have been achieved with graphene, some of them reaching theoretically predicted limits: room temperature (RT) μ of 2.5 × 105 cm2 V−1 s−1 (ref. 99) and μ ∼ 6 × 106 cm2 V−1 s−1 at 4 K,100 a Young modulus of 1 TPa and intrinsic strength of 130 GPa;18 impermeability for gases101 and so on. Graphene also has record high thermal conductivity κ (∼2000 to 5300 W m−1 K−1 (ref. 102) and can withstand high current densities (million times higher than copper).103

The surge in graphene research also paved the way for experiments on many other 2d crystals.5 One can use similar strategies to graphene to get new materials by mechanical5 and liquid phase exfoliation of LMs38 or CVD. An alternative strategy to create new 2d crystals is to start with existing ones (e.g. graphene) and use them as atomic scaffolding for modification by chemical means (graphane33 or fluorographene34). The resulting pool of 2d crystals is huge, and covers a range of properties: from the most insulating to the best conductors, from the strongest to the softest. Suitable properties may be used depending on the targeted application. E.g., to cover a range of various conductance properties (but keeping the strength) one might use combinations of graphene and fluorographene, the latter being insulating, but almost as strong as the former.

For the long-term future, opportunities can be envisioned, combining conducting graphene with semiconducting and optically active 2d crystals, to create hybrid multilayer superstructures. If 2d crystals hold a wide variety of properties, the sandwiched structures of 2, 3, 4… layers of such materials can further offer longer term prospectives. By assembling 2d structures, one can engineer artificial 3d crystals, displaying tailored properties. Since such 2d based heterostructures104,105 can be assembled with atomic precision and individual layers of very different identity can be combined together, the properties could in principle be tuned to fit any application. Furthermore, the functionality of those stacks is embedded in the design of such heterostructures. First proof of principle devices are already available,106 such as vertical tunnelling transistors106 which show promising electrical characteristics.107,108 Starting with fundamental studies, the aim is to learn how to tune properties of such hetero- or hybrid systems in order to target a specific functionality.

Exploiting the full potential offered by the electronic and mechanical properties of GRMs in applications requires extensive fundamental studies. Graphene transistors and interconnects have an opportunity to complement and extend current Si technology. One route towards the use of graphene transistors for logic devices relies on creating a controllable band gap. The limited on/off current ratio (ION/IOFF) may be resolved in new transistor designs, which exploit the modulation of the work function of graphene,109 or carrier injection from graphene into a fully-gapped semiconductor,110 by gaining control over vertical (rather than planar) transport through various barriers,106 or using graphene as a gate, electrode, or interconnect. For the latter application of graphene, its electrical and thermal conductivities play an important role, so that studies of those properties should be intensified, especially in polycrystalline CVD-material.

Nature offers a very broad class of 2d crystals. There are several LMs which retain their stability in the form of monolayer and whose properties are complementary to those of graphene. Transition metal oxides (TMOs) and transition metal dichalcogenides (TMDs) also have a layered structure.111 Atoms within each layer are held together by covalent bonds, while van der Waals (vdW) interactions hold the layers together.111 LMs include a large number of systems with interesting properties.111E.g., NiTe2 and VSe2 are semi-metals,111 WS2, WSe2, MoS2, MoSe2, MoTe2, TaS2, RhTe2, PdTe2 are semiconductors,111 h-BN, and HfS2 are insulators, NbS2, NbSe2, NbTe2, and TaSe2 are superconductors.111 Moreover, there are other LMs such as Bi2Se3, Bi2Te3 that show thermoelectric properties111 and may behave as topological insulators (TIs).112 Atomic layers of these materials can be produced,5 using mechanical or liquid-phase exfoliation, see section 4 for more details on production.

A wider variety of 2d materials are also being explored, such as the graphene analogue of silicon (i.e., silicene),113,114 germanium (i.e., germanene),115 phosphorus (i.e., phosphorene)116 and tin (i.e., stanene).117,118 Another large LM class is that comprising the MXenes.119,120 These are derived by exfoliating the so called MAX Phases, i.e. layered, hexagonal carbides and nitrides having the general formula: Mn+1AXn, (MAX) where n = 1 to 3, M is an early transition metal, A is an A-group (mostly IIIA and IVA, or groups 13 and 14) element and X is either carbon and/or nitrogen.

Chemical modification of graphene33 allows the creation of another class of 2d crystals with a non-zero bandgap, such as graphane33 and fluorographene.34 Modification of these materials is interesting, e.g. p-doped graphane could be an electron–phonon superconductor121 with a critical temperature (Tc) above 77 K.121

There is also a growing number of graphane analogues, such as germanane,122 and stanane.123Ref. 122 synthesized mm-scale crystals of hydrogen-terminated germanane from the topochemical deintercalation (i.e., selective for a specific chemical element) of CaGe2 resulting in a lattice of Ge atoms with an analogous geometry to the sp3-hybridized graphane surface, in which every Ge atom is terminated with either H or OH above or below the layer.124 Germanane is thermally stable up to 75 °C.122 Above this T, dehydrogenation and amorphization begin to occur.122

2.1. Electronic transport

Graphene's promise to complement or even replace semiconductors in micro- and nanoelectronics is determined by several factors. These include its 2d nature, enabling easy processing and direct control of the charge carriers, fast moving (quasi-relativistic) electronic excitations yielding a high μ (almost equal between electrons and holes) – both at RT and low T, and high κ. Graphene crystals have two well-established allotropes, single layer graphene (SLG), where charge carriers resemble relativistic Dirac particles,125 and bilayer graphene (BLG), where electrons also have some Dirac-like properties, but have a parabolic dispersion.125 However, unlike SLG, where the absence of a gap is protected by the high symmetry of the honeycomb lattice, BLG is more versatile: a transverse electric field can open a gap126–128 and its low-energy band structure can be qualitatively changed by strain.129 Each of these has advantages and disadvantages for a given application, and one has to learn how to control and exploit them to create functional devices.

Concerning μ, further research is needed to understand the effects of defects and charge inhomogeneities, as well as development of doping techniques. The influence of various dielectric substrates or overgrown insulators also needs further basic understanding in order to optimize device performance. Further studies of transport regimes and optoelectronic effects in gapped BLG are needed for FET applications. Considering the possible use of an electrically induced gap in BLGs for quantum dots (QDs) and engineered QDs-based circuits (e.g., for quantum information processing130), a detailed understanding of the influence of disorder and Coulomb interaction on the T dependence of conductivity is required, including the nature of variable range hopping in gapped BLGs, which can be generically described by an exponential increase of resistance following R ∼ exp[(T0/T)p] (ref. 131) with T0 a constant depending on the localization length and density of states, whereas the exponent is given by p = 1/2 for the Efros–Shklovskii mechanism132 and p = 1/3 for the Mott hopping regime;131 the dominating regime depending on the material.133

Besides the studies of sample-average graphene parameters, such as, Rs, it is highly desirable to get insights into local properties of graphene used in devices. This can be achieved by means of several non-destructive techniques: Raman spectroscopy,134–136 Kelvin probe microscopy,137 local compressibility measurements,138 and non-contact conductivity, using capacitive coupling of a probe operated at high frequency.139 The application of such techniques to graphene is natural, due to its 2d nature. These techniques can be used to study GRMs to reveal the role of inhomogeneity in carrier density, the role of particular substrates, and can shed light on the role of structural defects and adsorbents in limiting device performance.

The peculiar properties of electrons in SLG (their similarity to relativistic Dirac particles) make a p–n junction in graphene transparent to electrons arriving at normal incidence.140,141 On one hand, this effect, known as Klein tunneling,141 makes it difficult to achieve a complete pinch-off of electric current, without chemical modification or patterning.142 On the other hand, it offers a unique possibility to create ballistic devices and circuits where electrons experience focusing by one or several p–n interfaces.143 The development of such devices requires techniques of non-invasive gating [see, e.g., ref. 144]. Another method to improve quality of graphene is to suspend it over electrodes (also used as support) and then clean it by current annealing.144,145 This enables one to achieve highly homogeneous carrier density, and micron-long mean free paths, enabling a detailed investigation of electronic properties at very low excitation energies.145

Understanding the transport properties of graphene also includes its behaviour in the presence of a strong – quantising – magnetic field. As a truly 2d electron system, graphene displays the fundamental phenomenon of quantum Hall effect (QHE),146–150 which consists in the precise quantisation of Hall resistance of the device.150 Both integer and several fractional QHE (FQHE) states have been observed,151,146,148 the latter requiring very high crystalline quality and pure material,151 where the Coulomb interactions between electrons can become very strong, leading to the formation of correlated states of matter.152 The QHE robustness in SLG opens a possibility to explore one, up to now, impossible regime of quantum transport in solid-state materials: the interplay between QHE and superconductivity in one hybrid device made of graphene and a superconductor with a high critical magnetic field (e.g., a NbTi alloy153). Moreover, the particular robustness of QHE in graphene on the Si face of SiC154 (still waiting for a complete understanding155) makes it a suitable platform for a new type of resistance standard.154

One of the issues in the fabrication of GFETs is electrostatic gating. Atomic Layer Deposition (ALD) of high-[scr K, script letter K] [where [scr K, script letter K] is the dielectric constant] dielectrics (Al2O3, HfO2) is one possibility worth further exploration, due its accurate control of layer thickness.156 After such processing, graphene can be transferred to a Si substrate in which deep trenches previously filled with metal (e.g., W) form the back-gate, and the source and drain are subsequently deposited on the graphene itself. Such an approach offers a possibility to build devices with complex architectures. However, ALD uses alternating pulses of water and precursor materials157 and, since graphene is hydrophobic,158 the deposition of a uniform, defect-free dielectric layer is difficult,159 and requires further optimization. Another promising technological advance is offered by photochemical gating.160 There are several polymers where UV light converts Cl atoms into acceptors, whereas thermal annealing returns them into a covalently bound state.161 Due to easy charge transfer between graphene and environment, UV illumination can modulate carrier density in graphene covered by such polymers, enabling non-volatile memory cells.162

For device applications, graphene contacts with metals and semiconductors require further studies: charge transfer between materials, formation of Schottky barriers, and graphene p–n junctions. The contacts play a crucial role for several devices: for superconducting proximity effect transistors,163 where they determine how Cooper pairs penetrate graphene, and for transistors used to develop quantum resistance standard, also needing very low resistance contacts to reduce overheating at the high-current performance of the resistance standard. Chosen to match the work functions of graphite and metals, the most common combinations are Cr/Au,164 Ti/Au164,165, Ti/Pt,166 and Ti/Pd/Au,167 the latter exhibiting lower contact resistances in the 10−6 Ω cm−2 range.167 The best results to date, down to 10−7 Ω cm−2, were obtained for Au/Ti metallization with a 90 s O2 plasma cleaning prior to the metallization, and a post-annealing at ∼460 °C for 15 min.168

2.2. Spectroscopic characterization

Spectroscopy is an extremely powerful non-invasive tool in graphene studies. Optical visibility of graphene, enhanced by an appropriately chosen substrate structure,169–171 makes it possible to find flakes by inspection in an optical microscope. While a trained person can distinguish SLG from FLG by “naked eye” with high fidelity, Raman spectroscopy has become the method of choice for a more systematic analysis.134–136

The Raman spectrum of graphite was measured 44 years ago.172 Since then, Raman spectroscopy has become a commonly used characterisation technique in carbon ST, as a method of choice to probe disordered and amorphous carbons, fullerenes, nanotubes, diamonds, carbon chains, and poly-conjugated molecules.173 The Raman spectrum of graphene was first reported in 2006.134 This triggered a huge effort to understand phonons,134,136 electron–phonon,134,136,174 magneto-phonon,175–177 and electron–electron178 interactions, and the influence on the Raman process of number134 and orientation134,136 of layers, electric179–181 or magnetic182,183 fields, strain,129,184 doping,179,185 disorder,136 defects,186 quality187 and types187 of edges, functional groups.188 The graphene electronic structure is captured in its Raman spectrum that evolves with the number of graphene layers (N).134 The 2D peak changes in shape, width, and position for increasing N (see Fig. 16), reflecting the change in the electron bands. The 2D peak is a single band in SLG, whereas it splits in four in BLG.134 Since the 2D shape reflects the electronic structure, twisted multi-layers can have 2D peaks resembling SLG.134 FLGs can also be characterized by the interlayer shear mode189 (see Fig. 17), i.e. the C peak that probes the interlayer coupling.190 This peak scales from ∼44 cm−1 in bulk graphite to ∼31 cm−1 in BLG (see Fig. 17).190 Layer breathing modes (LBMs) can also be observed in the Raman spectra of FLGs, via their resonant overtones in the range 80–300 cm−1.191 They can also be addresses through the stokes and anti-stokes combinations with the D' peak,134,196 or in twisted samples.2335

image file: c4nr01600a-f16.tif
Fig. 16 Raman spectra of SLG (1LG), BLG (2LG), TLG (3LG), and bulk graphite measured at 633 nm. Adapted from ref. 204.

image file: c4nr01600a-f17.tif
Fig. 17 (a) C peak as a function of number of layers. (b) Fitted C- and G-peak position as a function of inverse number of layers. Adapted from ref. 134.

It is important to note that, although being an in-plane mode, the 2D peak is sensitive to N because the resonant Raman mechanism that gives rise to it is closely linked to the details of the electronic band structure,135,136 the latter changing with N, and the layers relative orientation.192 On the other hand, the C peak and LBMs are a direct probe of N,190,191,193–196 as the vibrations themselves are out of plane, thus directly sensitive to N. Raman spectroscopy has provided key insights in the related properties of all sp2 carbon allotropes, graphene being their fundamental building block, and other carbon-based materials, such as amorphous, nanostructured and diamond-like carbons173,188,197–203 Raman spectroscopy has also huge potential for LMs,204,205 other than graphene, see section 2.8.1.

Studies of the magneto-phonon resonances206,207 enable to directly measure the electron–phonon coupling in SLG, BLG, and multilayers.177,206–208,209–211 Optical spectroscopy allows to study the split-bands in BLG,212,213 and the analysis of disorder-induced phonon-related Raman peaks134 provides information on sample quality complementary to that extracted from transport measurements.

Angle-resolved photoemission spectroscopy (ARPES) directly probes band dispersions and lattice composition of electron states, which determine the pseudospin symmetry of electronic states.214,215

Further improvement of the above-mentioned optical characterisation techniques and development of new approaches are critically important for in situ monitoring. Outside the visible-range and Infra-Red (IR) optical spectroscopy, detailed studies of defects in graphene can be addressed using scanning transmission electron microscopy (STEM), energy loss spectroscopy, low-angle X-ray spectroscopy, and resonant inelastic X-ray scattering (RIXS). The development of a standardised optical characterisation toolkit with the capability to monitor N, as well as quality and doping level, is one of the key needs for the progress in graphene mass manufacturing. Since there are several routes towards viable mass production, described in section 4, the suitable energy/wavelength range for the standardised spectroscopic characterisation toolkit is not known yet, thus spectroscopic studies of graphene need to be carried out over a broad energy range, from microwaves and far IR to UV and X-ray.

Scanning tunnelling microscopy (STM) is another important tool. Since electronic states in graphene can be directly addressed by a metallic tip,216 STM studies may be instrumental for understanding the morphology and electronic structure of defects: vacancies, grain boundaries (GBs), functionalised faults, and strongly deformed regions (‘bubbles’) resulting from processing or transfer. Such studies will be necessary for materials manufactured using each of the methods discussed in section 4, and to investigate the result of subjecting graphene to various gases. The use of STM under extreme conditions, such as strong magnetic fields, can probe local properties of electrons in Landau levels, and their structure close to defects.

2.3. Magnetism and spin transport

The control and manipulation of spins in graphene may lead to a number of novel applications and to the design of new devices, such as spin-based memories217 or spin logic chips.218 Graphene is uniquely suitable for such applications, since it does not show sizeable spin–orbit coupling,74 and is almost free of nuclear magnetic moments.219 Graphene currently holds the record for the longest spin relaxation length at RT, initially evaluated to be ∼5 μm,219 with promise for applications.220–222 At lower T, there are indications that the spin relaxation length could approach ∼100 μm.223 Further studies require the investigation of spin injection, diffusion, relaxation and of the interfaces between graphene and magnetic materials.

The magnetic properties of graphene are connected to the defects. As a 2d electronic system, graphene is intrinsically diamagnetic.224 However, defects in graphene, as well as localisation of electrons in or around defects (vacancies, edges and covalently bonded dopants) can generate localised magnetic moments which directly modulate the spin current, as it has been proven in the cases of hydrogen adatoms and lattice vacancies.225,226

Edge magnetism has been predicted in graphene nanoribbons (GNRs) for certain edge geometries;227 the structure of graphene nanomesh, obtained by using block-copolymer nanopatterning of graphene, was also theoretically shown to yield RT magnetic states affecting spin transport.228

An enhanced paramagnetic signal was measured in graphene crystallites,229 and it was found that magnetism is enhanced in irradiated samples,229 similar to graphite.230 Strong enhancement of paramagnetism was also observed in fluorographene.231

An unambiguous assessment of the nature and the formation of magnetic moments in graphene and in FLG, and the resulting control of their properties would be a major advance and would significantly expand graphene applications.

The necessary steps towards a full portrait of graphene's magnetic properties are a complete understanding of disorder as well as (quantum) confinement (as for graphene quantum dots, GQDs, and GNRs) on spin relaxation and dephasing. This requires the investigation of the limits of conventional spin relaxation mechanisms common to metals and small gap semiconductors, i.e. Elliot–Yafet (EY)232,233 and Dyakonov–Perel234 (DP). EY was originally232,233 derived for spin relaxation in metals, and relates the spin dynamics with electron scattering off impurities or phonons.232,233 Each scattering event changes the momentum, with a finite spin-flip probability, that can be derived by perturbation theory (assuming weak spin–orbit scattering). This gives rise to a typical scaling behaviour of the spin relaxation time proportional to the momentum scattering time. DP234 is an efficient mechanism of spin relaxation due to spin orbit coupling in systems lacking inversion symmetry.234 Electron spins precess along a magnetic field which depends on the momentum.234 At each scattering event, the direction and frequency of the precession change randomly. The scaling behaviour is opposite to EY, with a spin relaxation time which is inversely proportional to the momentum scattering time.235

Two other mechanisms of spin relaxation in graphene have been proposed.236,237 One involves local magnetic moments which produce resonances and fast spin relaxation based on the resonant scattering of electrons off magnetic moments, which can be due to nonmagnetic adatoms, organic molecules, or vacancies.236 The other is related to the interplay between spin and pseudospin quantum degrees of freedom when disorder does not mix valleys.237 Such strong contribution of spin/pseudospin entanglement is particularly important when defects or impurities at the origin of local Rashba spin–orbit coupling (i.e. a momentum-dependent splitting of spin bands in 2d systems)238 preserve the pseudospin symmetry and lead to very long mean free path.

The role of edges on spin scattering and relaxation has yet to be clarified, as well as the case when the injected spin-polarized charges flow in close proximity (and interact) with other extrinsic spins (in localized or more extended charged states, located below or on top of the graphene).

The role of the substrate, contacts and environmental conditions on spin relaxation needs to be clarified. A detailed comparison between exfoliated graphene on SiO2, graphene on BN, and graphene grown on SiC is still missing. This would allow the classification of materials and devices' parameters, which have, to date, shown limited spin transport. These results could be compared with spin transport and relaxation in suspended graphene devices, which would provide the reference clean system. This is of fundamental importance for further exploration of more complex uses of the spin degree of freedom inside technology.

A systematic comparison of spintronic systems based on SLG, BLG and FLG still needs to be carried out. With a focus on spin ensembles, the RT capability of graphene devices has to be ascertained.

The knowledge derived from these investigations could be exploited in multiple ways. E.g., novel studies could be carried out to induce (para-) magnetism by introducing localised defect states in a controlled way, or by decoration of the surface with magnetic atoms or molecules. In the search for fingerprints of local magnetic ordering states and magnetoresistance profiles, chemically modified graphene (CMG)-based materials should be produced to investigate the potential of new physical phenomena. Other topics include studying electronic transport in graphene in proximity of ferromagnetic materials or ferromagnetic insulators (magnetic oxides as EuO, EuS, high-Curie T (above which the magnetic state of the system is lost) Yttrium Iron Garnet (YIG), NiO, CoFe2O4). This may induce spin polarization, which could be exploited to demonstrate spin filtering effects.239 Related issues focus on the existence of nanomagnetism of magnetic materials deposited on graphene, and the understanding of the interfacial electronic structure of such contacts.

2.4. Polycrystalline graphene

The GB role in transport and optical properties240 needs to be fully investigated, especially in view of large-scale production. The theoretical exploration of the properties of large size realistic models is crucial for guiding experiments.

Microscopic studies of grain boundaries are needed to determine their precise lattice structure and morphology, as well as the related functionalization of broken carbon bonds by atoms/molecules acquired from environment. Grain boundaries in the 2d graphene lattice are topological line-defects consisting of non-hexagonal carbon rings, as evidenced by aberration corrected high resolution TEM investigations.241 Although they are expected to substantially alter the electronic properties of the unperturbed graphene lattice,242 so far little experimental insight into the underlying mechanisms is available. GB introduce tension in graphene nanocrystals,243 which, in turn, bears influence on the electronic properties, including local doping. From the point of view of electronic transport, GB generate scattering, possibly with a strongly nonlinear behaviour, but present knowledge on the precise effects is incomplete. Ref. 244 suggested a scaling behaviour of polycrystalline graphene,244 which gives μ ∼ 300[thin space (1/6-em)]000 cm2 V−1 s−1 at RT for an average grain size of 1 μm and clean GB. This also shows the need for a more detailed GB chemical characterization, since these are more chemically reactive and could drive an essential part of the resistance of the material. Indeed, e.g., CVD grown samples fall behind by about an order of magnitude compared to mechanically exfoliated ones.9 The internal GB structure, and the resulting broken electron-hole (e–h) and inversion symmetry may generate thermo-power245 and local rectification,246 which may affect high current performance. Depending on their structure, GBs have been initially theoretically predicted to be highly transparent,247 or perfectly reflective,247 while other studies suggest GB act as molecular metallic wires248 or filter propagating carriers based on valley-index.249

A comprehensive picture of GBs’ spectral properties is thus missing, and should be established using STM and atomic force microscopy (AFM), and local optical probes,250 given possible specific light absorption and emission,251,252 The use of graphene for energy applications, in solar cells, also requires understanding of the GB role in the charge transfer between graphene and environment. Moreover, optics, combined with electrochemistry, is needed to figure out ways to re-crystallize graphene poly-crystals, and to assess durability (i.e. until when the (opto)electronic and thermal properties are maintained without degradation). There is growing evidence that GBs degrade the electronic performance.253,254

2.5. Thermal and mechanical properties of graphene

Practical implementation of graphene requires the understanding of its performance in real devices, as well as its durability under ambient and extreme conditions. A specialised effort will be needed to study the reliability of graphene-based devices, such as electric or thermal stress tests, device lifetime, etc. To preserve performance, it is likely that some protection of the graphene and the metals will be needed to minimize environmental effects.

Due to the sp2 hybridization, pristine SLG is very strong, and it takes 48[thin space (1/6-em)]000 kN m kg−1 of specific strength (i.e. strength (force per unit area at failure) divided by density) before breaking18 (compare this to steel's 154 kN m kg−1 (ref. 255)). This makes graphene a desirable addition to lightweight polymers, and the enforcer of their mechanical properties. Moreover, as ultrathin stretchable membrane, SLG is an ideal material for nonlinear tuneable electromechanical systems. However, for the practical implementation of realistic graphene systems, a detailed study of mechanical properties of polycrystalline graphene is needed in vacuum, ambient environment, and of graphene embedded in polymers. Studies of mechanical properties of GBs between graphene nano-crystals will require a further improvement of scanning techniques. The durability of graphene in various systems will also depend on its ability to recrystallize upon interaction with various chemical agents, as well under various types of radiation, from UV and soft X-rays to cosmic rays.

The application of graphene in the electronics and optoelectronics requires detailed understanding of its thermal and mechanical properties. Several early experiments,256,257 indicate that graphene is a very good heat conductor, due to the high speed of acoustic phonons in its tight and lightweight lattice. Detailed studies of heat transfer through graphene and the interfacial Kapitza thermal resistance258 (i.e. the measure of an interface's resistance to thermal flow) between graphene and other materials (metallic contacts, insulating substrates, polymer matrix) are needed. Graphene performance at high current may lead to overheating, and quantitative studies (both experimental and theoretical) are needed to compare its performance with the standards set in electronics industry. Moreover, overheating upon current annealing may lead to its destruction, so that studies of thermal and thermo-mechanical properties are needed to assess its durability in devices, and optimise its use in realistic and extreme conditions. In particular, in situ studies of kinetics and dynamics at the break point (use of HRTEM would be appropriate) are a challenging but necessary step towards practical implementation.

Experimental studies need to be complemented by ab initio and multiscale modelling of nanomechanical and heat transport properties, and modelling of graphene at strong non-equilibrium conditions (see Section 2.10.4).

2.6. Artificial graphene structures in condensed-matter systems

Advances in the design and fabrication of artificial honeycomb lattices or artificial graphene (AG) pave the way for the realization, investigation, and manipulation of a wide class of systems displaying massless Dirac quasiparticles, topological phases, and strong correlations. Such artificial structures are currently created by three approaches: atom-by-atom assembling by scanning probe methods,259 nanopatterning of ultra-high- μ two-dimensional electron gases (2DEGs) in semiconductors,260 and optical trapping of ultracold atoms in crystals of light.261 Examples of AG structures realized so far are shown in Fig. 18. The interplay between single-particle band-structure-engineering,262 cooperative effects and disorder263 can lead to interesting manifestations in tunnelling264 and optical spectroscopies.262
image file: c4nr01600a-f18.tif
Fig. 18 (a) SEM image of AG structure realized by e-beam lithography and reactive ion etching on a GaAs/AlGaAs heterostructure. Electrons localize underneath the nanopillars (white dots; also shown in (d).260 (b) STM topography of a molecular graphene lattice composed of 149 carbon monoxide molecules.259 (c) A honeycomb optical lattice for ultracold K atoms.261 (e) Electron moving under the prescription of the relativistic Dirac equation. The light blue line shows a quasi-classical path of one such electron as it enters the AG lattice made of carbon monoxide molecules (black/red atoms) positioned individually by an STM tip (comprised of Ir atoms, dark blue). (f) Tight-binding calculations of the Dirac Fermion miniband structure of the AGs in (a) and (d).

One of the reasons for pursuing the study of AGs is that these systems offer the opportunity to reach regimes difficult to achieve in graphene, such as high magnetic fluxes, tuneable lattice constants, and precise manipulation of defects, edges, and strain. These can enable tests of several predictions for massless Dirac fermions,265,266 Studies of electrons confined in artificial semiconductor lattices, as well as studies of cold fermions and bosons in optical lattices, may provide a key perspective on strong correlation and the role of disorder in condensed matter science. AG systems might open new avenues of research on spin–orbit coupling, with impact on spintronics, and frontier issues related to novel topological phases.

These are centred on TIs,267–269 that have emerged as a promising class of materials in this regard.269 Strong spin–orbit coupling results in an insulating bulk and metallic edge or surface states (respectively for 2d and 3d systems). These states are “topological” in the sense that they are insensitive to smooth changes in material parameters, and also exhibit unique spin textures. One remarkable phenomenon is the Quantum Spin Hall Effect (QSHE), which for a 2d TI, consists of pairs of edge counter-propagating modes with opposite spins.112 Meanwhile, the surface state of a 3d TI exhibits spin-momentum locking, where the spin is perpendicular to the electron momentum. The spin texture of these states implies that backscattering by non-magnetic impurities is strongly suppressed, resulting in insensitivity to disorder and long coherence times. Due to this behaviour, TIs are promising for next-generation electronics, as well as for spintronics and quantum computing.270 The QSHE effect was first predicted in graphene,112 linked to the honeycomb topology of the lattice112 and to the contribution of a weak spin–orbit coupling.112 AG structures in systems properly engineered to display large spin–orbit coupling represent viable candidates to simulate TI states. There are further possibilities for research arising from the demonstration that single atoms can function as atomic-size gates of a 2d electron system at noble metal surfaces, whereby simple molecules, such as CO, function as repulsive potentials for surface electrons when shaped into open and closed quantum structures. Individual CO molecules arranged on Cu(111) were used as a tuneable gate array to transform a 2d gas of electrons moving through these lattices (Fig. 18).201

Control over every lattice position and potential would result in control of the spatial texture of the hopping parameter, ultimately allowing observation of electronic ordering into ground states, rarely encountered in natural systems. In AGs, molecular graphene, artificial lattices in semiconductors, and optical lattices of cold atoms, controlled densities of ‘artificial impurities’ can be introduced259 in otherwise perfect lattices. Studies of these artificial structures may provide insights on localization and μ degradation in graphene.

2.6.1. Honeycomb lattices in semiconductors. The goal is the creation of nanostructures on a 2DEG confined in high-μ semiconductor heterostructures, creating an in-plane potential with honeycomb geometry so that the miniband structure has well-defined (isolated) Dirac points. The lattice constant in graphene is fixed at ∼1.42 Å. In contrast, AG structures in semiconductors can have tuneable lattice period in the range 10–100 nm,260 so that it should be possible to change interaction regimes from one in which Mott–Hubbard physics (such as the Mott–Hubbard excitation gap and collection spin density modes271) manifests for weak inter-site interactions compared to π–π coupling,260 to one where inter-site interactions drive the creation of novel phases, and to the TI regime in materials with large spin–orbit interaction.

Semiconductor AG may also challenge current thinking in ICT, revealing new physics and applications of scalable quantum simulators for ICT based on semiconductor materials already used in real-life electronic and optoelectronic devices. Due to the embryonic nature of the field, any future activity will be high-risk, but has great potential for discoveries. In semiconductor materials the efforts should be directed to the realization of artificial lattices with small lattice constants and with tuneable amplitude, V0, of the potential modulation. The idea is that the energy range, Δ, in which the bands are linearly dispersing in AGs depends on the hopping energy, therefore this quantity is expected to exponentially increase as we reduce the lattice constant and/or decrease the amplitude of the potential modulation. One target could be the realization of AGs in the regime in which Δ approaches 1 meV. This requires lattice constants ∼20–40 nm (see Fig. 19).

image file: c4nr01600a-f19.tif
Fig. 19 (a) AG Minibands (energy is in meV) with a lattice period a = 60 nm, r0 = 0.2a (r0 is the width of the potential well) and V0 = 5 meV. A dash-dotted line is drawn at the Dirac-point energy. (b) Energy width, Δ, of the linear part of the spectrum near the Dirac point as a function of a. V0 is varied correspondingly (see the right vertical axis) in order to obtain an isolated Dirac point, i.e. without any other state inside the bulk BZ at the Dirac-point energy. Inset: magnification of the energy bands in panel (a) around the Dirac point energy. The blue dashed lines mark the energy limits of the linear dispersion approximation. Adapted from ref. 260.

One ambitious goal is to observe the dispersive intrasubband plasmon mode of the AG lattice by resonant inelastic light scattering or far-IR spectroscopy. Peculiar to plasmon modes in graphene, in fact, is the specific dependence of energy on electron density: ωplasmon(q) ∝ n1/4q1/2, where q is the in-plane wavevector.272 The difference with the classical square-root dependence n1/2q1/2 of 2d parabolic-band systems is a consequence of the ‘relativistic’ linear dispersion of Dirac fermions,273,274 The manifestations of Dirac fermions are particularly striking under the application of a perpendicular magnetic field.275 In AGs with lattice constant much smaller than the magnetic length (alB = ħc/eB) this is expected to lead to graphene-like Landau levels.

Such peculiar energy level structure, and the resulting anomalies in QHE experiments,145,146,148 have been largely explored in graphene276 where the lattice constant is a = 0.14 nm (lB ≈ 25 nm at 1 Tesla). Tight-binding calculations show that the Dirac Fermion physics occurs when lB/a > 1.275 In AGs with a ∼ 10–20 nm, a Dirac-Fermion Landau level structure is expected for magnetic fields of several Tesla. In molecular AG structures with a ∼ 1 nm, Dirac Fermion physics should emerge at much smaller magnetic fields. The occurrence of such phenomena can be investigated by conventional QHE and by optical spectroscopy. For lB/a < 1, commensurability effects, such as the Hofstadter butterfly,277–279i.e. a fragmentation of the Landau levels structure, begin to emerge and compete with the Dirac-Fermion physics of the honeycomb lattice.280 These effects prevail when lBa. The impact of the Hofstadter physics on the energy spectrum of a 2DEG in semiconductor heterostructures was studied in magneto-transport in a lateral superlattice of anti-dots arranged in a square geometry.280 Moiré superlattices arising in SLG and BLG coupled to h-BN provide a periodic modulation with length scales ∼10 nm enabling experimental access to the fractal Hofstadter butterfly spectrum.278,279,281 If met, these demanding limits will enable the occurrence of the physics linked to artificial massless Dirac fermions at T above liquid He. Finally, the impact of e–e interaction can be studied theoretically by exploiting advanced methods such as density-functional theory (DFT) and developing a Kohn–Sham DFT coded for 2d electrons moving in a model periodic potential (see Fig. 20)282 and experimentally by optical, transport and scanning probe methods. Additionally artificial topological order and spin-split counter-propagating edge channels can be pursued by creating honeycomb lattices in 2DEGs confined in InSb and InAs heterostructures, with a large spin–orbit coupling. In this area, the long-term vision is the establishment of a new field of quantum information processing and scalable quantum simulations based on nanofabricated AGs in high-μ semiconductor heterostructures.

image file: c4nr01600a-f20.tif
Fig. 20 Spatial distribution of electrons in AG (with one electron per pillar) calculated for two values of the potential well representing the pillar. The left and right panels show the results without and with e–e interactions.282
2.6.2. Honeycomb lattices with cold atoms. A different system for the experimental realization of artificial graphene is represented by ultracold atoms in optical lattices.283 Here, the role of electrons is taken by the atoms, which move in the periodic optical potential generated by the interference of different laser beams. The optical realisation of lattice potentials is an intrinsically clean method that allows for the production of disorder-free lattices with well-controlled topology, lattice spacing and potential strength. A suitable arrangement of laser beams can be used to produce honeycomb lattices allowing the simulation of AG. Hexagonal spin-dependent optical lattices, which can be seen as a triangular lattice with a bi-atomic basis where atoms occupy π+ and π polarized states (Fig. 21), were demonstrated with ultracold bosonic 87Rb in different internal states.284
image file: c4nr01600a-f21.tif
Fig. 21 (a) Lattice potential with alternating π+ (green spheres) and π (red spheres) polarization. The upper plot shows a cut through the 2d potential. (b) 1d potential along the channel of the orange dashed line in (a) for particles in different Zeeman states.284

A system of spin-polarised fermionic 40K atoms in honeycomb optical lattices was realized in ref. 261, where the presence of Dirac points in the energy spectrum was measured by momentum-resolved detection of the energy gap between bands. A controlled deformation of the hexagonal lattice was used to control the Dirac cones, moving them inside the Brillouin zone (BZ) and eventually merging them.

Besides the perfect knowledge of the potential landscape, this “atomic” approach to the simulation of AG has several additional advantages. Tuning interactions between particles with an appropriate choice of atoms or with Feshbach resonances285 (i.e. scattering resonances that occur when the energy of an unbound state of a two-body system matches that of an excited state of the compound system, see e.g.ref. 282,285,286), allows studying different interaction regimes, from ideal Dirac Fermion physics to strongly-interacting Mott–Hubbard physics. Multi-layer systems can be produced with an additional optical lattice in the orthogonal direction to the 2d honeycomb plane, with an adjustable tunnel coupling between the layers, which could allow the investigation of artificial BLG/FLG. Other interesting perspectives arise from the introduction of artificial magnetic fields for effectively-charged neutral atoms and spin–orbit coupling, which enrich the possibilities of Hamiltonian engineering in ultracold atomic systems.266

Several detection techniques can be used to probe the properties of atomic AG, including measuring transport of atoms across the lattice, even with single-atom (electron) imaging resolution. The excitation spectrum of the system can be directly measured with inelastic light scattering (Bragg spectroscopy),287 which gives access to the dynamical structure factor, while correlation functions of different order (1st order: phase correlations, 2nd order: density–density correlations) can be probed in time-of-flight experiments.286

In addition, a unique feature of ultracold atom experiments is the possibility of changing the lattice and interaction parameters from one experiment to the other, and even modifying them in real time, inducing rapid changes in the AG Hamiltonian enabling the investigation of non-equilibrium dynamics, which can give important information on the system properties.

Of particular relevance is the possibility to study the impact of disorder, which can be introduced in a controlled way by using additional inhomogeneous optical potentials, created e.g. with optical speckles or multi-chromatic optical lattices,288 see Fig. 22. This allows the investigation of the interplay between disorder and Dirac Fermion physics. An example related to 1d optical lattices is shown in Fig. 23, where the different images show the size of an ultracold cloud of non-interacting 39K Bose-condensed atoms for different evolution times and Δ: as Δ increases, the atom cloud stops expanding, becoming Anderson-localized.263

image file: c4nr01600a-f22.tif
Fig. 22 Schematic depiction of ultracold atoms trapped in artificial crystals produced by the interference of laser beams (optical lattices).

image file: c4nr01600a-f23.tif
Fig. 23 Localization transition for 39 K ultracold atoms in a quasiperiodic bichromatic lattice: as Δ is increased the atomic cloud becomes Anderson-localized.263

2.7. Atomic scale technology in graphene and patterned graphene

Tailoring electronic and optical properties in graphene can be achieved by lateral confinement of its 2d electron gas from the mesoscopic regime down to the molecular scale,289–295 The dominant approach consists in using inorganic resist to lithographically define GNRs,296,110,297,298 A resist-free approach can be achieved by focused ion beam lithography,299,300 However, the transport in ion-etched GNRs is strongly dominated by edge disorder and amorphization299–302 which calls for alternative approaches. Ultrasonically shredded graphene,303 carbon nanotube opening,304,305 AFM and STM tip-induced oxidation,306,307 and catalytic particle cutting,308–310 offer promising routes to 50–500 Å wide GNRs, but only the former has so far led to functional devices. 40 nm-wide GNRs grown on SiC have shown ballistic conductance on a length scales >10 μm,100 see section 2.7.1.

The ultimate goal of graphene-based nanotechnology is to achieve atomic-scale fabrication through techniques that are rapid enough to bridge the gap with standard nanofabricated features. A promising strategy should probably exploit electron and/or scanning probe microscopy techniques. A challenging objective is to investigate the suspended vs. supported cases and, in the latter, define the most suitable atomically flat substrate. However, chemical approaches should also be considered, either from the molecular synthetic or colloidal etching viewpoint. Next, atomic-scale imaging, such as STM, non-contact AFM, aberration-corrected HRTEM, should be developed in the specific realm of atomic-scale graphene devices. Electron transport, optical measurements and local, near-field measurements should be pushed to the limits to assess the properties in atomic-scale devices and identify the degrees of freedom able to control graphene behaviour, such as magnetic field, gate effects, optical excitations, near-field coupling to metallic surfaces, etc. These experimental issues should be guided by a theoretical description and simulations, in particular regarding the bridging between atomic/molecular scale and the mesoscopic regime.

2.7.1. Graphene nanoribbons. Nanofabrication applied to graphene has already produced a new physical system: GNRs. ION/IOFF can reach high values [up to 104],311 at RT,311 with the extra asset that all GNRs are found to be semiconducting, in contrast to nanotubes.312 Transport measurements in shredded GNRs have shown that scattering by substrate potential fluctuations dominates the edge disorder.313 One approach to produce GNRs with high crystallinity and smooth edges is based on e-beam etching at high energy (80–300 kV) in a TEM,314–317 Progress was also made in chemical synthesis,318–320 This draws a bridge between top-down patterning and atomically-precise chemical design.

To achieve patterning of GNRs and bent junctions with nm precision, well-defined widths and predetermined crystallographic orientations, STM lithography should be further improved. The latter can be used only for the demonstration of operational principles of new devices, since it should be difficult to incorporate it into a production line, despite the good stability and reproducibility even under ambient conditions307,321 The short de Broglie wavelength [∼0.5 nm] of He ions300,322 gives He ion lithography an ultimate resolution better than 0.5 nm,322 very attractive for GNRs300,323 Using 30 kV He ions, clean etching and sharp edge profiles, ∼15 nm GNRs were obtained, with little damage or doping,323 so that He-techniques may be considered further for GNR production. While many of the promising applications of graphene do not require precise nanoscale processing, there exist numerous applications, e.g. in digital nanoelectronics and spintronics, for which the precise engineering of GNRs307 or antidot lattices324 is mandatory. This is a challenging task, as the properties of GNRs and other graphene nano-architectures depend strongly on the crystallographic orientation of edges,142 the width,325 and the edges atomic structure,326 including edge disorder.142 Precise, reproducible and fast patterning is fundamental for mass production of devices.

Patterning of graphene not only concerns the removal of material (anti-dot lattices, nanomesh, and GNRs) but also suspending it on supports, holes or gates. To date, only few nano-processing methods327,328 have been reported to meet the very strict criteria for nanopatterning, i.e. crystallographic orientation control and atomic scale precision, see Fig. 24. These rely on local probes (STM, AFM) see Fig. 24, or crystallographically selective chemical reactions, or their combinations.

image file: c4nr01600a-f24.tif
Fig. 24 AFM lithography of graphene.306

The usual method for the production of patterns on the 10–100 nm scale is e-beam lithography, followed by plasma etching. Several groups have used this technique to make GNRs,206 single electron transistors (SET)329 and FETs.330 To open a practically relevant band gap, graphene must be patterned to critical dimensions in the range of a few nm. However, 20 nm is on the threshold of what can easily be achieved using conventional e-beam lithography, due to known electron scattering effects in common e-beam resists.331 Other top-down approaches, such as reduction of graphite oxide,681 unzipping of carbon nanotubes (CNTs)304,332Fig. 25a, or liquid-phase exfoliation (LPE)303 of graphite (Fig. 25d,e), have so far lacked control over the size and edge structure.

image file: c4nr01600a-f25.tif
Fig. 25 Top-down fabrication of GNRs via (a) CNT unzipping332 (b) STM lithography (adapted from ref. 307), (c) catalytic hydrogenation, using thermally activated nickel nanoparticles, (d) exfoliation of chemically modified328 and (e) expanded graphite662 (i.e. with larger interlayer distance than graphite due to intercalation of nitric337 and sulfuric acid338). Bottom-up fabrication: (f) chemical synthesis336 (g) schematic diagram of GNRs grown on SiC.100 (h) STM image of GNR edge showing helical edge structure.100

A high precision control of graphene edges could create narrow constrictions, down to the atomic size contacts,333,334 and this enables one to operate a GNR as a quantum wire,335 for use in quantum information processing, in conjunction with QDs. Simultaneously, bottom-up synthesis offers an alternative route towards the production of GNRs (see Fig. 25f): GNRs with lengths of 40 nm were reported.336 40nm wide GNR have been achieved on SiC,100 see Fig. 25g,h. In these GNRs, the transport is dominated by two modes.100 One is thermally activated, while the other is ballistic and T independent.100 At RT, the resistance of both modes is found to increase abruptly at a particular GNR length—the ballistic mode at 16 μm and the thermally activated one at 160 nm.100

Another goal is to investigate the effects of patterning on graphene, to fully control the balance between engineering of desirable properties, against introduction of performance inhibiting defects and artefacts. High-resolution (few nm-scale) lithography enables periodic patterns of voids (‘antidots’) and networks of GNRs.339 The transport, microwave, and far infrared (FIR) properties of such systems require a dedicated investigation. While the study of GNR devices and the optimization of their performance will contribute much to the development of graphene-based nanoelectronics (see section 5) and THz plasmonics, progress made towards atomic-scale technology would make graphene a strong platform for non-CMOS approaches to Boolean information processing, by inspiration of the mono-molecular electronics paradigm.340 Transport measurements may suggest new ways to implement Boolean logic into designed, atomically-defined graphene nanostructures.

2.7.2. Graphene quantum dots. QDs are sub-micron-size objects which can be incorporated in electronic circuits and then controlled electrically. There are two main physical effects that distinguish the QD electrical properties from other electronic system: size quantisation of electronic states into a discrete spectrum, and charge quantisation, the phenomenon known as Coulomb blockade.341 The ability to move electrons in/out the dot one by one makes it possible to use them as SETs. By trapping an odd number of electrons (e.g., one) one can create an electrically controlled localised spin and use it for quantum information processing.

The advantage of graphene as QD material lies in its reduced dimensionality, therefore large charging energy, which protects the quantised charged state of the dot. This enables SET operation at high T.342 Coulomb blockade effects and size quantization have been observed in GQDs.329,343,344 It is now necessary to achieve full control on GQD-based circuits. Besides a further development of atomic scale technologies on graphene, this also requires understanding of the properties of electronic states on graphene edges (functionalised and with dangling bonds).

An additional possibility to create GQDs130 is related to the unique properties of BLG.126–128,149 In BLG, one can use a transverse electric field created by external electrostatic gates to open a gap, reaching up to 200 meV.126 It has been demonstrated that one can confine electrons in small regions of BLGs using a combination of top/bottom gates, and then operate the charging states of such QDs electrostatically.345,346 Since spin relaxation in high-quality graphene is slow, in the order of 1 ns, see ref. 219, further studies of gap control and electron confinement in gapped BLGs are needed.

2.7.3. Patterning- and proximity-induced properties in graphene. Decoration of graphene with nanoparticles opens up a range of possibilities to modify its charge carrier properties, by proximity effects with superconductors, ferromagnets or coupling to strong spin–orbit entities. From a fundamental point of view, interesting topological transport effects were predicted for graphene decorated by 5d transition metal ad-atoms, with very high magneto-electric ratios.347 By covering graphene with superconducting islands one can induce superconductivity,163,348–350 through the Andreev reflection process [a proximity effect mechanism providing phase correlation in non-interacting electrons at mesoscopic scale],351 whereas by changing the carrier density one can control the Tc and current of the induced superconducting state,351 as well as induce a superconductor – quantum insulator transition,352,353 The newly acquired properties of graphene, due to its patterning with other materials, require detailed studies, aiming at determining new functionalities of the hybrid structures. Other routes are: (a) to exploit progresses in high critical magnetic field electrodes,354 to inject Cooper pairs in the edge states of graphene in the QHE regime; (b) combine proximity superconductivity and suspended graphene to extremely high-Q factors, controlled by the ac Josephson effect.163

Large-scale periodic patterning of graphene may also be done using deposition of nanoparticles, and this would change the high-frequency response of the system, up to the THz range. A superlattice potential can modify the properties of graphene,355–359 The block-copolymer (BC, i.e. a polymer derived from two (or more) monomeric species), technology360 can be used to create “soft” modulations of graphene, in contrast to “hard” modulations caused by the antidots. In particular, we envisage graphene sheets gently suspended on a regular array of “needles”, fabricated with the BC technology. These novel structures may lead to new phenomena arising from the nature of the modulation, and its tunability. Manufacturing and characterisation of patterned graphene flakes is both a challenging and promising direction of fundamental research, requiring a combination of graphene-specific techniques with methods developed for more conventional materials.

2.8. 2d crystals beyond graphene

There are many examples of 2d crystals. A number of studies5,38 have reported exfoliation of LMs to atomically thin layers. These include h-BN, TMDs,111,361–375 and possible TIs such as (bismuth telluride) Bi2Te3, (bismuth selenide) Bi2Se3 or antimony telluride (Sb2Te3).269,376 Other classes of layered material exist. Examples are transition metal oxides (LaVO3, LaMnO3) transition metal trichalcogenides (NbSe3, TaSe3), transition metal chalcogenide phosphides (Li7MnP4, MnP4), and many others. Each class consists of a range of material types, with its own set of properties. Monolayer MoS2 (1L-MoS2) has a direct band gap377 (while bulk 2H-MoS2 has indirect band gap378) that allows optical applications and, when used in a lateral FETs,379 ION/IOFF up to 108.379 This material has excellent electrostatic integrity that allows flexible electronics,380 and may enable 100[thin space (1/6-em)]000 times less power consumption in standby state than traditional Si transistors.379

Concerning digital electronics Mo and W-based dichalcogenides are 2d semiconductors with band gaps ranging from the visible to the near-infrared (NIR), offering serious perspectives for performant FETs.

The fabrication of the first top-gated SL-MoS2 FETs with RT mobilities ∼60–70 cm2 V−1 s−1, large ION/IOFF (∼108) and low sub-threshold swings (74 mV dec−1) were reported in 2011.379 This was followed by studies focussed on understanding381,382 the charge transport regime (mainly variable range hopping) and band transport regime,383 as well as further optimizing the performances of the devices, with the finding that few-layer samples display the best RT μ > 100 cm2 V−1 s−1.384–386 Still, improvement of the device characteristics are needed, as well as improvement on the overall quality of the 2d materials, together with environmental conditions for achieving non-invasive contact and gating, aspects that are strongly driving the current flow and dissipation.387

However, progress towards realization of functional devices is being made with advances in low power electronics,388 photodetectors,389 low frequency noise devices390 and spintronics391 (see also ref. 392 for an additional discussion).

Reduced dimensionality is a great advantage for electronic applications. Quantum-mechanical confinement usually translates into different electronic and optical properties from the bulk. Mechanical properties are also tuned by ultra-thin thickness, and the large surface–volume ratio would certainly affect chemical reactivity. With this in mind, we foresee potential applications in optoelectronics,393 catalysis,394 batteries,395,396 or supercapacitors,397,398 ultrasensitive sensors for pressure changes,399 gas storage or separation, lubricants,400 and many others. Titania nanosheets could be ideal for ultrathin high-[scr K, script letter K] dielectrics, with maximum values of ε ≈ 125 for thickness of a few nm,401 better than conventional dielectric oxides,402,403 TMOs, in particular MnO2 flakes, have excellent properties for batteries404 and supercapacitors.405,406 They also have photoelectrochemical properties,407 with photon-to-electron conversion efficiencies comparable to those of dye-sensitized solar cells (DSSCs).408 The small thickness may facilitate charge separation of excited e–h pairs, although the low conductivity could favour recombination, if longer migration distances are required. Electrical conductivity can be enhanced by combination with graphene, which could enable high-performance energy storage flexible devices. Moreover, the good mechanical properties,409,410 of some 2d crystals (i.e. BN and MoS2) make them attractive as fillers to reinforce plastics.411 Thin films prepared from the exfoliation of layered compounds may lead to efficient thermoelectric devices.38,396

Ultrathin films of tin (stanene)117,118,123 and phosphorus (phosphorene),116 see Fig. 26, are gaining attention because they have properties of potential interest for the engineering of new concept nanodevices.412

image file: c4nr01600a-f26.tif
Fig. 26 Crystal structure of few-layer phosphorene. (a) Side view of few-layer phosphorene. b,c, Side and top views of 1L phosphorene. Adapted from ref. 116.

Calculations have shown that a phosphorene monolayer is a semiconductor with a direct band gap of 0.9 eV,116 with a strong band gap dependence on N.116 Phosphorene has promise for electronics. μ ∼ 286 cm2 V−1 s−1 at RT was reported in Ref. 116. This is lower than its bulk counterpart (black phosphorus) where the e and h μ are ∼1000 cm2 V−1 s−1 at RT413 and could exceed 15[thin space (1/6-em)]000 cm2 V−1 s−1 for e and 50[thin space (1/6-em)]000 cm2 V−1 s−1 for h at low T.413μ could increase significantly upon surface passivation, in a high-[scr K, script letter K] dielectric environment.414

The electronic properties of silicene and germanene (the Si and Ge equivalent of graphene) have also been studied theoretically412,415–417 both materials being predicted to be gapless semiconductors with linear energy dispersion relations near the K points, like graphene. Structural, vibrational and electronic structure of silicene and germanene grown on various substrates have been studied using DFT, and found to be influenced by their interaction with the underlying substrate.418 See section 4.14 for the production and processing of these compounds.

2.8.1. Characterisation of new 2d crystals. The physical and chemical properties of 2d crystals are yet to be fully investigated. In view of the potential applications, in the short term, priority is to understand and optimize band gaps, electron conductivity, chemical activity, and dielectric, magnetic, mechanical and thermal properties. The role of defects (point defects, dopants, grain boundaries, stacking faults, etc.) and edge terminations must also be addressed. Due to the reduced dimensions and large surface-volume ratio, environment effects could be important and need to be studied, especially for reverse engineering of 2d crystals to build superstructures (section 4.14).

The newly found/produced 2d crystals must be subjected to the same exhaustive studies as graphene. Their structure must be tested using TEM and grazing incidence XRD. Some of these techniques require improvements to get molecular-scale information. Of particular interest would be the development of tools sensible to light elements, enabling chemical differentiation. In this sense, recent developments on annular dark-field TEM419 are very promising. It is most natural to determine the electronic band structure of 2d crystals using ARPES. There are several European facilities with capacity to perform such studies, on a massive scale: BESSY (Berliner Elektronen-Speicherring Gesellschaft für Synchrotronstrahlung) in Germany, Diamond Light Source in the UK, SOLEIL in France, DAFNE (Double Annular Factory For Nice Experiments) in Italy, and several others. Since electronic states in transition metal compounds may feature strong spin–orbit coupling, some of the studies will involve spin-resolved ARPES. Moreover, the development of a scanning ARPES instrument with submicron-resolution is highly desirable for speeding up such studies: this will enable the investigation of free-standing monolayers left upon lift-off of a bulk layered crystal, without the necessity to transfer those onto a substrate.

The studies of individual flakes need to be performed using a broad range of optical techniques, in particular, absorption, reflectivity, ellipsometry and luminescence. Raman scattering can probe phonons,204,205 as well as electronic excitations,204,205 and can be used to determine the N.205E.g., the Raman spectrum of bulk MoS2 consists of two main peaks at ∼382 and ∼407 cm−1 (Fig. 27)204 assigned to E12g in-plane and A1g out of-plane modes, respectively.420 The former red shifts, while the latter blue shifts with the number of layers.421 Moreover, they have opposite trends when going from bulk MoS2 to 1L-MoS2, so that their difference can be used to monitor N.421 However, the trends are not fully understood, and more work is needed to clarify the changes with N. Raman spectroscopy of C and LB modes is also a useful tool to probe these materials.2336 These modes change with N, with different scaling for odd and even N.205 With the increase of N, the frequency of the observed C mode of multilayer MoS2 blue shifts, while that of LBMs red shifts,204 as shown in Fig. 27. The C and LB frequencies, ω(N), of a LM with N layers depend on N as follows: (N ≥ 2N0, and N0 is an integer: 1, 2, 3, 4…). This formula can be generally applied to any LM.205

image file: c4nr01600a-f27.tif
Fig. 27 Raman spectra of 1L-MoS2, 2L-MoS2, 3L-MoS2, and bulk MoS2 measured at 532 nm. Adapted from ref. 204.

Metallic NbSe2 and NiTe2 can be studied using FIR and microwaves. Doped LMs can be subjected to FIR magneto-spectroscopy,422,423 to characterise the effective masses and to develop non-contact methods for quality assessment.

Transport measurements (T dependent resistivity, Hall effect) of individual flakes will require the development of methods for non-destructive deposition of metallic contacts and the implementation of the 2d crystals in FET-type devices. These studies may appear to be material sensitive and will require the development of low-T deposition processes: some of the dichalcogenides start losing Se and Te already at few hundred °C.424 Semiconducting 2d crystals, such as MoS2, offer opportunities for low-power electronics. Because of their atomic-scale thickness and lower dielectric constant than in Si, 2d semiconductors offer higher electrostatic control and could overcome issues related to short-channel effects in Si.425

All of the new 2d crystals will need to be investigated using surface scanning techniques, such as AFM and STM. STM studies can shed light onto the spatial structure of the electronic states near the Fermi level, as well as their accessibility from the environment. STM can also provide information on the electronic structure of defects in dichalcogenides (such as S, Se, or Te vacancies) and on the influence of oxygen on these compounds.

2.8.2. Modelling of physical properties of new 2d crystals. To understand the physical properties of the new 2d crystals, a multi-scale modelling approach is needed for each particular material: a combination of microscopic modelling based on first principles, effective minimal tight-binding models, and effective Fermi liquid theory for electrons at low energies. The existing literature on electronic properties of TMDs addresses band structure,426 Fermi surface nesting,427 and lattice reconstruction in bulk crystals, but very few studies were devoted to isolated monolayers and bilayers of such crystals,428 and the properties of the 2DEGs in them.

2.9. Hybrids of graphene and other 2d crystals

The concept of “materials on demand”: an assembly of graphene and other 2d crystals into hybrid super-structures, Fig. 28, needs to be explored. This would allow the creation of a large number of different multilayers with properties tailored for novel, multitasking applications. From the simulation point of view, there is a need for development of new models of electronic interactions for modelling transport at the mesoscale, effects of disorder, and device simulation. DFT and quantum Monte Carlo simulations can determine the band structure and microscopic charge distribution in monolayers of various TMDs, and CMG. Developments may enable to predict optical properties, electronic correlations, and photochemical reactions, as well as to interpret the experimental findings. Multi-scale approaches such as quantum mechanics/molecular mechanics (QM/MM) can be applied for 2d crystals in solution, or interactions with macromolecules (e.g. for lab-on-a-chip applications that require control on decoration with molecular sensors). New possibilities to find stable TMDs with ferro- and antiferromagnetic properties need to be tested. The electronic properties of monolayers and bilayer of ternary compounds BxNyC1−xy need to be modelled. This can be done both ab initio, and by developing phenomenological models for the alloy properties. Using the input from the ab initio band structure, minimal tight binding models can be developed for each particular compound, and used to describe their optical properties (see section 2.5). Effective Fermi liquid theories for electrons at low excitation energies for each 2d compounds need to be developed and used to analyse transport and correlation properties, including the Landau level spectrum and transport in the QHE regime. Using DFT and effective low-energy theories, the electronic properties of the edges of the layers and states formed around defects can be studied. These low-energy models, in conjunction with group theory can also be used to model the electron–phonon coupling.
image file: c4nr01600a-f28.tif
Fig. 28 From 2d crystals to superstructures.
2.9.1. Electronic transport in lateral and vertical hybrid superstructures. Vertical and lateral transistors are the first and most natural application of atomically thin heterostructures and multilayer systems. Vertical heterostructures and tunnel devices have been used for many years, from the Esaki diode429 to cascade lasers.430 2d-based heterostructures offer a prospect of extending the existing technologies to their ultimate limit of using monolayer-thick tunnel barriers and quantum wells. At the same time, since the doping-dependent screening properties of graphene can be controlled electrically, graphene sheets and thin ribbons in multilayers can be used as gates with widely variable properties, a functionality hardly offered by any other material. New heterostructures will offer unique opportunities to study transport properties of complex, interacting systems (e.g. exciton condensation431,432) and to create transistors with significantly improved transfer characteristics,433 as well as sensors and other applications. Vertical devices can also be scaled to one nm laterally, as far as lithography techniques allow. Tunnelling and resonant tunnelling devices. The feasibility of using multilayer structures for tunnelling devices was demonstrated106 (Fig. 29) showing that BN can act as an excellent defect-free tunnel barrier.106 1L-BN separating two graphene electrodes provides a high-quality tunnel barrier and allows biases as large as 1 V without electrical breakdown.106 The first experiments on exfoliated SLG/BN/SLG structures showed non-linear tunnelling IV curves,106 see Fig. 30. Several other architectures have been realized with modulated tunnel barriers.434–440
image file: c4nr01600a-f29.tif
Fig. 29 I(V) characteristics of a SLG/4L-BN/SLG device. Adapted from ref. 106.

image file: c4nr01600a-f30.tif
Fig. 30 Graphene field-effect tunnelling transistor. (a) Schematic of the proposed devices. (b) The corresponding band structure with Vg = 0. (c) The same band structure for a finite Vg and zero Vb. (d) Both Vg and Vb are finite. The cones illustrate graphene's Dirac-like spectrum. Adapted from ref. 106.

Theoretical understanding of transport properties of such vertical FETs, leading to the full control of their operation, will require a substantial dedicated effort. One challenge is to develop a quantitative microscopic description of single-particle tunnelling processes, based on atomistic approaches. The other is to take into account several factors important for different parts of the I(V) spectrum: orientation mismatch of SLG flakes, contribution of phonon-assisted inelastic tunnelling, and defect-assisted tunnelling.

Aiming at practical applications, further investigations are needed to explore vertical tunnelling structures of graphene, h-BN and materials from the TMD family in various nonlinear electronic elements, such as frequency multipliers. Tunnelling experiments are the first steps towards the production of other devices, as they will allow finding optimum thickness and learning about achievable quality of one-atom-thick barriers.

There are various types of devices where quantum tunnelling may be used (e.g. tunnelling magnetoresistance devices or resonant tunnelling diodes). Atomically thin, smooth and continuous barriers offered by the use of 2d crystals can improve quality and characteristics of any existing or considered scheme involving quantum tunnelling. Investigation of resonant tunnelling is a logical continuation of the tunnelling experiments. Modulating the tunnelling barrier height by using different materials (e.g. heterostructures like SLG/h-BN/MoS2/BN/SLG or SLG/h-BN/SLG/BN/SLG) may create additional states in the barrier, which would allow resonant tunnelling, see ref. 106. Such devices are most interesting to get negative differential resistance conditions, useful for various non-linear components. Resonant tunnelling through impurities and defects enables to map the wavefunction of the latter.441

Vertical graphene-based structures represent a new approach to develop functional electronics. Rapid response and ultra-small sizes could be achieved in vertical transistors. Indeed, electron transfer through nm thick barriers can be extremely fast (and, possibly, coherent). Ballistic tunnelling transistors may allow one to overcome the most significant drawback of GFETs: the low ION/IOFF. However, performances achieved to date are poor: ION/IOFF is smaller than 20, while the delay time is four orders of magnitude larger than that expected from ITRS.12 The tunnelling devices would have a highly insulated off state with no dissipation, which should allow not only individual transistors but integrated circuits at RT.442 The latter is difficult to achieve for horizontal transport in graphene and remains a distant goal. The ideas currently under consideration include several architectures for tunnelling/hot electron transistors. The simplest is metal/h-BN/SLG/h-BN/SLG, where the metal contact (separated from the bottom graphene by thick, tunnelling non-transparent h-BN) serves as a gate and two SLG (acting as emitter and collector) are separated by thin h-BN layer. The operation of the device relies on the voltage tunability of the tunnelling density of states in graphene, and of the effective height of the tunnel barrier adjacent to the graphene electrode. It is interesting to experiment with several different dielectrics in heterostructures, such as metal/h-BN/SLG/MoS2/SLG. Higher quality heterostructures and dielectrics with smaller tunnelling barrier might bring ION/IOFF to 105–106, as required by modern electronics.

Another interesting idea is to attempt the development of a hot electron transistor, similar to those discussed in semiconductor electronics.443 Few-atom-thick transistors based on a 2d tunnelling barrier and graphene may allow much better quality, and become more successful in applications. The transit time through such sandwiches is expected to be ≪1 ps,443 whereas there are no limits for scaling down in the lateral dimension to true nm sizes. In a metal/h-BN/SLG/h-BN/SLG system, the thickness of the active part of the devices would be less than 10 atoms (∼3 nm) and should allow a ballistic current controlled by the central graphene electrode. The assembly of 2d crystals into superstructures may allow stacks of several transistors in series (metal/h-BN/SLG/h-BN/SLG)N in a vertical integrated architecture. Light emission and photovoltaics. Superstructures of 2d crystals can be used to develop tunnelling LEDs and photovoltaic cells. Here, we refer to superstructures composed of two conducting layers separated by a barrier with a modulated profile. Quantum wells are one example of heterostructures made by joining, directly at the atomic level, different materials, usually in layers.444,445 A fundamental experimental quantity is the “band offset ratio”, i.e. the ratio of the difference in conduction band energies to the difference in valence band energies.444,445 Heterostructures for which the difference in the band gap energies is largely dominated by the conduction band offset are called “Type I” quantum wells.444,445 Those where e and h have their lowest energies in different materials are called “Type II” quantum wells.444,445 Type-I quantum wells (Fig. 31) can be used for injecting e/h, with subsequent recombination leading to light emission.
image file: c4nr01600a-f31.tif
Fig. 31 SLG/h-BN/MoS2/h-BN/SLG structure: blue is the valence band and pink is the conduction band. Adapted from ref. 106.

Using more complicated structures, both type-I and type-II quantum wells of various configurations can be created. As the band-structure of 2d crystals depends on N, by changing the thickness of one component, one could tune the optical properties. Using various thicknesses of different materials one can target LEDs of different colours. As individual heterostructures can be combined in one stack with individually contacted layers, LED at different wavelengths could be combined in one structure.

Similarly, PV devices may be created by placing two metallic 2d crystals (e.g. graphene) within tunnelling proximity of each other. By applying a bias (or by exploiting the proximity effect of other metallic 2d crystals) electric field could be created inside the barrier. Any e–h pair excited by light will be separated and contribute to photocurrent.106 Similarly to the case of LED, it should be possible to create heterostructures with various band gaps, sensitive to photons of different energies. Moreover, plasmonic nanostructures can improve the performance of graphene photovoltaic devices.446 In situ characterization methods. Hybrid systems will require advanced characterization, which should involve both high spatial and/or point resolutions and coupling, preferentially in situ. This will limit the contact with air, preventing contamination. Two kinds of in situ coupling are worth considering: (i) coupling several characterization methods to investigate a single object, e.g. HRTEM + Raman spectroscopy + electrical measurements, in order to accurately correlate the structural features and the physical behaviour. (ii) Coupling one or several characterization methods (e.g., HRTEM imaging and electrical measurements) with one or several treatment methods (e.g., mechanical and/or thermal stresses) in order to correlate the variation of the behaviours with the structure changes. Considering in situ TEM experiments, the above could be achieved by using sample holders equipped with various facilities (e.g., able to apply thermal or mechanical stresses). However, there is the need to develop other sample holders that will allow a larger panel of possible tests to be applied to the samples under study. When the in situ coupling is technically difficult (e.g., coupling TEM and UV-Raman, which cannot be done through an optical fiber, or coupling TEM and high magnetic field inducer), a chip-based sample holder technology will be highly preferred. Indeed, the latter will allow the chip to be transferred from a characterization system to another, each of them equipped with the appropriate sample holder bearing the same in situ treatment methods. This will allow various investigations on the same samples under the same conditions.

The chemical functionalization of graphene with reactive molecules and the deposition of supramolecular assemblies require studying the self-organization process and the molecule/graphene interface in several conditions: At the liquid–solid interface, by wetting–dewetting processes, but also in connection with ultra-high vacuum (UHV) conditions. This is already possible with UHV systems combining several sources of molecule deposition (sublimation, liquid-valve injection), with surfaces characterization techniques such as STM and X-ray and UV Photoelectron Spectroscopy. Several STM techniques are currently used: low temperature STS, Spin-polarized STM and Fourier-Transform Scanning Tunnelling Spectroscopy (STS), which can allow a local dispersion and surface Fermi measurement. Synchrotron sources can be used for high resolution ARPES, X-ray magnetic circular dichroism (XMCD) and also spin-polarized low-energy electron microscopy, particularly useful for the ferromagnet/graphene interfaces. Hybrid structures for active plasmonics. Combinations of 2d heterostructures with plasmonics would allow for creation of active optical elements. 2d heterostructures are ideally suited to be used with plasmonic structures, as they can be positioned exactly at the maximum of electric field. Such elements are of great importance in different areas of science and technology: from displays, to frequency modulators. Despite great progress, active optics still relies heavily on either liquid crystals,447 which guarantee deep modulation in inexpensive and small cells, but are quite slow,448 or non-linear optical crystals,449 which are fast,450 but bulky and expensive.450 Thus, inexpensive, fast and small active optical elements would be of considerable interest.

The target is the design and fabrication of a new generation of active plasmonic metamaterials with optical properties electrically controlled by 2d heterostructures. Plasmonic metamaterials of various configurations could be achieved by sandwiching between SLG/h-BN heterostructures (see Fig. 32). The conductivity of graphene can be changed by at least two orders of magnitude with electrostatic doping.451 This could modulate the optical properties of the under-lying plasmonic structure. The combination of 2d heterostructures with plasmonics could result in fast, cheap and small active optical elements.

image file: c4nr01600a-f32.tif
Fig. 32 Active plasmonic structure: Au dots sandwiched between SLG/h-BN layers.

2.10. Multiscale modelling of graphene-based structures and new 2d crystals

The optimisation of active plasmonic materials will require multiscale modelling of their properties, taking into account plasma modes and single-particle excitations, as well as their coupling with flexural vibrations of individual layers. Further studies should include modelling of heating of superstructures and their cooling by lateral and vertical heat transfer.

Modelling of physical properties of new 2d crystals and hybrid devices constitutes an important research direction. A specialised effort is needed to provide timely interpretation of the characterisation of new 2d crystals, assessment/prediction of functional properties, and guidance of technological effort in creation of hybrid structures (‘materials on demand’, see section 4.15). This will require the implementation of a multiscale modelling approach, in which the materials band structure and local microscopic parameters computed using ab initio simulations are incorporated in the mesoscale description of electronic transport, thermal, mechanical, and optical characteristics, which then enter into the finite-element modelling of operational devices or technological processes.

Many fundamental questions are open in the field of nano-electronics and new materials. Due to the complexity of both, these cannot be answered by conventional simplified approaches. The research of novel functional materials is highly interdisciplinary covering the domains of chemistry, material science, physics, and engineering with their methods and scope of length scales. Advanced knowledge of such fields has necessarily to be combined. In addition, the complexity of quantum laws in nano-electronics complicates upscaling attempts to the point that, at the cross-road of new materials and nanoelectronics (especially for beyond-CMOS applications), only multiscale modelling approaches can progress knowledge sufficiently fast in the near future. The development of the necessary theoretical and computational methods, as well as improvement of the existing technologies, is therefore needed and critical to the STR.

2.10.1. Ab initio computations. The theoretical exploration of structural and electronic properties of GRMs requires the extensive use of first-principles (or ab initio) computational methods. Especially, the modelling of material imperfections, such as structural defects (vacancies, Stone-Wales, grain boundaries), chemical modifications of the material surface (adatoms, molecular adsorbates), substrate or contact effects becomes crucial at the nanoscale, and such task can only be achieved using the proper level of electronic structure description, including exchange and correlation effects with DFT, or extensions, such as the GW-correction452 and the time-dependent density functional theory (TDDFT),453 for grasping excited states and optical spectra or low dimensional transport. The portfolio of such ab initio methods (mostly developed by European consortia) is rich and diversified, offering several levels of accuracy and scalability that will be appropriate depending on the system under study.

We provide a brief description of the general contours of the various levels of approximations to deal with electron correlations, together with a list of useful codes and a brief technical description of their respective capabilities, in the context of GRM studies. Most common ab initio methodologies encompass the following approaches, all of which will require further development: Hartree–Fock theory (H–F)454 and Post-Hartree–Fock (PH–F)455 quantum chemistry methods. Usually, H–F has insufficient accuracy for most purposes, but it provides a good starting point for quantum chemistry configuration interaction methods, Møller–Plesset456 perturbation theory and coupled-cluster457 methods. PH–F is highly accurate, but scales poorly with system size.458 Such codes generally allow optimisation of geometry and calculation of a wide range of other properties.

DFT is the most “standard” ab initio approach, its main limitation being band gap underestimation.459 Either plane-wave or localised basis sets can be used, but this choice determines the system size that can be simulated. The computational cost of conventional DFT calculations scales with the cubic power of the number of atoms, O(N3), but linear-scaling [O(N)] DFT methods have been developed for insulators.460 DFT results can be mapped to tight-binding, to enable transport calculations (see, e.g., ref. 461, 462). Most DFT codes allow molecular and lattice dynamics calculations. The long standing difficulty to deal with vdW forces463 was recently overcome,464 and several codes, such as AB-INIT or SIESTA (described below) now offer functionals that allow for an accurate description of vdW, a critical issue when dealing with electronic properties of LMs (such as BN/graphene, graphene/graphene, etc.).

GW is the approximation to the self-energy of quasi-particles in a many-body system, to obtain accurate excitation energies.465 The methods generally scale as O(N4) but ref. 466 reported a O(N3) implementation.467 GW can use the Bethe–Salpeter equation468 to describe excitonic effects,469,470 but scaling is O(N6).

TDDFT is the time-dependent formulation of the Kohn–Sham,471,472 equations, to calculate response functions.472 Poles of density–density response function correspond to excited states. TDDFT is cheaper, but less accurate than GW, for determining excited-state energies.

Quantum Monte Carlo (QMC) is the most accurate total energy method for condensed matter physics.473 Scaling is O(N3), like standard DFT, but with a much greater prefactor.473 QMC is able to exploit massively parallel computers. Excitation energies are accurate, but must be calculated one by one as differences of total energy. QMC molecular and lattice dynamics will be available in the near future.

There is a manifold of codes suitable for the study of structures and electronic features of GRMs. Among the most accurate DFT methods, usually based on plane-waves basis set, one can consider AB-INIT, CASTEP, VASP or ADF described below.

AB INITio (ABINIT):474 A package that allows to calculate a broad number of molecular and periodic solids’ properties as total energy, electronic structure and charge density, optimized geometries and dynamic simulations using calculated DFT forces and stresses, dynamical matrices, Born effective charges, dielectric tensors, excited states etc. The programs use codes based on DFT, Density-Functional Perturbation Theory(DFPT), Many Body Perturbation Theory (GW approximation and Bethe–Salpeter equation) and TDDFT.

CAmbridge Serial Total Energy Package (CASTEP):475 a full-featured code based on a first-principles quantum mechanical description of electrons and nuclei. It uses the robust methods of a plane-wave basis set and pseudopotentials.

Vienna Ab initio Simulation Package (VASP):476 a plane-wave all-electron code using the projector-augmented wave method to describe the electron–core interaction. The code uses fast iterative techniques for the diagonalization of the DFT Hamiltonian and allows total-energy calculations and structural optimizations for systems with thousands of atoms and ab initio molecular dynamics simulations for ensembles with a few hundred atoms extending over several tens of ps.

Amsterdam Density Functional (ADF):477,478 A suite based on DFT that uses Slater-type orbitals479 as basic functions to make electronic structure calculations, also able to perform TDDFT. ADF can calculate various molecular properties like IR, Raman, VCD (vibrational circular dichroism), UV and XAS (X-ray absorption spectroscopy) spectra, NMR and EPR (electric paramagnetic resonance) parameters, solvent and environmental effects.

Such methods are limited in terms of system size, and mainly restrict the exploration of material models few hundreds of atoms large. O(N) methods are implemented using localized basis sets, gaining in upscalability but at the cost of a reduced accuracy. Among the most useful: SIESTA, CONQUEST, ONETEP, QUANTUM ESPRESSO, FLEUR, OCTOPUS.

Spanish Initiative for Electronic Simulations with Thousands of Atoms (SIESTA):480 a method and its computer program implementation, to perform efficient electronic structure calculations and ab initio molecular dynamics simulations of molecules and solids. SIESTA's efficiency stems from the use of strictly localized basis sets and from the implementation of linear-scaling algorithms which can be applied to suitable systems. Its accuracy and cost can be tuned in a wide range, from quick exploratory calculations to highly accurate simulations matching the quality of other approaches, such as plane-wave and all-electron methods. VdW interaction has been recently implemented.481

Concurrent O(N) QUantum Electronic Simulation Technique (CONQUEST):482 A linear-scaling code based on DFT, to perform electronic structure calculations on very large systems, up to millions of atoms, with the possibility of choosing different levels of accuracy.

Order-N Electronic Total Energy Package (ONETEP):483 a linear-scaling code for quantum-mechanical calculations based on non-orthogonal generalized Wannier functions,484,485 expressed in terms of periodic cardinal sine functions, in turn equivalent to a basis of plane-waves. ONETEP combines the advantages of the plane-wave approach (controllable accuracy and variational convergence of the total energy with respect to the size of the basis) with computational effort linear with the size of the system.486

QUANTUM opEn Source Package for Research in Electronic Structure, Simulation, and Optimization (QUANTUM ESPRESSO487): an integrated suite of codes for electronic-structure calculations and materials modelling, based on density-functional theory, plane waves, and pseudopotentials (norm-conserving, ultrasoft, and projector-augmented wave). QUANTUM ESPRESSO builds upon electronic-structure codes developed and tested by the original authors of novel electronic-structure algorithms, and applied over twenty years by some of the leading materials modelling groups worldwide.

Jülich FLAPW code family (FLEUR):488 A code designed to calculate properties of solids, based on the Full-potential Linearized Augmented Plane-Wave (FLAPW) method.489

General Atomic and Molecular Electronic Structure System (GAMESS-UK):490 Gaussian-basis quantum chemistry code, with a variety of techniques for PH–F calculations.

OCTOPUS:491 A program aimed at the ab initio virtual experimentation on, in principle, and ever-increasing range of system types. Electrons are described quantum-mechanically within DFT, or TDDFT, when doing simulations in time. Nuclei are described classically as point particles. Electron–nucleus interaction is described within the pseudopotential approximation.

Other methods are more suitable for dealing with dynamical properties of materials, including the formation of stable structures and or the highly accurate description of electron–electron correlations. Those include:

Cambridge quantum Monte Carlo computer program system (CASINO):492 A code based on QMC algorithms for continuum electronic structures calculations using Slater–Jastrow493 pairing, and backflow wavefunctions or combinations of them. It is also possible to extend the range of wavefunctions available.

Cornell-Holland Ab-initio Materials Package (CHAMP):494 A based on QMC algorithms designed to perform electronic structure calculations on a wide variety of systems, such as atoms, clusters, solids, nanostructures and molecules.

Car–Parrinello Molecular Dynamics (CPMD):495 A plane-wave and pseudopotential DFT code designed for performing molecular dynamics simulations, which can be used to perform TDDFT calculations.

CRYSTAL:496 A code designed to compute the electronic structures of periodic systems based on H–F, DFT or hybrid approximations. Atom centred Gaussian functions form the basis set that is linearly expanded into Bloch functions.

2.10.2. Mesoscale modelling. Despite the intensive software development, the computational cost of ab initio methods still remains very expensive. This originates from the complexity of electrons represented by wave functions, hence possessing an inner structure with widely variable properties compared to simple particles used in classical molecular dynamics. The requirement of self-consistency is only one consequence of the quantum nature, which slows down such methodology. Although many concepts exist to weaken this impact, there is a practical limitation to sizes of systems, at present ∼1–2 nm, treatable ab initio. This, however, is not the length scale on which one discusses functional materials (at least one-two orders of magnitude above).

The characteristics and fundamental properties of interest for functional materials are co-defined on a larger length scale beyond the ab initio scope. This is because they are additionally influenced by other facts governing properties on such length scales, such as low concentration dopants, impurities or structural defects, or simply because the relevant structures may reach these dimensions themselves. Special attention has to be paid to the interfaces or interaction of layers in multilayer systems.

Multiscale modelling497 is capable of bridging length scales. The concept is based on the observation that not all interactions must necessarily be treated within the first principles framework. This allows one to introduce a hierarchy of interactions, which might be founded either on very general considerations or just adapted and valid for the presently studied properties. Based on this, a hierarchy of levels of treatment may be introduced. The lowest (microscopic) level deals with the smallest objects at the highest accuracy. It can be identified with the full ab initio level. Multiscale modelling defines first the models on each level and second the interfaces for transferring relevant information to the respective upper (or even lower for feedback loop) level, where they are further processed. The advantage is that not all information available on the computationally heavy lower level enters the upper-level modelling, but only relevant condensed information, which is precisely where multiscale modelling benefits from. In addition, modelling of interactions on the upper macroscopic level replaces respective couplings on the more refined lower level. This allows one to reduce the work at the lower level by treating smaller parts (non-interacting subsystems) there. E.g., a finite range impact on electrostatics and on electronic properties can be expected from impurities or dopants depending on the local surrounding of a host crystal. Additional long-range parts such as the Coulomb interaction might be separable and can be treated on the upper level. The information on local electronic properties can still be obtained with massively parallel ab initio methods using large supercells. On the other hand, the evolution of a system as a whole, composed of millions of atoms, including a certain distribution of such dopants, is unpredictable by ab initio methods when the whole system is included at the same level. The problem is solved by combining local ab initio codes with specialised solvers addressing mesoscale range of distances. Such combined codes include:

Learn on the Fly [LOTF]:498 hybrid quantum/classical molecular dynamics which combines quantum-mechanical embedding and classical force model optimization into a unified scheme free of the boundary region, and the transferability problems which these techniques, taken separately, involve. It is based on the idea of augmenting a unique, simple parametrized force model by incorporating in it, at run time, the quantum-mechanical information necessary to ensure accurate trajectories. The scheme was tested on a number of Si systems composed of up to approximately 200[thin space (1/6-em)]000 atoms.499

SMEAGOL:500 DFT-based transport code which calculates transport properties of atomic scale devices, using state of the art electronic structure calculation schemes.

Transport-Spanish Initiative for Electronic Simulations with Thousands of Atoms [TranSiesta]:480 Extension of SIESTA with recursive Green functions solver. The TranSiesta method is a procedure to solve the electronic structure of an open system formed by a finite structure sandwiched between two semi-infinite metallic leads. A finite bias can be applied between both leads, to drive a finite current.

Tight-Binding Simulation [TB_sim]:501,502 tight-binding code which can be interfaced with SIESTA for more realistic descriptions of structures and disorder.503 It is able to compute the structural, electronic, optical and transport properties of various nanostructures.

For a broader range of applications, including electronic and heat transport, optics, optoelectronics, thermomechanics, etc., alternative hybrid codes will need to be developed.

The treatment of disorder deserves particular attention. Indeed, to design a realistic scenario of the influence of disorder on transport properties of materials and devices one has to consider different length scales and modelling strategies simultaneously. First, a microscopic picture of the atomic structure is necessary to access electronic properties. This can only be provided with state-of-the-art first principles simulations. These can be carried out using simple unit cells in clean systems. When considering crystal imperfections or dopants, larger supercells with few impurity atoms or defect sites are necessary.

The interface part is an essential ingredient of the modelling. It defines which information is exchanged, i.e. which features of the ab initio-simulation are strong enough to be important on the macroscopic length scale. From these simulations one extracts electronic structure parameters which represent at best the interactions at this level. For the efficiency of the multiscale approach it is very advantageous if the extracted parameters are generic. This should be considered when setting up the modelling strategy to reduce or, at best, avoid feedback effects. Quantum transport in chemically modified and disordered graphene-based materials can be successfully simulated by means of novel types of O(N) algorithms implemented in the Landauer–Büttiker504 or Kubo–Greenwood505 formalisms (see, e.g., ref. 503, 506, 507). New directions of work should include the effect of spin–orbit coupling, T effects or high bias regimes.

2.10.3. High performance computing. In the longer term, having a strong and realistic simulation capability will provide a strategic tool to support product development in all fields of applications. In this perspective, supercomputing will be very useful for material characterization and device simulation at a realistic level. The use of High Performance Computing (HPC) is certainly necessary for making advances in frontier developments in the fields of first-principles calculations and multiscale methodologies. To make these new computational schemes useful, supercomputers must be used intensively. However, one should note that the grand challenge is not just a question of computing faster and faster. E.g., for simulating complex graphene-based devices (NEMS, sensors, transistors, or circuits), in addition to high performance computing, more physical inspection of models and suitable use of first principles are needed.

GRM research will greatly benefit from a connection to supercomputers facilities, especially for material simulation to serve as guidance applications. Improvements are required for multiscale modelling and reverse engineering, such as decreasing the problem complexity, ideally to O(N). In DFT this has already been achieved for insulators with the advent of O(N) codes.508 For other methods it is a far greater challenge. Furthermore, it is essential to add functionality (the ability to calculate a range of properties) to these codes, whilst retaining the favourable scaling.

Other improvements include increasing the ease of use (availability of documentation, support, examples, well-designed and clear input and output, ease of installation, etc.), robustness (lack of bugs and numerical reliability), and “reverse engineering” tools to discover the technological principles of a device, object, or system through analysis of its structure, function, and operation.509

Broad cooperation between groups authoring the existing codes is needed, for comparison between codes, for testing and optimisation purposes, development of better interfaces between codes, including better data standards, developments of methods that can exploit massively parallel computer facilities, development of “code libraries” and allowing problems to be distributed among members of a community.

Developments made to first-principles algorithms and codes will be equally useful in other areas of physics, chemistry and materials science. First-principles codes and algorithms aim to be general.

2.10.4. Further development of field-theory and kinetic theory methods. Analytical methods will play an equally important role in the multiscale modelling of GRMs, providing a consistent scalability of materials parameters and a systematic description of GRMs in strongly non-equilibrium states.

The effects of e–e correlations, as well as the interplay between e–e interaction and disorder can be efficiently addressed using the renormalisation group approach.510 The latter enables one to follow non-trivial (non-linear) scaling of materials parameters upon variation of the length scales at which the electron system is studied. It is based upon the microscopic input (phenomenological or provided by DFT modelling), permitting one to formulate a medium-energy effective field theory describing the physical system, and it is followed by the analysis of the renormalisation flow of essential constants in the theory, which may (or may not) be length scale dependent. The renormalisation group approach can be applied to the studies of both quantum effects, like localisation in graphene, the interactions-driven phase transitions into states with spontaneously broken symmetry, and to studies of classical problems related to the renormalisation of flexural deformations.

The development of efficient methods of kinetic theory is also necessary, in order to treat graphene in strongly non-equilibrium conditions. As a thin material with a weak coupling to the substrate/environment, graphene may easily overheat.511 Moreover, the energy relaxation of non-equilibrium carriers in graphene (photoexcited or injected by tunnelling) requires extensive modelling effort.512,513 Tackling graphene at high current, reaching breakdown, is a more challenging theoretical problem, which requires the use of a combination of the microscopic dynamics computation with the mesoscale kinetic theory.

2.10.5. Correlations in multiple graphene layers. e–e interaction is relevant in BLG,514 where novel phases are expected.515 A similar situation is expected in FLG with (translational or rotational) stacking faults.

Apart from fundamental questions, a deeper understanding is relevant for potential applications, because of possible bandgaps induced by correlations that may compete with gaps in graphene nanostructures due to their spatial confinement.

The role of electronic correlations in GRMs is still poorly understood, and its full investigation will thus be a major challenge over the next years. e–e interactions might lead to novel, yet unexplored phases, with possible magnetic order or unusual topological properties, due to an expected time reversal symmetry breaking. The interplay between topology and interactions opens a new research field in theory (advanced analytical and numerical techniques, such as Dynamic Mean Field Theory-DMFT-extensions)516 and experimental physics, where novel experimental techniques are required for probing these phases (e.g. Kerr rotation measurements517).

2.11. Graphene for high-end instrumentation

2.11.1. Graphene for high energy physics instrumentation, Tokamaks and Stellarators. In addition to the impermeability of graphene to gases101 or liquids,101 some other properties of graphene, such as its resistance to radiation damage,518 may be used for high-end applications such as devices for high-energy physics. These characteristics can be used in various fields, in containers used for nuclear repositories, or high energy experiments that require ultra-low background noise. Specifically, graphene could be used as the container of Liquid Xe in experiments searching for dark matter or double beta-decay without neutrinos. Graphene hardly contributes to the background noise due to presence of small traces of radioactive elements (radiopurity). It absorbs less than 10% of UV light (10% at 266 nm[thin space (1/6-em)]519,520), the sparkling wavelengths for Xe521 when a charge particle passes through, and it is an excellent conductor, which enables the container itself to act as detector.

Another option is to the substitution of graphite as first wall in nuclear fusion reactors (as the Tokamak ITER -International Thermonuclear Experimental Reactor-,522 presently under construction). Graphite is used to prevent the injection of heavy ions523 with a large number of electrons, as impurities into the plasma. These ions are the result of sputtering processes at the reactor walls. Graphite can only provide light C ions as impurities, which translate into lower losses of the plasma energy. Graphite-based thin films, with low Secondary Electron Yield (SEY), a limiting factor for the achievement of high luminosity in accelerators for positively charged particles, have been developed at CERN.524

2.11.2. Graphene for metrology. Quantum resistance. Electric resistance measurements in metrology are always referred to the von Klitzing constant (25[thin space (1/6-em)]812.807449 Ω),150,525 measured in QHE experiments of semiconductors.150 The fact that graphene is an ideal 2d electron system and not only a quasi-2d system, such as that realized in semiconductor quantum wells and heterostructures, paves the way for new applications of QHE phenomena. The existence of a RT QHE in graphene276 demonstrates that the cyclotron gap can be larger than in any other material, since this gap is not limited by higher electric subbands. The combination with ferromagnetic materials may allow RT applications of quantum Hall phenomena, like quantum resistors or dissipationless current transport.526 The peculiarity of graphene that simultaneously p- and n-type regions can be present in one device allows new designs for quantum Hall resistance standards.527

The QHE is observed in graphene at lower magnetic fields and higher currents than semiconductor QHE devices.150 This may enable a new, simpler and cheaper QHE standard of resistance, that would spread to any metrology institutions and calibration benches (Fig. 33). The von Klitzing constant could be measured in graphene and compared to semiconductors, enabling validation of the von Klitzing relation150 with a quantum Wheatstone bridge,528 a circuit made of two series arrays of two Hall bars parallel to each other.

image file: c4nr01600a-f33.tif
Fig. 33 7 × 7 mm2 wafer of SiC with 20 Hall bar devices. The contact configuration for the smaller device is shown in the enlarged image.154 Quantum current standard. Due to the robust Coulomb blockade in GQDs, those can be used in SET nano-circuits.329,343,344 When driven at high-frequency, such SETs can be operated as single electron turnstiles,529 which pass exactly one electron charge per each external field cycle. This enables a fundamental current standard based on charge quantization.329,343,344 Standard for optical absorption coefficient. Another use of graphene in metrology is as a standard for the optical absorption coefficient (≈2.3% for SLG in visible light990). It is for some time now that CNTs are used in radiometry530 and photometry.531 The useful property in photometry is the high absorption in a broad spectral range, from UV to IR and THz. CNTs are, therefore, interesting as coating in thermal detectors, and this enables the measurement of optical power in spectral regions where there are no detectors based on semiconductors.532 The extension of these applications to graphene is the next step.

2.12. Perspectives

Fundamental and blue-sky studies of GRMs have a twofold aim. First, the fundamental mechanisms that determine and may limit their potential in already foreseen electronic and optoelectronics applications should be uncovered. Second, the next generation of GRM-based nanostructures for the development of electronic devices beyond CMOS needs to be developed. Research is needed to gain in-depth understanding of microscopic properties of defects in crystal graphene, GBs in polycrystalline graphene, or the influence of environment (such as various substrates), all factors jeopardizing the performances of mass-produced devices. The engineering of electrical circuits incorporating GNRs and GQDs will be essential to exploit the new functionalities of graphene in high-end instrumentation and metrology.154 The study of 2d crystals beyond graphene5,38 is also key to enhance graphene's properties by combining this material with monolayers of 2d crystals in superstructures, which will allow broadening of the range of functional applications of graphene and hybrid superstructures in post-CMOS electronics.

The objectives foreseen in the next few years are:

• To establish the fundamental limits for functional graphene nanostructures in electronics beyond CMOS.

This can be achieved via microscopic characterisation of single- and poly-crystalline graphene and graphene-based nanostructures, studies of kinetic processes, influence of defects, disorder, and influence of substrate/environment on electronic properties.

• To explore the use of LMs in electronics.

This will include microscopic characterisation of electronic properties of 2d crystals, studies of their transport and vertical tunnelling properties, and multi-scale modelling of graphene-based and hybrid structures.

The long-term objectives are:

• To exploit graphene for both classical and quantum information processing in the post-CMOS era.

For classical information processing, the development of vertical transistors and atomic scale metal–semiconductor field effect transistor (MESFETs) where graphene is used as active component, as interconnect, or transparent gate is a key objective. For quantum information processing, the exploitation of the long-lived spin coherence of electrons in graphene is needed to develop monolayer and bilayer QD qubits, readouts, and their scalable circuits.

• To build graphene-based metrological applications and high-end electronic instrumentation.

For metrology and standardisation, it is fundamental to develop transferable table-top quantum resistance standards based on the quantum Hall effect, and stable quantised current sources. High-end instrumentations will include highly sensitive bolometers, ultra-sensitive Hall probes, and new types of scanning probes.

• To produce new inorganic 2d crystals, systematically investigate their physical properties, and evaluate the potential for optoelectronic applications.

Using the experience developed with graphene, the long-term vision is to explore a broad range of LMs beyond graphene. The issues that need to be addressed will include material stability, compatibility with nanofabrication processes, and the entire range of structural, optical and electrical characterisation.

• To study electronic, optical, and thermo-mechanical properties of hybrids combining graphene with various 2d crystals.

The microscopic charge transfer and transport in multilayer hybrid structures, optical processes and hot carrier dynamics, electron spin memory and relaxation need to be studied. Understanding these processes will enable the development of devices such as transistors and switches, LEDs, PV cells and photo-detectors.

The timeline is shown in Fig. 34. Timescales: 23 years: understanding fundamental kinetic processes and influence of defects. 47 years: Understanding of electronic, optical, and thermo-mechanical properties of hybrid structure combining graphene with various 2d crystals that have electronic bandgaps and establishing their fundamental limitation. 710 years: Integration of vertical hybrid devices and development of graphene-based metrological systems and high-end electronic instrumentation.

image file: c4nr01600a-f34.tif
Fig. 34 Timeline for fundamental research and development of materials and components. Key technological targets for the three main areas: Science and Technology, Components & Circuits and Systems (reported on Y axis). The forecasts for when the targets are predicted to be met can be read on the X axis. The horizontal arrows indicate the time evolution of each specific target, while the dotted lines indicate how specific targets are linked or are necessary for others inside the same or different STR area. The symbols indicate the involvement and/or the benefits of/for other themes and/or applications.

3. Health and environment

Nanosafety, defined as all the safety issues associated with nanotechnology, is crucially required to translate any future development of new nanotechnologies into action, from industrial applications to health care approaches.533–536 GRMs are not devoid of possible risks on health or environment, and cannot be excluded from these two domains of investigation. It is of fundamental importance to explore the level of toxicity and to establish, if required, constrains for safety of use.537–542

The GRM small size and unique physico-chemical properties may pose potential health and environment risks. Determining and resolving safety and toxicity issues will not only be beneficial for integration into devices, but also in light of possible biomedical applications, such as direct nano-interfacing devices with cells and tissues.

Initial studies have assessed the in vitro and in vivo toxicity of graphene and graphene oxide (GO), while for other 2d crystals very little has been done to date. Some of these studies showed that there are no particular risks for using GRMs, conversely others have evidenced that some forms might become a health hazard.537–542 When manipulating a nanomaterial, inhalation is one of the possible ways of contact. Some GRMs have aerodynamic size that may lead to inhalation and deposition into the respiratory apparatus with implications on the formation of granulomas and lung fibrosis.543 On the other hand, the possible biological effects and responses can differ depending on the chemico-physical properties (i.e. N, lateral size, stiffness, hydrophobicity, and surface functionalization), the administered dose and the purity. Limited data542 are available on the difference in biological behaviour between large (few micron lateral size) and small (few hundred nanometer later size) sheets of graphene, or SLG versus FLG. These aspects certainly deserve thorough studies.

3.1. In vitro impact

Ref. 544 reported that graphene induces stronger metabolic activity than single-wall carbon nanotubes (SWNTs) on neuronal cells. Signs of cell membrane damage associated to necrosis (i.e. damage resulting in the premature death of cells in living tissues and organs),545 apoptosis (i.e. programmed cell death induced by biochemical events that lead to changes on cell morphology)546 and reactive oxygen species (ROS), generated in a concentration and time dependent manner, were measured.544 The comparison between graphene and CNTs indicates544 that the shape of the material plays a primary role in determining its biological effects. Ref. 547 tested GO and carboxylated GO (i.e. GO after acid treatment to generate more COOH groups) on monkey kidney cells. At a concentration of 25 μg ml−1, GO accumulated mainly at the cell surface provoking significant destabilization of the cytoskeleton (i.e. the cellular scaffolding contained into the cytoplasms that plays an important role in intracellular and cellular division).547 Carboxylated GO was instead internalized by the cells, accumulating in the perinuclear region, without affecting the morphology of F-actin filaments (a protein, found in all eukaryotic cells, that forms microfilaments).547 GO affected the cell viability at 50 μg ml−1, while carboxylated hydrophilic GO was not toxic up to 300 μg ml−1.547 The lack of a careful definition of the materials (e.g. calling pristine graphene what is instead GO) creates confusion when cytotoxic profiles of different nanocarbons are to be compared. Other studies on different cell lines reported dose and time dependent GO effects on cell viability only at concentration above 80 μg ml−1,548–550 and changes in cell morphology only after long time culturing.551Ref. 552, 553 reported that GO induced time and concentration dependent cell mortality, as documented by the early and late presence of apoptotic cells.

To expand the analysis of the cellular responses induced by GO, sheets of different lateral sizes (i.e. 350 nm and 2 μm) were tested.554 Two phagocytic cell lines (i.e. with the capacity of engulfing solid particles) were able to internalize nano-sized and micro-sized GO.554 Little differences on cell viability were observed up to 20 μg ml−1.554 However, the presence of manganese as contaminant, due to inaccurate purification procedures during the oxidative process to prepare GO, induced a GO-independent high cell death.554 This shows how accurate purification steps are relevant in avoiding false positives, that might be erroneously associated to an undesired effect of the nanomaterial itself. In addition, in the phagocytes tested, the uptake and sub-cellular localizations of the two GOs of 350 nm and of 2 μm, were different. The larger GO sheets induced also a stronger inflammatory response. The different behaviours of the two types of GOs can be exploited on different biomedical contexts. The high inflammatory responses of micro-sized GO (2 μm) can be used to promote adjuvant effects (that modify the effect of other agents) in vaccine systems to strengthen weak immune responses. The low inflammatory profiles associated to nano-sized GO (350 nm) can be useful for applications in cancer therapy, where improved biocompatibility is necessary. The study of the effects on macrophages (i.e. cells of the immune system capable to phagocytose, i.e. engulf and digest, cellular debris and pathogens555) was then extended to graphene dispersed in 1% surfactant (Pluronic F108).555 In this experiment, murine macrophages underwent apoptosis through decrease of mitochondrial potential and ROS increase. To identify the mechanisms that trigger macrophage apoptosis is important as it provides information to develop strategies to control cell death induced by graphene. Besides Pluronic, other polymers were shown to improve the biocompatibility of GO and RGO.556

In developing graphene and GO as therapeutic delivery systems of molecules via systemic administration, it is important to evaluate their interaction with blood cells.557,558 When tested, both graphene and GO showed little haemolysis (i.e. rupturing of red blood cells) (max. dose of 75 μg ml−1).557 Haemolytic activity was completely suppressed by coating GO with the polysaccharide chitosan.558 In addition, platelets exhibited normal morphology and were not activated or aggregated. The coagulation pathways were also not influenced, predictive of low risk of thrombosis once intravenously administered. However, two additional studies reported opposite results and evidenced potential thrombotoxicity risks of GO.559,560 These cytotoxic effects were modulated by reducing the carboxylic functions of GO with thermal annealing,559 or by transforming them into ammonium groups.560 This second transformation led to a new modified GO with no effect on platelet stimulation, nor lysis of erythrocytes (i.e. red blood cells) and absence of thromboembolysis (i.e. embolisms due to blood coagulation). What is unclear is the efficiency of the transformation of the carboxylic groups into amines as GO contains also other types of oxygenated functions that should not undergo the chemical transformation following the Curtius rearrangement (i.e. transformation of the carboxylic groups into amines),561,562 applied to GO. When GO was treated with hydrazine to generate RGO, it was highly cytotoxic, lowering cell viability.549

Alternatively to the reduction with hydrazine, other chemical strategies were used to modify GO. Covalent modification with polyethylene glycol (PEG) chains (PEGylation) had the effect to modulate the cytotoxic effect on a series of cell lines, without affecting cell viability up to 100 μg ml−1.563,564 Cytotoxicity of GO was also reduced by functionalization with dextran, a biocompatible polymer widely used for surface coating of biomaterials.565

Ref. 566 carried out a biocompatibility assessment of both CVD and LPE graphene using cultured human lung epithelial cell line A549. No detectable changes in the cellular morphology of A549 cells growing on graphene thin films or cells exposed to graphene flakes (0.1 to 5 μg mL−1) for 4 to 72 h was observed.566 In contrast, the same work outlined how carbon black is significantly more toxic than graphene.566

This set of results represents a significant first step towards the experimental assessment of the in vitro toxicity of graphene and GO. However, to avoid the generalization often found for other carbon-based nanomaterials, such as CNTs,539 it is necessary to take into consideration the great variability of the materials tested. It is essential to compare graphene and GO as well as other GRMs and correlate their impact on cells to their physico-chemical characteristics and, in case, to the chemical changes introduced.

3.2. Cytotoxicity effects on graphene-coated surfaces

The evaluation of the cytotoxic effects of graphene and GO has also important implications on their use as support for tissue regeneration, cell growth, cell differentiation, and for the development of implantable devices (Fig. 35).567,568
image file: c4nr01600a-f35.tif
Fig. 35 Schematic representation of cells (coloured in purple and with their nucleus in blue) in contact with a surface of graphene (represented as black balls and sticks).

Preliminary studies already explored graphene ability to interface neurons.569 Mouse hippocampal cells were cultured on a graphene substrate to test if the neural functions were affected during their development and maturation in vitro.569 The results revealed high biocompatibility of such growth surface, with low toxicity in the absence of morphological changes.569 The potential use of graphene in neuronal interfacing thus represents a promising approach in future developments of neural prosthesis. Similarly, different surfaces covered by graphene were also able to induce progenitor cell differentiation.570 In particular, graphene allowed to control and to accelerate the proliferation of human mesenchymal stem cells (i.e. multipotent connective tissue cells that can differentiate into a variety of cell types). Cell viability was maintained and no change in cell morphology was observed during the differentiation process, which was enhanced in the case of the cells in direct contact with graphene in comparison to other types of surfaces, like glass or Si wafers.570 This study represents an interesting example where bone regeneration can be achieved in the absence of growth factors, usually required to boost this process.570 It also hints at the safety of graphene when it remains intact after cell growth. This opens the possibility of actively integrating it into devices design to favour tissue regeneration or repair.

3.3. In vivo impact, biodistribution and pharmacokinetics

Additional studies investigating toxicity in vivo,559,560 complemented those in vitro,559,560 directed to assess the impact of GO at the cellular level. Intravenous administration of GO in mice elicited blood platelets aggregation and extensive pulmonary thromboembolism.559,560 This behaviour was associated to the presence of charged functional groups on GO, as the aggregation properties were significantly lowered, when it was chemically transformed into RGO.560In vivo toxicity of GO was evaluated in mice and rats following intravenous administration after labelling with a radiotracer.571 Significant pathological changes, including inflammatory cell infiltration, pulmonary aedema and formation of granulomas were found at a dose of 10 mg kg−1.571 These results were confirmed by another study in which GO, injected in mice, induced granulomas in lungs, spleen and liver.553

Pulmonary toxicity is a major concern in the industrial production of nanomaterials, as their volatile nature favours inhalation and might directly cause acute damages and long term diseases. It was reported572 that GO provokes severe and persistent injury in mice lungs following direct injection into the organs.572 However, this form of toxicity was reduced when graphene was dispersed in a surfactant (i.e. Pluronic).572 In another study,543 the exposure risks of the respiratory system were also suggested for graphene nanoplatelets (25 μm lateral size) dispersed in bovine serum albumin.543 Following the inhalation, the nanoplatelets were found to deposit beyond the respiratory tract. Acute inflammatory responses in mice, inflammation and frustrated macrophage phagocytosis (occurring when a phagocyte is unable to engulf its target because it is physically too large) were reported.543 The inflammogenicity (i.e. the ability to sustain long-term inflammatory activity) was attributed to the inhalable aerodynamic diameter,543 an index that determines the respirability of a particle and the site of deposition. The differences in these two studies, both testing graphene dispersed in surfactants, are likely due to the different morphology of the materials used, in terms of N, surface area, dispersion procedures, and/or the presence of contaminants. These parameters can favour variable toxicity profiles. The ability to manipulate graphene's toxicity profile is supported by the evidence that extensively purified GO does not affect polymorphonuclear leukocytes (i.e. white blood cells containing segmented lobular nucleus) even after 7 days of peritoneal administration.573 No giant cells, inflammatory response and granuloma were observed in comparison to pristine multi wall nanotubes (MWNTs).574 The modulation of toxicity in vivo can also be achieved through functionalization. It was reported that GO PEGylation can reduce toxic effects in mice.564

The study of the biodistribution, accumulation and elimination of graphene and GO in vivo is another key step to predict and avoid any user's risk.574,575 Biodistribution studies have shown that GO accumulates predominantly in the lungs,571,574 while low uptake was observed in the reticuloendothelial system – RES (i.e. the part of the immune system that consists of the phagocytic cells located in reticular connective tissue. The cells are primarily monocytes and macrophages, and they accumulate in lymph nodes and the spleen. The Kupffer cells of the liver are also part of the RES).571,574 GO exhibited long circulation time (several hours) in comparison to other carbon forms like CNTs.571,574 While no pathological modifications in the organs were reported after 14 days following the injection of 1 mg per kg of body weight,571,574 significant changes were observed in the lungs at a dose of 10 mg kg−1. PEGylated GO instead accumulated mainly in the tumour of tumour-bearing animals, with lower uptake by RES, without significant toxic effects.574 The same type of material was intravenously injected, after being radiolabelled, to better elucidate the organ biodistribution and the excretion routes.575 Following an initial accumulation in the RES, a gradual elimination was observed between 3 and 15 days.575 After three months, the GO sheets were completely eliminated without toxic effects.575 No signs of abnormality were observed in kidney, liver, spleen, heart and lung.575 This GO was further studied to evaluate more thoroughly the pharmacokinetics and the long-term biodistribution. These parameters were analyzed by radiolabelling the GO with 125I.575 After intravenous administration, the GO sheets were found to accumulate into the RES including spleen and liver.575 They were gradually cleared by both renal and faecal elimination.575 A dose of 20 mg kg−1 did not provoke toxicity in a period of 3 months, as proved by measuring the biochemical parameters in blood, and the haematological markers.575 Histological examination of the different organs did not show damage or lesions, with the exception of an increase in spleen and liver colour intensity due to the accumulation of brown material.575

Similarly to the in vitro tests, many parameters may affect graphene and GO toxicity measures in vivo. The variability of the samples thus far studied is high. Thus, good practice requires to report the morphological and physico-chemical characteristics of each type of tested samples, avoiding generalization, which may introduce strong biases in the final statements on “toxicity” or “non-toxicity”. In essence, toxicity of graphene seems to be associated to its surface functionalization. However, it is necessary to use animal models, and to explore higher doses of material and other routes of administration. These studies will allow understanding the effects of graphene and GO when deliberately administered for therapeutic strategies, or when accidental exposure occurs.

3.4. Bacterial toxicity

Another issue is the toxicity of graphene, GO and RGO towards microorganisms. In ref. 541, 576, 577 different types of bacterial and fungal strains were tested, since any antibacterial activity may be translated into interesting applications in the development of antimicrobial products (see Fig. 36).
image file: c4nr01600a-f36.tif
Fig. 36 Bacteria (light blue) when in contact with a graphene surface (represented as black balls and sticks) are inactivated (i.e. they lose disease-producing capacity),576 leading to their detachment from the surface, as indicated by the arrow pointing towards the dead bacteria (represented in dark blue).576

GO and RGO showed antimicrobial activity when tested against Gram-negative E. coli576 (i.e. bacteria that do not retain crystal violet dye in the Gram staining protocol,578 (a procedure named after H. G. Gram in which organisms -bacteria- are differentiated according to cell wall composition, exploiting staining, a technique used to enhance contrast in microscopic images), due to structural properties of their cell walls) and Gram-positive S. aureus577 (i.e. bacteria stained dark blue or violet by Gram staining),576 although RGO was more efficient in inactivating both types of pathogens.577 However, Gram-negative bacteria were more resistant.577 Similar behaviours were observed on E. coli, with GO showing more bactericidal activity.541 Effects on the metabolic activity at different concentrations were complemented by the damage of the microorganism cell membrane, as assessed by TEM.577

To expand further these studies, the antimicrobial mechanism was analyzed using different materials, such as graphite, graphite oxide, GO and RGO.579,580 The higher antibacterial activity was measured for GO,579,580 which had the smallest average lateral size amongst the different types of graphenes.579,580 Direct contact of bacteria with graphene sheets induced a loss of membrane integrity. No ROS generated by superoxide anions (O2) were detected.579,580 But oxidation of glutathione, a redox state mediator in bacteria, was observed.579,580 These results suggest that the GO antimicrobial action contributes to both membrane disruption and oxidative stress. The physico-chemical characteristics of GRMs seem to play an important role in the efficiency of bacterial killing. These results were questioned by ref. 581 where GO was added to E. coli: bacteria grew faster by forming dense biofilms around GO. Only the combination with Ag nanoparticles induced cell death.581 In view of these results, the bacteriostatic properties (i.e. the ability to stop bacteria from reproducing) of graphene, GO and RGO certainly need further studies. Comparison between the available data is difficult, as the conditions of cell cultures and type of starting materials differ in the various experiments. New studies are thus needed to assess the antimicrobial role as a function of physico-chemical characteristics.

3.5. Biodegradation

The assessment of the bioaccumulation and biodegradation of novel nanomaterials is another fundamental aspect for their safe implementation. The study of the biodegradability of carbon-based materials is of paramount importance to translate their use into new biomedical devices or therapeutic tools, and, in parallel, to anticipate the possible risks when integrated into novel devices. A careful characterization will allow elucidating and clarifying the role of surface modification and its biological impact, eventually proposing solutions for the treatment and use of graphene, improving the knowledge on toxicology, and prospecting it into clinical applications, including therapy, diagnostics and imaging. As recently demonstrated for CNTs,582 GO can undergo degradation by the enzyme horseradish peroxidase in the presence of hydrogen peroxide.583 RGO seems instead more resistant to this enzyme. This has important implications on the design of safer graphene derivatives to minimize the risks for human health and environment.

Some microorganisms are also able to modify GO.584 The family of bacteria called Shewanella, which consists of metal-reducing bacteria, was reported to affect GO.584E. coli585 and S. aureus were not inhibited,585 but they were able to reduce GO585 by microbial respiration, providing a unique nontoxic approach to the synthesis and modification of graphene flakes. The mechanism of reduction involved both direct extracellular electron transfer and electron mediators at the interface between cells and flakes. This work should be extended to other microorganisms, not only to modify, but eventually to degrade graphene.

3.6. Environmental impact

To our knowledge, to date there are almost no studies that focus on the ecotoxicity and environmental impact of graphene, GO and RGO.541 The majority of data in literature concerns fullerenes.586 These studies highlight the effect of fullerene ingestion and its associated toxicity in several model organisms: fresh water crustaceans,586 marine copepods586 and fish.586 Several data are also currently available on the ecotoxicological effects of CNTs in various organisms.587 Most studies on CNT ecotoxicity were conducted on representative species of the aquatic environment,588–590 and, to a lesser extent, on terrestrial organisms.591,592 Based on the experience acquired with these carbons, it is mandatory to explore the impact of GRMs on the environment. It is necessary to develop and broaden our knowledge in this area in order to better understand the potential effects on the environment in the short or long term, and to adopt adequate disposal procedures. With the development of graphene technology, it is essential to assess exposure in real-life conditions and to fully understand the graphene life-cycle, otherwise any assessment of occupational or environmental effects will remain uncertain.

Thus far, only few studies addressed the influence of graphene on terrestrial plants.593 Phytotoxicity (i.e. toxic effects on plant growth) at concentrations of GO between 500 and 2000 μg ml−1 was studied in cabbage, tomato, red spinach and lettuce.593 Root and shoot growth, biomass, shape, cell death and generation of ROS were evaluated.593 After 20 days of exposure, the plants treated with GO displayed a reduced number of leaves.593 The growth and biomass were inhibited with visible symptoms of necrosis and increased ROS.593 The only exception was lettuce seedlings.593 These effects were dose and time dependent.593 Particular attention needs to be devoted to possible adverse effects of graphene on plant cells. Similarly, GO sheets were tested on the germination of rice seeds in comparison to other carbon nanomaterials such as CNTs and fullerenes.594 While, e.g., CNTs were found to favour an increase in water content during germination,594 GO retarded the development and reduced the water content.594 A possible blockage of the pore seeds by micro-sized GO layers likely caused less water uptake and inhibited germination. This was reflected by a reduced growth of rice seedlings, shorter roots and shoots. The dimensions and form of carbon nanomaterials influence the development of the plants, thus can be of concern during the processes of handling and disposal. In vivo cytotoxicity studies of functionalized GO were performed on zebrafish.594 Zebrafish is one of the model animals for screening the impact of nanomaterials in the environment (since it is a vertebrate, it has a high genetic homology with humans, and the assays can be automated to provide high-throughput along with high-content analysis). Besides analyzing the GO biodistribution and demonstrating the possibility of its use for whole-body imaging, biocompatibility was also assessed.594 Functionalized GO did not induce any significant abnormality on embryonic development,594 nor affected the survival rate, neither provoked malformations.594 To our knowledge, ref. 594 is the first example of evaluation of GO using a relevant model in an environmental context. Although a generalization is impossible at this stage, this can be considered the starting point to intensify the studies in this direction.

3.7. 2d crystals and hybrids

To our knowledge, there are no studies on health and environment impact of other 2d crystals and their hybrids with graphene or other materials. Some studies are available on BN nanotubes595 or cubic-BN.596 BN nanotubes coated with polylysine were tested on a model of muscle cells595 (C2C12 mouse myoblast – an embryonic cell which becomes a muscle cell or fiber-cell line). The tubes are able to penetrate these cells without affecting their viability,595 nor interfering with the formation of myotubes (i.e. multinucleated fibres that form muscle fibres), during cell differentiation.595,596 Alternatively, BN nanotubes were functionalized with dendritic structures bearing carbohydrate ligands to their periphery by non-covalent adsorption.597 The generated complexes interacted with proteins and cells without modifying cell proliferation and viability.597 Morphological alterations in different cell populations exposed to BN nanotubes are cell-type dependent.598

3.8. Perspective

Research is necessary to explore the biological responses and the potential toxicity of GRMs looking at their interactions with living systems. This will allow better understanding of the differences between GRMs, eventually establishing a correlation of their impact on health and environment with their physico-chemical characteristics. These studies could guide the safe design, production and manufacturing of GRMs, minimizing health and environment risks. To date, only a few studies focus on the ecotoxicology and environmental impact of GRMs. It is imperative to broaden our knowledge in this area in order to better identify the potential effects in the short and long term. With the development of large-scale production and the arrival on the marketplace, it is essential to assess exposure in real conditions and fully understand the GRM life-cycle. Without this, any assessment of health risks due to occupational or environmental exposure will remain uncertain.

The activities on health and environment need to focus on a thorough exploration of the biological responses and the toxicity effects of GRMs by taking into consideration their physico-chemical properties and bio-nano-interactions. The variability between the samples is currently high. The toxicity of GRMs seems to be closely associated to their surface functionalization. In relation to surface characteristics, size is a second important parameter. Other key factors can be also associated to the toxicity of new nanomaterials, such as generation of reactive oxygen species, indirect toxicity because of adsorption of biomolecules or signalling from cells that accumulate nanoparticles to adjacent non-exposed cells (e.g. paracrine signalling599 – a form of cell–cell communication in which a cell produces a signal to induce changes in nearby cells, altering the behaviour or differentiation of those cells), and physical toxicity associated with GRMs interaction with lipids (and other molecules) constituting cell membranes, tissues and organs. In addition, the study of cellular and tissue uptake as a function of GRM size and chemical functionalization is very important. Lateral dimensions of GRMs might affect the population of the receptors involved in the mechanisms of penetration dependent on energy (e.g. endocytosis, an energy dependent process by which cells absorb molecules or nanoparticles by engulfing them, or phagocytosis). If passive processes, as in CNTs,600 happen, it is necessary to understand how the 2d form affects membrane organization (e.g. membrane disruption or sliding between lipid bilayers).601

At the environmental level, the impact on organisms widely used as natural biosensors, such as lichens, and on a certain number of aquatic species, or microorganisms, specifically sentinel species that can be affected once 2d crystals are released, need to be assessed. New living organism models need to be applied to investigate the water/sediment interface and terrestrial behaviour of 2d crystals. All these studies will offer a safer design, production and manufacturing of 2d crystals in order to minimize health and environment risks, Fig. 37.

image file: c4nr01600a-f37.tif
Fig. 37 Possible interactions of GRMs with cells, organs (i.e. lung) and living organisms (i.e. amphibian). GRMs can have an environmental impact on living organisms like amphibians as indicated by the left arrow pointing to the frog on the bottom left panel. GRMs can have an impact on human health leading to disease related to respiration, as illustrated by the arrow pointing to the lung (bottom right panel). GRMs can interact at the cellular level as indicated by the top arrow pointing to a cell represented with its subcellular components. Possible interactions of GRMs at the cellular level can occur, following a clockwise direction from the top left panel, at the plasma membrane level (membrane contact), during cell uptake at the cellular membrane (cell uptake), in the cytoplasm where degradation may occur (biodegradation), and at the nuclear membrane (nucleus entry). Adapted from ref. 542.

In this context, several interconnected targets should be pursued: (i) elucidation of how GRMs interact with cells at the cellular and molecular level, with the assessment of the role of the bio-corona (i.e. proteins surrounding a nanomaterial when in contact with a biological fluid); (ii) addressing GRM effects on specific tissues, such as the immune system, nervous system or placenta, and determination of biomarkers for possible pathogenic risks; (iii) identification of any possible GRM hazard in relation to their physico-chemical properties, with a particular focus on the most important exposure routes (i.e. lung, skin); (iv) understanding the processes that control GRM biostability and biodegradation, key to pave the way to nano-interfacing devices; (v) investigation of GRM's potential impact on aquatic species (i.e. amphibians), terrestrial organisms and microorganisms; (vi) development of a standardized and validated GRM testing strategy, to enable regulation.

In the long term, knowledge of GRMs’ biotoxicity and environmental risk will favour: (i) manufacturing, development and applications, (ii) assessment of technological risk versus risk perception to harmonize technical/social visions; (iii) development of tools for the governance of key changes in social landscape for new technology; (iv) converging technologies, i.e. integrative process of orienting S&T capabilities towards societal needs.

The timeline is shown in Fig. 38. Timescales: 23 years: Investigation and understanding the effect of GRMs on different cells and identifying possible hazard. 47 years: Development of GRMs regulation. 710 years: Assessing the impact of GRMs and validation of safety issues for technological development.

image file: c4nr01600a-f38.tif
Fig. 38 Timeline for the study of impact on health and environment of GRMs.

4. Production

The industrial use of GRMs will require large scale and cost-effective production methods, while providing a balance between ease of fabrication and final material quality, defined as on-demand tailored properties according to the final use. One advantage of graphene is that, unlike other nano-materials, it can be made on large and cost-effective scale by bottom up (atom by atom growth) or top-down (exfoliation from bulk) techniques,602 see Fig. 39. Prior to large-scale production, developing laboratory scale protocols for synthesis with targeted properties is essential.
image file: c4nr01600a-f39.tif
Fig. 39 Schematic illustration of the main experimental setups for graphene production. (a) Micromechanical cleavage (b) Anodic bonding (c) Photoexfoliation. (d) Liquid phase exfoliation. (e) Growth from SiC. Schematic structure of 4H-SiC and the growth of graphene on SiC substrate. Gold and grey spheres represent Si and C atoms, respectively. At elevated temperatures, Si atoms evaporate (arrows), leaving a C-rich surface that forms graphene. (f) Precipitation from carbon containing metal substrate. (g) CVD process. (h) Molecular beam epitaxy. Different carbon sources and substrates (i.e. SiC, Si, etc.) can be exploited. (i) Chemical synthesis using benzene as building blocks. Adapted from ref. 602.

The combination of graphene with other materials in heterostructures, as well as functionalization, will play a key role in validating GRMs as a disruptive nanotechnology platform for real-world devices. Table 2 summarize the state of the art in term of size and μ for some of the production strategies presented in Fig. 39, together with targets applications.

Table 2 State of the art (as of August 2014) of the main production approaches and foreseen applications
Method Crystallites Size, μm Sample Size, mm μ Applications
Micromechanical cleavage 1000 1 2 × 105 cm2 V−1 s−1 Fundamental research and proof of principle devices
106 cm2 V−1 s−1
(@T = 4 K)
2 × 104 cm2 V−1 s−1
LPE of graphite 0.01–1 0.1–1 100 cm2 V−1 s−1 Inks, coatings, paints, batteries, supercaps, solar cells, fuel cells, composites, sensors, TCs, photonics, flexible electronics and optoelectronics, bio-applications
(∞ as overlapping flakes) (for a layer of overlapping flakes)
LPE of GO >1 >1 1 cm2 V−1 s−1 Inks, coatings, paints, batteries, supercap, solar cells, fuel cells, composites sensors, TCs, photonics, flexible electronics and optoelectronics, bio-applications
(∞ as overlapping flakes) (for a layer of overlapping flakes)
Growth on SiC 100 100 6 × 106 cm2 V−1 s−1 RF transistors
(6′′) @T = 4 K other electronic devices
CVD 50[thin space (1/6-em)]000 1000 6.5 × 104 cm2 V−1 s−1 Photonics, nanoelectronics, TCs, sensors, bio-applications, flexible electronics
@T = 1.7 K
3 × 104 cm2 V−1 s−1

The main targets are:

• Development of scalable synthesis protocols that enable tuning of electronic, structural and optical properties for different applications.

• Enlargement of scope of GRMs applications by adding new functionalities. This implies developing a strategy for covalent and non-covalent functionalization with molecules, clusters and nanocrystals.

• Systematic exploration of 2d crystals. Hundreds of LMs have not yet been exfoliated and could exhibit interesting properties that would be useful in a range of applications.

• Development of hybrid structures combining graphene and other 2d crystals.

In the short term, it is imperative to meet graphene specifications required by the targeted applications and develop scalable strategies to build graphene heterostructures with improved functionalities. The main long-term goal is to achieve on-demand large-scale heterostructures of 2d hybrid systems, with control of growth, pattern formation, functionalisation and self-assembly, without compromising the quality of the layers.

The production effort should not be limited to lab-scale synthesis, but should also aim at recognizing and understanding the sources of differences in materials properties compared with mechanically exfoliated flakes, as well as improving properties, such as electrical and thermal conductivity, μ, carriers concentration, functionality and homogeneity.

In the short term, these objectives should be targeted:

• Development of scalable synthesis protocols for GRMs. This includes the development of functionalization processes that enable tuning of electronic, structural and optical properties for different applications. This will constitute a breakthrough in GRMs growth technology, paving the way for the production of 2d-based devices on an industrial scale.

• Development of scalable liquid exfoliation synthesis and functionalization of GRMs, optimizing their properties for different applications (energy, composites, electronics, optoelectronics, and bio-medicine). Processing of liquid exfoliated GRMs in ultrathin few layer films. Formulation and characterization of inks suitable for printings and coatings.

• Developing bottom up approaches to grow GRMs from molecular precursors on metals (CVD), (molecular beam epitaxy, MBE), etc., for energy, electronics, optoelectronics and spintronics. Cluster and nanoparticle decoration of graphene films. Optimization of reliable transfer techniques on different target substrates (flexible and non-flexible).

• Development of growth of graphene on commercial SiC wafers, both polarities and politypes, assessing the homogeneity of thickness, maximum carrier μ, concentration and functionalizations. Implementing surface engineering intercalation protocols leading to high μ through intercalation by hydrogen and/or other elements.

• Development of methods to build multistacked heterostructures formed by a few layers that can be the bases of new devices. Hybrids of BN/graphene and graphene/other 2d crystals.

The objectives to be pursued in longer time frames (∼10 years) are:

• Further optimization of specific graphene properties enabling additional applications, among them tailoring GNR bandgaps, increasing μ of graphene grown on metals and SiC, improving the synthetic processes to achieve larger areas. Also, as the synthetic methods progress and reach maturity, implementation of technological processes leading to in situ/ex-situ patterning, continuous growth/synthesis, on-line characterization and continuous, deterministic transfer.

• Production of other 2d crystals whose properties and applications are unknown today. This will determine many activities in the long term, not limited to graphene but enabling a complete exploration of the Flatland promise.

In the following sections we overview the production strategies and processing and placement methods of GRMs, starting with the historical development602 going through the analysis of the state of the art,602 and future goals.

4.1. Graphene production

4.1.1. Dry exfoliation. Dry exfoliation is the splitting of LM into atomically thin sheets via mechanical, electrostatic, or electromagnetic forces in air, vacuum or inert environments. Mechanical exfoliation for research purposes and new concept devices. Micromechanical cleavage (MC), also known as micromechanical exfoliation, has been used for decades by crystal growers and crystallographers.603,604 In 1999 ref. 605 reported a method of cleaving graphite, yielding films of several layers of graphene. Ref. 605 also suggested that “more extensive rubbing of the graphite surface against other flat surfaces might be a way to get multiple or even single atomic layers of graphite plates”. This was then firstly demonstrated, achieving SLG using an adhesive tape, by ref. 5, as for Fig. 39a).

MC is now optimized to yield high quality layers, with size limited by the single crystal grains in the starting graphite, of the order of millimeters.125 The number of layers can be readily identified by elastic (see Fig. 40a)171 and inelastic134 light scattering (Raman spectroscopy, see Fig. 40b). μ up to 107 cm2 V−1 s−1 at 25 K were reported for a decoupled SLG on the surface of bulk graphite,606 and up to 106 cm2 V−1 s−1 on current-annealed suspended SLGs,607 while RT μ up to ∼20[thin space (1/6-em)]000 cm2 V−1 s−1 was measured in as-prepared SLGs.608 Suspended SLGs, cleaned by current annealing, can reach μ of several 106 cm2 V−1 s−1 (ref. 609). μ > 105 cm2 V−1 s−1, with ballistic transport at the micron level, was reported for SLG encapsulated between exfoliated h-BN layers.99

image file: c4nr01600a-f40.tif
Fig. 40 (a) Optical micrograph of MC flake, with regions of different thickness. (b) Evolution of Raman spectra with N.134 The spectra are normalized to have the same G peak intensity.602

Although MC is impractical for large-scale applications, it is still the method of choice for fundamental studies. Indeed, the vast majority of basic results and prototype devices were obtained using MC flakes. Thus, MC remains ideal to investigate both new physics and new device concepts. Anodic bonding. Anodic bonding is widely used in the microelectronics industry to bond Si wafers to glass,610 to protect them from humidity or contaminations.611 When employing this technique to produce SLGs,612,613 graphite is first pressed onto a glass substrate, and a high voltage of few KVs (0.5–2 kV) is applied between it and a metal back contact (see Fig. 39b), and the glass substrate is then heated (∼200 °C for ∼10–20 min).612,613 If a positive voltage is applied to the top contact, a negative charge accumulates in the glass side facing the positive electrode, causing the decomposition of Na2O impurities in the glass into Na+ and O2 ions.612,613 Na+ moves towards the back contact, while O2 remains at the graphite–glass interface, establishing a high electric field at the interface. A few layers of graphite, including SLGs, stick to the glass by electrostatic interaction and can then be cleaved off;612,613T and applied voltage can be used to control N and their size.612,613 Anodic bonding was reported to produce flakes up to ∼1 mm width.613 This method may also be used for other LMs. Laser ablation and photoexfoliation. Laser ablation is the use of a laser beam to removal of material, via evaporation and/or sublimation, from a solid surface614, Fig. 39c). In the case of LMs, such as graphite, if the laser beam irradiation does not induce evaporation and/or sublimation of the carbon atoms, but the detachment of an entire or a part of a layer, the process is called photoexfoliation.615

Laser pulses can in principle be used to ablate/exfoliate graphite flakes, Fig. 39c. Tuning the energy density permits the accurate patterning of graphene.616 The ablation of a defined N can be obtained exploiting the energy density windows required for ablating a SLG616 and N-layer graphene (NLGs).616Ref. 616 reported that energy density increases for decreasing N up to 7LG. Ref. 616 argued that the N dependence of the energy density is related to the coupling of heat with NLGs via phonons, with the specific heat scaling as 1/N. For N > 7 the ablation threshold saturates.616 Laser ablation is still in its infancy616,617 and needs further development. The process is best implemented in inert or vacuum conditions,618,619 since ablation in air tends to oxidize the graphene layers.616 Promising results were recently demonstrated also in liquids.620 Thus, photoexfoliation could be alternative and complementary technique to LPE.

Laser irradiation has room for further optimization. This technique was tested to produce flakes from direct laser irradiation of GO.621 New protocols are needed to prepare graphene flakes in liquid, overcoming the limitations of LPE, exploiting high boiling point solvents and surfactants. The laser irradiation approach is of general validity. It can be extended to other LMs with weak interlayer coupling, see section 4.12.2.

4.1.2. Liquid phase exfoliation. Graphite can also be exfoliated in liquid environments exploiting ultrasounds to extract individual layers, Fig. 39d. The LPE process generally involves three steps: (1) dispersion in a solvent; (2) exfoliation; (3) “purification”. The third step is necessary to separate exfoliated from un-exfoliated flakes, and usually requires ultracentrifugation. Exfoliation can be done via chemical wet dispersion followed by ultrasonication (Fig. 39d), both in aqueous250,624,625,627 and non-aqueous solvents.35,622,623,625,626,629

The LPE yield can be defined in different ways. The yield by weight, YW [%], is the ratio between the weight of dispersed graphitic material and that of the starting graphite flakes.35 The yield by SLG percentage, YM [%], is the ratio between the number of SLG and the total number of graphitic flakes in the dispersion.35 The Yield by SLG weight, YWM [%], is the ratio between the total mass of dispersed SLG and the total mass of all dispersed flakes. YW does not give information on the “quality” (i.e. the composition of the dispersion, e.g. the presence of SLG, BLG, etc.) of the dispersion, since it considers all the graphitic material (SLG, FLG and thicker flakes), thus it does not quantify the amount of SLG, but only the total amount of graphitic material in dispersion. YM [%], and YWM [%] are more suitable to quantify the amount of dispersed SLGs.

In order to determine YW it is necessary to calculate the concentration c [g L−1] of dispersed graphitic material. c is usually determined via optical absorption spectroscopy (OAS),35,58,624–627 exploiting the Beer–Lambert Law: A = αcl, where l [m] is the length of the optical path and α [L g−1 m−1] is the absorption coefficient. α can be experimentally determined by filtering a known volume of dispersion, e.g. via vacuum filtration, onto a filter of known mass,35,624–627 and measuring the resulting mass using a microbalance. The filtered material is made up of graphitic flakes, surfactants or solvents and residuals from the filter.35,624 Thermogravimetric analysis (TGA) is used to determine the weight percentage of graphitic material, thus enabling the measurement of c.58–61 However, different values of α have been estimated both for aqueous59,60 and non-aqueous-based dispersions.35,626Ref. 35 derived α ∼ 2460 m L mg−1 m−1 for a variety of solvents, i.e. N-methylpyrrolidone, NMP, dimethylformamide, DMF, benzyl benzoate, γ-butyrolactone, GBL, etc., while later ref. 626 reported α ∼ 3620 mL mg−1 m−1 for NMP. Ref. 624 gave α ∼ 1390 mL mg−1 m−1 for aqueous dispersions with sodium dodecylbenzene sulfonate (SDBS), while ref. 627 reported a higher value ∼6600 mL mg−1 m−1, still for aqueous dispersions but with sodium cholate (SC). Ref. 627 assigned this discrepancy to the difference in concentration of the two dispersions. However, α cannot be dependent on the concentration (indeed it is used for its determination), thus more work is needed to determine its exact value.602

Y M is usually determined via TEM and AFM. In TEM, N can be counted both analyzing the edges134 of the flakes and by using electron diffraction patterns.134 AFM enables the estimation of N by measuring the height of the deposited flakes and dividing by the graphite interlayer distance, 0.34 nm.602 However, the estimation for the height of SLG via AFM is dependent on the substrate.602 Indeed, on SiO2, a SLG has a height of ∼1 nm,5 while on mica is ∼0.4 nm.628 Raman spectroscopy can be used for the determination of YM[thin space (1/6-em)]35,625,629 and to confirm the results obtained with TEM and/or AFM. YWM [%] requires the estimation of the SLGs area other than N.35 Although this is more accurate (giving quantitative and qualitative information on SLGs), with respect YW and YM, its determination is time consuming. However, for a semi-quantitative evaluation, YM and YW must be reported if YWM is not. LPE of graphite. Ultrasound-assisted exfoliation is controlled by hydrodynamic shear-forces, associated with cavitation,630i.e. the formation, growth, and collapse of bubbles or voids in liquids due to pressure fluctuations.630 After exfoliation, the solvent–graphene interaction needs to balance the inter-sheet attractive forces.

Solvents ideal to disperse graphene are those that minimize the interfacial tension [N m−1]631 between the liquid and graphene flakes. In general, interfacial tension plays a key role when a solid surface is immersed in a liquid medium.631–633 If the interfacial tension between solid and liquid is high, there is poor dispersibility of the solid in the liquid.631 In the case of graphitic flakes in solution, if the interfacial tension is high, the flakes tend to adhere to each other and the cohesion (i.e. the energy per unit area required to separate two flat surfaces from contact631) between them is high, hindering their dispersion in liquid.602 Liquids with surface tension (i.e. the property of the surface of a liquid that allows it to resist an external force, due to the cohesive nature of its molecules631) γ ∼ 40 mN m−1 (ref. 35), are the “best” solvents for the dispersion of graphene and graphitic flakes, since they minimize the interfacial tension between solvent and graphene.602 The same solvents have been investigated for the debundling of CNTs.634–636

Ref. 637 determined via wettability and contact angle the surface energy, ϱ [mJ m−2], of different graphitic materials, finding ϱ ∼ 46 mJ m−2, ∼55 mJ m−2, ∼62 mJ m−2 for RGO, graphite and GO. The slight difference being due to the different surface structure of GO, RGO and graphite. Ref. 638 reported that the contact angle measurements are not affected by N.

The majority of solvents with γ ∼ 40 mN m−1 (i.e. NMP, DMF, Benzyl benzoate, GBL, etc.) [see ref. 35 for a more complete list] have some disadvantages. E.g., NMP may be toxic for reproductive organs,639 while DMF may have toxic effects on multiple organs.640 Moreover, all have high (>450 K) boiling points, making it difficult to remove the solvent after exfoliation. As an alternative, low boiling point solvents,641 such as acetone, chloroform, isopropanol, etc. can be used. Water, the “natural” solvent”, has γ ∼ 72 mN m−1,631 too high (30 mN m−1 higher than NMP) for the dispersion of graphene637 and graphite.637 In this case, the exfoliated flakes can be stabilized against re-aggregation by Coulomb repulsion using linear chain surfactants, e.g. SDBS,624 or bile salts, e.g. SC642 and sodium deoxycholate (SDC),250,625 or polymers, e.g. pluronic,643etc. However, depending on application, surfactants/polymers may be an issue, e.g. compromising, decreasing, inter-flake conductivity.644

Thick flakes can be removed following different strategies based on ultracentrifugation in a uniform645 or density gradient medium (DGM).646 The first is called differential ultracentrifugation (sedimentation based-separation, SBS),645 while the second density gradient ultracentrifugation (DGU).646 SBS separates various particles on the basis of their sedimentation rate645 in response to a centrifugal force acting on them. SBS is the most common separation strategy and, to date, flakes ranging from few nm to a few microns have been produced, with concentrations up to a few mg ml−1.626,647 High concentration is desirable for large-scale production of composites35 and inks.629YM up to ∼70% can be achieved by mild sonication in water with SDC followed by SBS,250 while YM ∼ 33% was reported with NMP.629 This YM difference is related to the difference in flake lateral size. In water-surfactant dispersions flakes are on average smaller (∼30 nm250 to ∼200 nm624) than in NMP (∼1 μm35,629), since the viscosity (ν) at RT of NMP (1.7 mPa s648) is higher than water (∼1 mPa s648). Larger flakes in a higher ν media experience a higher frictional force645,646 that reduces their sedimentation coefficient, making it more difficult for them to sediment. This decreases YM in NMP compared to water.

Control on N is achieved via DGU: graphitic flakes are ultracentrifuged in a preformed DGM646,649 see Fig. 41a,b. During this process, they move along the cuvette, dragged by the centrifugal force, until they reach the corresponding isopycnic point, i.e., the point where their buoyant density equals that of the surrounding DGM.646 The buoyant density is defined as the density (ρ) of the medium at the corresponding isopycnic point.649 Isopycnic separation has been used to sort CNTs by diameter,650 metallic vs. semiconducting nature651, and chirality.652 However, unlike CNTs of different diameter, graphitic flakes have the same density irrespective of N, so another approach is needed to induce a density difference: coverage of the flakes with a surfactant results in an increase of buoyant density with N, see Fig. 41c. Fig. 41d is a photograph of the cuvette after the isopycnic separation with Sodium Deoxycholate (SDC). To date, YM up to ∼80% was reported by using isopycnic separation.642

image file: c4nr01600a-f41.tif
Fig. 41 Sorting of graphite flakes via isopycnic separation. Formation of (a) step gradient and (b) linear density gradient. (c)The flake-surfactant complexes move along the cuvette, dragged by the centrifugal force, until they reach their corresponding isopycnic points. The buoyant density of the flake-surfactant complexes increases with N. (d) Photograph of a cuvette containing sorted flakes. Adapted from ref. 602.

Another method is the so-called rate zonal separation (RZS).653,654 This exploits the difference in sedimentation rates of nanoparticles with different size654,655 shape656 and mass,654,655 instead of the difference in nanoparticle density, as in the case of isopycnic separation. RZS was used to separate flakes with different size653 (the larger the size, the larger the sedimentation rate).

Other routes based on wet chemical dispersion have been investigated, such as exfoliation in ionic liquids (ILs),657,658 1-hexyl-3-methylimidazolium hexafluorophosphate (HMIH)657 or 1-butyl-3-methylimidazolium bis(trifluoro-methane-sulfonyl)imide ([Bmim]-[Tf2N]).658 These are a class of purely ionic, salt-like materials,659 defined as salts in the liquid state (below 100 °C), largely made of ions.659Ref. 657 reported concentrations exceeding 5 mg mL−1 by grinding graphite in a mortar with ILs, followed by ultrasonication and centrifugation. The flakes had sizes up to ∼3–4 μm, however no YM data was shown. Ref. 657 used a long ultrasonication process (>24 hours), probably because of the IL high ν. In SBS ν plays a fundamental role. Flakes in a higher ν medium have a lower sedimentation coefficient with respect to water. The sedimentation coefficient is commonly measured in Svedberg (S) units (with 1S corresponding to 10−13 s), i.e. the time needed for particles to sediment out of the fluid, under a centrifugal force.79E.g., for a flake dispersed in [Bmim]-[Tf2N] (ρ = 1.43 g cm−3, ν = 32 mPa s), the sedimentation coefficient is ∼55 times smaller than in water.602 There are no reports to date showing that exfoliation via ultrasonication in ILs can have the same YM as in water,250 or organic solvents.629 Moreover, the resultant flakes contain oxygen functional groups,658 probably due strong non-covalent interactions, or covalent functionalization with [Bmim][Tf2N] itself.658 A step forward for the production of flakes without these functional groups was reported in ref. 660 where oxygen-free flakes were made by grinding graphite in 1-butyl-3-methylimidazolium hexafluorophosphate, [BMIMPF6]. ILs were then removed by mixing with Acetone and DMF.658 Controlling grinding time and IL quantity, ref. 658 reported GQDs with size from 9 to 20 nm and thickness between 1 and 5 nm.

Alternative processes rely on exfoliation of fluorinated graphite,661 expanded graphite,662 and non-covalent functionalization of graphite with 1-pyrenecarboxylic acid.663

LPE is cheap and easily scalable, and does not require expensive growth substrates. A range of applications for graphene lie in conducting inks629 (Fig. 42a), thin films35 (Fig. 42b) and composite materials624 (Fig. 42c). For these, graphene is best prepared as flakes, so that the active surface is maximised. The resulting material can be deposited on different substrates (rigid and flexible) following different strategies, such as drop and dip casting (Fig. 42d), rod (Fig. 42e) and spray coating (Fig. 42f), screen and ink-jet629 printing (Fig. 42g), vacuum filtration,35 Langmuir–Blodgett,662 and other techniques.

image file: c4nr01600a-f42.tif
Fig. 42 (a) Graphene ink. (b) Graphene-based TCF and (c) graphene polymer composite. (d) Dip casting, (e) rod coating, (f) spray and (g) inkjet printing of graphene inks.602

High quality graphene inks and ink-jet printed TFTs with μ ∼ 90 cm2 V−1 s−1 have been demonstrated,629 paving the way towards a fully graphene-based printable electronics.629

LPE flakes have limited size due to both the exfoliation procedure, that induces in-plane fracture, and the purification process, which separates large un-exfoliated flakes. To date, LPE-SLGs have area mostly below 1 μm2 (ref. 35, 250, 624–626, 629, 641).

LPE can also be optimized to produce GNRs, with widths <10 nm.303Ref. 303 ultrasonicated expanded graphite664 in a 1,2-dichloroethane solution of poly(m-phenylenevinylene-co-2,5-dioctoxy-p-phenylenevinylene), and ultracentrifuged, resulting in flakes and GNRs of different shapes. However, the GNR production mechanism via LPE is not well understood. Thus, more work is needed to improve it.

The goal is now to further develop LPE to get control on-demand of N, flake thickness and lateral size, as well as rheological (i.e., density, viscosity, and surface tension) properties of the resulting dispersions. A combination of theory and experiments is needed to fully understand the exfoliation process in different solvents, in order to optimise the separation of flakes in centrifugal fields, so to achieve SLG and FLG with well-defined morphological properties at a high rate.

A very desirable step is the development of techniques capable of manipulating individual flakes. Optical tweezers (OT) can trap, manipulate, control and assemble dielectric particles, single atoms, cells and nanostructures.250,665–669 These can be used to trap graphene layers and/or GNRs in liquid environments. The coupling of OT with a Raman spectrometer (Raman Tweezers250), can test solutions composition and sort N in optofluidic channels. The assessment of exfoliation yield is essential to allow further improvements. Detailed structural characterisation of the exfoliated sheets can be done by aberration-corrected HRTEM and STEM, electron energy loss spectroscopy (EELS) and in situ TEM. These can characterise the exfoliated materials down to the atomic level. The effect of structural defects on the electrical properties can also be investigated in situ.

Ref. 670 reported a scalable method, based on shear mixing of graphite, Fig. 43. During rotation, the shear mixer acts as a pump, pulling both liquid and solids into the mixing head where centrifugal forces drive them towards the edge of the rotor/stator (see Fig. 43b).670 This is accompanied by intense (power density of ∼100 W l−1) shear as the materials are driven between the rotor and screen and then out through the perforations in the stator and into the main volume of the liquid. This method can produce ∼1.4 g h−1 of mostly FLG with YW = 3.35% and could be applied to other LMs.670

image file: c4nr01600a-f43.tif
Fig. 43 (a) Silverson model L5M high-shear mixer with mixing head in a 5 liters beaker of dispersion containing graphitic flakes. Close-up view of (b) mixing head (diameter = 32 mm) and (c) mixing head (diameter = 16 mm) with rotor (left) separated from stator. (d), bottles containing the final dispersions. Adapted from ref. 670. LPE of graphite oxide. LPE is a versatile technique and can be exploited not only for the exfoliation of pristine graphite, as reported in section, but also for the exfoliation of graphite oxide and graphite intercalated compounds (GICs), which have different properties with respect to pristine graphite,602 as discussed in this and the next section.

The oxidation of graphite in the presence of potassium chlorate (KClO3) and fuming nitric acid was developed by Brodie in 1859 while investigating the reactivity of graphite flakes.671 This process involved successive oxidative treatments of graphite in different reactors.671 In 1898, Staudenmaier modified Brodie's process by using concentrated sulphuric acid and adding KClO3 in successive steps during the reaction.672 This allowed carrying out the reaction in a single vessel, streamlining the production process.673 However, both methods were time consuming and hazardous, as they also yielded chlorine dioxide (ClO2) gas,104 which can explosively decompose into oxygen and chlorine.674 Graphite oxide (GrO) flakes were already investigated by Kohlschtter and Haenni in 1918,675 and the first TEM images reported in 1948 by Ruess and Vogt676 showed the presence of single sheets of GO.

In 1958, Hummers modified the process using a mixture of sulphuric acid, sodium nitrate and potassium permanganate.677 Avoiding KClO3 made the process safer and quicker with no explosive byproducts.677 These aggressive chemical processes disrupt the sp2-bonded network and introduce hydroxyl or epoxide groups,678–680 in the basal plane, while carbonyl and carboxylic groups, together with lactone, phenol and quinone attach to the edges (see Fig. 44). However, the introduction of these functional groups is essential for the GO production and subsequent liquid dispersion.

image file: c4nr01600a-f44.tif
Fig. 44 GO synthesis and reduction. Graphite can be oxidized with different procedures in the presence of strong acids. The GO flakes have the basal plane functionalized with epoxy and hydroxyl groups, both above and below it, and the edges with a variety of functional groups. This makes GO sheets defective. A partial restoration of the electronic properties is obtainable following different reduction strategies.602

GO flakes can be produced via sonication,303,681 stirring,682 thermal expansion,683etc., of GrO. The aforementioned functional groups make them strongly hydrophilic, allowing their dispersion in pure water,303,681 organic solvents,682,684 aqueous mixtures with methanol, acetone, acetonitrile685 or 1-propanol and ethylene glycol.686 However, although large GO flakes, up to several microns,687 can be produced, they are defective679 and insulating, with Rs ∼ 1012 Ω □−1(Ohm per square), or higher.688

GO is luminescent under continuous wave irradiation.689 Visible excitation gives a broad PL spectrum from visible to near-infrared,564 while blue emission690 is detected upon UV excitation. This makes GO an interesting material for lighting applications (e.g. light emitting devices691) and bio-imaging.564

Several processes have been developed to chemically “reduce” GO flakes, i.e. decrease the oxidation state of the oxygen-containing groups in order to re-establish σ and κ as close as possible to pristine graphene.602 In 1962, the reduction of GrO in alkaline dispersions was proposed for the production of thin (down to single layer) graphite lamellaes.680,692 Other methods involve treatments by hydrazine,303,693 hydrides,686,694p-phynylene,695 hydroquinone694etc., as well as dehydration696 or thermal reduction.679,684,697 UV-assisted photocatalyst reduction of GO was proposed,698 whereby GO reduces as it accepts electrons from UV irradiated TiO2 nanoparticles.698

The charge transport in RGO is believed to take place via variable-range hopping (VRH).690,699 Individual RGO sheets have been prepared with σ ∼ 350 S cm−1,700 while higher values (1314 S cm−1) were achieved in thin films,701 because in the latter RGO flakes are equivalent to resistors in parallel.693 These σ are much bigger than those of organic semiconductors (e.g. poly(β′-dodecyloxy(-α,α′-α′,α′′)terthienyl) (poly(DOT)) ∼ 10−3 S cm−1 for charge carriers, n ∼ 1021).702

It is important to differentiate between dispersion-processed flakes, retaining the graphene electronic properties, such as those reported in ref. 35, 250, 624–627, 629, and GO flakes, such as those in ref. 303, 681–684.

GO and RGO can be deposited on different substrates with the same techniques used for LPE graphene, discussed in section, 11.1. GO and RGO are ideal for composites,703 due the presence of functional groups, which can link polymers.703

Ref. 704 reported RGO sheets with σ ∼ 103 S m−1, high flexibility, and surface areas comparable to SLG, thus interesting for a range of electronic and optoelectronic applications. Thin films of RGO have been tested as FETs,705 transparent conducting films (TCFs),706 electro-active layers,707–709 solar cells,710 ultrafast lasers,711,712etc. Patterning has been used to create conductive RGO-based electrodes.690

Heating-driven reduction has the potential to produce good quality graphene structures. Laser heating in an oxygen-free environment (Ar or N2) can be done with a spatial resolution down to a few μm and T up to 1000 °C. This may enable graphene micro-patterns fabrication. This methodology could pave the way to large-scale production of patterned graphene.

Other strategies to create reactive dangling bonds directly on edges or GNRs have been developed.708 Thus, GO and CMG are also attractive for bio/medial applications, for the development of new biosensors, for bio-labelling and bio-imaging,564 for tissue engineering, for drug delivery and as antibacterial.

Another option is to induce magnetism by chemical functionalization. Theoretical studies predicted that defective graphene could be semiconducting and magnetic.713–715Ref. 716 reported a mixture of disordered magnetism regions (ferro, super-paramagnetic and antiferromagnetic) on graphene using nitrophenyl functionalization. The aim is to induce long-range ferromagnetic order by controlling the chemisorbed sites for spintronics.716

Functionalized graphene (FG) could be used as substrate for the deposition and organization of supramolecular layers and (or) enhance the local reactivity by inducing a curvature. Molecules are used either for doping,717,718 or can exploit graphene itself as substrate for the self-organization of supramolecular layers,719,720 and (or) Moiré patterns, e.g. in the case of graphene/Ru(0001).721,722

The outstanding issue is to understand the electronic interaction between molecules and graphene and the balance between molecule–molecule and molecule–substrate interaction for the realization of supramolecular networks.

A demonstration of the possibility to functionalize graphene with individual molecules is the molecular spin valve device reported in ref. 723, made by decorating a graphene nanoconstruction with TbPc2 magnetic molecules.723 These experiments open a wide research field and several intriguing questions on spintronics. LPE of intercalated graphite. GICs are formed by periodic insertion of atomic or molecular species (intercalants) between graphene layers.724,725 GICs are typically characterized in terms of a ‘staging’ index m, i.e. the N between two adjacent intercalant layers. Thus, e.g., a stage 3 GIC (see Fig. 45) has each 3 adjacent graphene layers sandwiched by 2 intercalant layers726 (the latter can also be more than 1 atom thick).
image file: c4nr01600a-f45.tif
Fig. 45 Graphite intercalation compounds. In stage 1, SLG alternate with intercalant layers. In stage 2, stage 3, etc., 2, 3, etc. graphene layers separate two intercalant layers. Adapted from ref. 602.

GICs have a long history since the first recorded production by Schafhaeutl in 1840.724Ref. 726, 727 summarized the historical development of GICs. The first determination of stage index by X-ray diffraction was done in 1931 by Hoffman and Fenzel.728 Systematic studies started in the late 1970s.

Intercalation of atoms or molecules with different m gives rise to a wide variety of electrical,726 thermal726 and magnetic properties.726 GICs have potential as highly conductive materials.726,729–731 GICs with metal chloride or pentafluoride intercalants, such as antimony pentafluoride (SbF5) and Arsenic pentafluoride (AsF5), received much interest since the 1970s.726,729–731E.g., AsF5-GICs has slightly higher σ (6.3 × 105 S cm−1)729 than bulk Cu730,731 (5.9 × 105 S cm−1),729 while the graphite in plane σ is ∼4.5 × 104 S cm−1.732 The σ increase is assigned to injection of carriers from the intercalate layer, with low μ, to the graphite layers, with high μ.726

GICs can be superconducting733 with transition temperatures up to 11.5 K for CaC6 GICs at ambient pressure,734 and higher with increasing pressure.735 Moreover, GICs are also promising for hydrogen storage, due to a larger interlayer spacing.736 GICs are already commercialized in batteries,737 in particular, in Li-ion batteries since the 1970s.738–740 GICs have also been used as negative electrodes (anode during discharge) in Li-ion batteries with the introduction of solid electrolytes.741,742

The most common production strategies include two-zone vapour transport726,743,744 exploiting T differences between graphite and intercalants745 and, sometimes, the presence of gases,745e.g. Cl2 for intercalation of AlCl3.726 GICs can be produced in single (for binary or ternary GICs) or multiple steps, the latter when direct intercalation is not possible.746 Hundreds of GICs with donor (alkali, alkali earth metals, lanthanides, metal alloys or ternary compounds, etc.) or acceptor intercalants (i.e. halogens, halogen mixtures, metal chlorides, acidic oxides, etc.) have been reported.726,744

The intercalation process increases the graphite interlayer spacing, especially for low stage index GICs,747,748E.g., K, Rb or Cs-GICs have interlayer distance ∼0.53–0.59 nm, while larger intercalants, such as dimethylsulfoxide, give an interlayer distance ∼0.9 nm,748i.e. 1.5 to ∼3 times larger than the ∼0.34 nm spacing in pristine graphite. This makes GICs promising to produce graphene via LPE, even without ultrasonication.628,747–750 However, although the exfoliation process is often called spontaneous,628,750 due to the absence of ultrasonication, it requires mechanical energy, often provided by stirring.628,750 To date it is possible to exfoliate GICs with lateral sizes ∼20 μm with YM ∼ 90%,749 and μ ∼ tens cm2 V−1 s−1.749

Note that many GICs tend to oxidize in air,726,751 and require a controlled ambient for their processing.726,751 This, coupled with the additional steps for GIC production, is one of the primary reasons why GICs are not yet extensively used to produce graphene via LPE. However, ref. 752 recently reported FeCl3 intercalated FLGs air-stable for up to one year.

Aspects of the intercalation mechanism, key for applications, still need to be clarified. This has implications for life duration of Li-ion batteries.753–755 The role of the solvent and the search for novel strategies for intercalation are also crucial, particularly to achieve large quantities of LPE graphene. Some GICs were shown to be spontaneously soluble in polar solvents without need of sonication or high shear mixing.628,750

Other open questions are: what is the role of the intercalant, the charge transfer to the graphene layer and the modification of the graphene band structure.

4.2. Growth on SiC

The production of graphite from SiC, Fig. 39e, was reported by Acheson as early as 1896 (ref. 59) for lubricant applications.59 The growth mechanism has been investigated since the 1960s.760 Both polar surfaces, the Si-terminated (0001) and the C-terminated (000[1 with combining macron]) surface, annealed at high T under UHV tend to graphitize, because of Si evaporation.756,757Ref. 758–760 reported the production of graphene films by thermal decomposition of SiC above 1000 °C. This is not a self-limiting process and areas of different film thicknesses may exist.759

On the (0001)-surface (see Fig. 46) the graphene layer is grown on top of a C-rich (6√3 × 6√3)R30° reconstruction with respect to the SiC surface, called the buffer layer.761 This consists of C atoms arranged in a graphene-like honeycomb structure,761 but without graphene-like electronic properties, because ∼30% are covalently bonded to Si.761

image file: c4nr01600a-f46.tif
Fig. 46 (a) Top view of covalently bound stretched graphene (CSG) model on SiC(0001). (b) graphene on bulk-truncated SiC(0001) surface, Adapted from ref. 761.

The buffer layer can be decoupled from the Si(0001)-face by hydrogen intercalation,763 becoming a quasi-free-standing SLG with typical linear π bands.763 In contrast, the interaction between graphene and the C(000[1 with combining macron])-terminated face is much weaker.761

Growth of graphene on SiC is usually referred to as “epitaxial growth”,764 even though there is a large lattice mismatch between SiC (3.073 Å) and graphene (2.46 Å) and the carbon rearranges itself in a hexagonal structure as Si evaporates from the SiC substrate, rather than being deposited on the SiC surface, as would happen in a traditional epitaxial growth process. The term “epitaxy” derives from Greek, the prefix epi means “over” or “upon” and taxis means “order” or “arrangement”. In 1928 Royer765 used the term “epitaxy” referring to the “oriented growth of one substance on the crystal surface of a foreign substance”. If the growing crystal and the substrate have the same lattice constants these are lattice matched.766 The use of “epitaxial” as the adjectival form of epitaxy has been subjected to some criticism already in the sixties, because it is incorrect from the philological point of view.767 Epitactic is the correct form.767 In 1965 epitaxic was recommended by ref. 768. However, the word “epitaxial” is now widely used and any attempt to change it is unrealistic.602 We will thus use “epitaxial” as adjectival form of epitaxy. There are two general epitaxial growth processes depending on the substrate, homo- and hetero-epitaxy. In the case of homoepitaxy the substrate is of the same composition and structure as the growing film, whereas in the case of heteroepitaxy the substrate is of a different composition and may not be perfectly lattice matched.602

It would be desirable to grow graphene on a lattice matched isostructural substrate in order to minimize defects, like misfit dislocations, as in the case of traditional semiconductors.769 However, with the exception of graphite, where the growth would be referred to as homoepitaxy, and would not be very useful for obvious reasons, there are few substrates that are isostructural and nearly lattice matched to graphene. There are two potential substrates that might meet the aforementioned requirement, h-BN and hexagonal closed packed (hcp) Co.602 H-BN has the lowest lattice mismatch ∼1.7%. Cobalt metal (hcp at T < 400 °C) also has a small lattice mismatch ∼2%.602 There are other hcp metals like Ru, Hf, Ti, Zr, but these have much larger lattice mismatch770 than that between Co and graphene, and there are face centre cubic metals like Ni, Cu, Pd, Rh, Ag, Au, Pt and Ir that have a range of lattice mismatch on the (111) planes.602 Therefore, from an epitaxial growth perspective, it would be desirable to grow on oriented single crystal Co (see section 4.3, 4.4) as performed by ref. 771. Growth on Co would also require transfer to other non-metallic substrates, discussed later. SiC substrates could be a natural substrate were if not for the fact that the lattice mismatch between graphene and SiC is also very large, ∼25% for both 4H-SiC (Si-face) and 6H-SiC (C-face). There have been reports of growth of LMs on highly non-lattice-matched substrates as buffer layers, due to their weak bonding to the underlying substrates,772–774 In this case, the films grow parallel to the substrate because of the anisotropic nature of their chemical bonds. Growth of graphene on SiC might be described in a similar manner.772–774

The growth rate of graphene on SiC depends on the specific polar SiC crystal face.775,776 Graphene forms much faster on the C-face than on the Si-face.775,776 On the C-face, larger domains (∼200 nm) of multilayered, rotationally disordered graphene are produced.777,778 On the Si-face, UHV annealing leads to small domains, ∼30–100 nm.778 The small-grain structure is attributed to morphological changes of the surface in the course of high-T UHV annealing.759 Indeed, ref. 779, via energy-resolved maps of the local density of states of graphene grown on Si-face, revealed modulations on two different length scales, reflecting both intra-valley and inter-valley scattering due to in-plane atomic defects. These defects in UHV annealed SiC are related to the relatively low growth T and the high graphitization rates in the out of equilibrium UHV Si sublimation process.775

Different strategies were proposed to control the Si sublimation rate. Ref. 780 used Si in a vapour phase establishing thermodynamic equilibrium between the SiC sample and the external Si vapour pressure to vary the T of the phase transition from the Si-rich (3 × 3) to the C-rich (6√3 × 6√3)R30° structure, the buffer layer, and the final graphene layer. The resulting domains were an order of magnitude larger than those grown under UHV.762

Ref. 759 used the “light bulb method” to grow graphene, exploiting a 80-year old process first developed to extend the lifetime of incandescent light bulb filaments.781 This uses Ar in a furnace at near ambient pressure (1 bar) to reduce the Si sublimation rate.759 Indeed, in Ar no sublimation was observed until 1500 °C,759 whereas Si desorption started at 1150 °C in UHV,759 thus enhancing surface diffusion, with complete surface restructuring before graphene formation.759 The resulting films on the Si-face have ∼50 μm domains,759 almost 3 orders of magnitude larger than in UHV annealing.777,778

Si sublimation can also be controlled by confining SiC in a graphite enclosure (either in vacuum775 or in an inert gas775) limiting the Si escape, maintaining a high Si vapour pressure. This keeps the process close to thermodynamic equilibrium, producing either SLG775 or FLG775 films over large (cm scale) areas, both on Si- and C-face. High T annealing can also give GNRs and GQDs.782,783

To date, graphene grown on the Si-face has a RT μ up to ∼500–2000 cm2 V−1 s−1 (ref. 775), with higher values on the C-face (∼10[thin space (1/6-em)]000–30[thin space (1/6-em)]000 cm2 V−1 s−1).775–777 For near-intrinsic samples (8.5 × 1010 cm−2)784 RT mobilities up to ∼150[thin space (1/6-em)]000 cm2 V−1 s−1 on C-face785 and ∼5800 cm2 V−1 s−1 on Si-face785 were reported. Ref. 100 reported an exceptionally high μ in a 40-nanometre-wide GNR grown on the (0001) face of SiC. These GNRs have shown ballistic conductance (at 4 K) on a length scale greater than ten micrometres with μ ∼ 6 × 106 cm2 V−1 s−1.100 This is equivalent to Rs ∼ 1 Ω □−1 (ref. 100).

Graphene on SiC has the benefit that SiC is an established substrate for power electronics.786 Top gated transistors have been fabricated from graphene on SiC on a wafer scale.787 High frequency transistors have also been demonstrated with 100 GHz cut-off frequency788 higher than state-of-the-art Si transistors of the same gate length.433 Graphene on SiC has been developed as a resistance standard based on QHE.146,148,151

A drawback for this technology for large scale production is the SiC wafers cost (∼$150–250 for 2′′ wafer789 at 2011 prices, compared to ∼$5–10 for same size Si wafers) and their smaller size (usually no larger than 4′′) compared to Si wafers. One approach to reduce substrate costs is to grow thin SiC layers on sapphire, the latter costing less than ∼$10 for 2′′ wafer,790 and subsequently perform thermal decomposition to yield FLG.791 Thus far, FLGs produced in this way have inferior structural and electronic quality compared to bulk SiC. Another approach is to grow SiC on Si.792 However SiC on Si is usually cubic,793–795 making it challenging to achieve continuous high quality graphene, due to bowing and film cracking as a consequence of high residual stress.219,220,796,797Ref. 798 grew SLG on 3C–SiC(111) with domains ∼100 μm2, combining atmospheric pressure growth759 with hydrogen intercalation,763 demonstrating that large domains can grow on 3C–SiC(111).

Future challenges are the control on the layer thickness homogeneity (currently not 100% monolayer)602 probably via better control of unintentional mis-cut angles, understanding and control of unintentional doping caused by the substrate, together with a better understanding of the effect of structural in-homogeneities (e.g. steps, wrinkles, BLG inclusions) on transport, and the mechanisms limiting μ. Other targets are the growth of graphene on pre-patterned SiC substrates, and a better control of growth on the SiC C-face, so to have SLG also on this face. The aim is also to tune the properties of graphene grown on SiC via interface engineering (e.g. SiC surface hydrogenation, etc.) and better understanding of defects generated during growth and/or interface manipulation and identification of methods to heal them. Doping of graphene by insertion of heteroatoms needs also to be addressed, with the aim to have control on the procedures and consequently on the properties of the graphene flakes. The achievement of many of the aforementioned goals will require a careful investigation and improvement of the substrates. A critical point to be addressed is the transfer of SLG directly from a SiC surface to a target substrate, where the difficulty arises from the strong binding of graphene to the SiC surface.761,775,776 This issue was addressed by ref. 799 that developed a method in which a film consisting of SLG or BLG grown on SiC is exfoliated via the stress induced with a Ni film and transferred to another substrate. Other points to be addressed are routes to improve growth of cubic SiC as substrate, and the growth of insulating SiC layers on cheap on-axis n-type substrates, in order to replace expensive semi-insulating substrate materials.

The graphene quality and N are strongly dependent on the growth and annealing conditions. The advantage over standard CVD is the graphene quality control achievable via tuning of carbon source thickness and annealing conditions. In addition, all the process steps occur in fully semiconductor compatible environment. Thus, industry can then take benefit of the versatility of this method to integrate graphene in their process flow. The long term goal is a totally controlled graphene nano-structuring, so to produce GNRs and GQDs on demand. This is motivated by the prospect of band gap creation in graphene.

4.3. Growth on metals by precipitation

Carbon can be deposited on a metal surface by a number of techniques, e.g.: flash evaporation, physical vapour deposition (PVD), CVD, and spin coating. The carbon source can be a solid,800,801 liquid,802,892,803 or gas.849 In the case of a pure carbon source, flash evaporation804 or PVD805 can be used to deposit carbon directly on the substrate of interest, if non-carbide forming but with a finite carbon solubility, before before diffusion at high T followed by precipitation of graphite (graphene) upon cooling. When the solid source is a polymer, it can be spun on the metal substrate at RT, followed by high T annealing and growth.801

Segregation from carbon-containing metal and inorganic substrates is another approach to graphene growth,806–808 see Fig. 39f. The first reports of synthetic growth of graphite, i.e. not extracted from mined natural sources, on transition metals date back to the early 1940s.806–808 It was not until the 1970s, however, when the details of the growth process were elucidated. Ref. 809 identified, via a combination of Auger and low-energy electron diffraction (LEED), SLG formed from carbon precipitation, following high T annealing of Co, Pt, or Ni. This process exploits the T-dependent solubility of interstitial carbons in transition metals (i.e. Ni(111),810 Ru(001),811 Ir(111),812 Pt(111),813 Pd(100),813etc.) or inorganic chemical compounds (i.e. LaB6[thin space (1/6-em)]814 ) to achieve layer-by-layer growth of graphene on the surface. The amount of carbon that can be dissolved in most metals is up to a few atomic%.815

Materials for substrates are selected among those which do not form carbides, in order to eliminate the competition between forming a carbide and graphite/graphene growth. Elements like Ti, Ta, Hf, Zr and Si, etc. form thermally stable carbides, as shown by the phase diagram,816–820 thus are not “ideal” substrates for graphite/graphene growth. Moreover, all these have a large (>20%) lattice mismatch with graphene. The transition metal is first annealed to high T (>1000 °C) in UHV, where the bulk solubility of interstitial carbon is high, then cooled to decrease solubility, resulting in its segregation as a graphene film.821

Graphite can also be obtained from carbon saturated molten iron during the formation of steel.822 In accordance to the process just described, Fe is supersaturated with carbon, and the excess carbon precipitates.822 This is usually referred to as “Kish graphite”,823 derived from the German “Kies”, used by steel workers to refer to the “mixture of graphite and slag separated from and floating on the surface of molten pig iron or cast iron as it cools”.227

Significant attention has been devoted to the use of inexpensive metals such as Ni,824–826 and Co.827 Growth on noble metals such as Ir,812 Pt,813 Ru,811 and Pd,813 aimed at a better understanding of the growth mechanisms and/or at obtaining samples suitable for fundamental studies, e.g. for STM,828 that require a conductive substrate. It would be desirable in any case to have a stable metal that can promote graphene single crystal growth, and the use of (111) oriented Ni or Co could help in this sense, since they facilitate the hexagonal arrangement of carbon atoms.

The monolayer-graphite/graphite growth process on Ni was first investigated in 1974 in ref. 809. SLG on Ni(111) was observed at T > 1000 K by Auger analysis, followed by graphite formation upon cooling. During high T annealing, carbon diffuses into the metal until it reaches the solubility limit. Upon cooling, carbon precipitates forming first graphene, then graphite.809 The graphite film thickness depends on the metal, the solubility of carbon in that metal, the T at which the carbon is introduced, the thickness of the metal and the cooling rate.

To get large metal grains with crystalline orientation i.e. Ni(111),829 an annealing of the metal surface is often performed. Carbon segregation in Ni(111) was investigated in ref. 830, 831 with control of number of surface C atoms by adjusting the annealing T.831

The graphene–metal distance and its nano-rippling on the metallic substrate determine the so-called graphene–metal Moiré superstructure.812,832 The latter is due to a mismatch between substrate and graphene lattice, and depends on the metal substrate. For lattice mismatches between graphene and substrate below 1%, commensurate superstructures, where the resulting broken symmetry is a doubling of the unit cell along one axis (i.e. 1/2, 0, 0), are formed.833 This is the case for Ni(111)821 and Co(0001).834 On the other hand, larger mismatches yield incommensurate (total loss of symmetry in a particular direction, i.e. (0.528,0,0)) moiré superstructures, such as in Pt(111),835 Ir(111),836 or Ru(0001)832,837 Indeed, graphene grown on Ir(111) yields flakes of well-defined orientation with respect to the substrate.812 On the contrary, graphene obtained via high-T segregation of C on Ru(0001) has a spread of orientations.837 Moreover, the graphene/Ru lattice mismatch results in a distribution of tensile and compressive strains.838 This causes corrugation, due to buckling, and the formation of ∼1.7 Å humps.838 The Moiré superstructure could be eliminated by the adsorption of oxygen on the metal surface acting as intercalant.839

Growth of graphene by precipitation requires careful control of the metal thickness, T, annealing time, cooling rate, and metal microstructure. Ref. 840 reported growth on Ni, Co and Ru on sapphire. Through the suppression of grain boundaries, ref. 840 demonstrated uniform growth on Ru by a surface catalyzed reaction of hydrocarbons, but not on Ni and Co.840 Both SLG and FLG were observed on Ni and Co, presumably due to the higher solubility of carbon and incorporation kinetics in comparison to Ru at the same T.840 However, ref. 771 grew graphene on epitaxial Co on sapphire, achieving SLGs, in contrast to FLGs in ref. 840. An alternative strategy for SLG growth on high C solubility substrates was proposed by ref. 841 using a binary alloy (Ni–Mo). The Mo component of the alloy traps all the dissolved excess C atoms, forming molybdenum carbides and suppressing C precipitation.841 Graphene was also grown on Ru(0001) on sapphire.842

One of the shortcomings of the growth on metals is that most applications require graphene on an insulating substrate. Ref. 843 suggested that graphene can be grown directly on SiO2 by the precipitation of carbon from a Ni film deposited on the dielectric surface. This process has favourable perspectives but needs further refinement.

The aim of the work on graphene produced by carbon segregation is eventually to achieve a full control on graphene quality and N. These are strongly dependent on the growth and annealing conditions and a full control on the latter is still missing. Deeply connected with this first issue are the requirements to obtain high quality graphene, where defects such as grain boundaries, pentagon–heptagon pairs, point defects,844 wrinkles,845 or local deformations are avoided or minimized. A further open issue is the transfer of graphene produced via carbon segregation onto arbitrary substrates and a process optimized for the various graphene applications still requires significant effort.

4.4. Chemical vapour deposition

Chemical vapour deposition is a process widely used to deposit or grow thin films, crystalline or amorphous, from solid, liquid or gaseous precursors of many materials. CVD has been the workhorse for depositing materials used in semiconductor devices for decades.846

The type of precursor is usually dictated by what is available, what yields the desired film, and what is cost-effective for the specific application. There are many different types of CVD processes: thermal, plasma enhanced (PECVD), cold wall, hot wall, reactive, and many more. Again, the type depends on the available precursors, the material quality, the thickness, and the structure needed; cost is also an essential part of selecting a specific process.

The main difference in the CVD equipment for the different precursor types is the gas delivery system.847 In the case of solid precursors, the solid can be either vaporized and then transported to the deposition chamber,847 or dissolved using an appropriate solvent,847 delivered to a vaporizer,847 and then transported to the deposition chamber.847 The transport of the precursor can also be aided by a carrier gas.847 Depending on the desired deposition temperature, precursor reactivity, or desired growth rate, it may be necessary to introduce an external energy source to aid precursor decomposition.

One of the most common and inexpensive production methods is PECVD. The creation of plasma of the reacting gaseous precursors allows deposition at lower T with respect to thermal CVD. However, since plasma can damage the growing material, one needs to design the equipment and select process regimes that minimize this damage. The details of the growth process are usually complex, and in many cases not all of the reactions are well understood. There are many different ways to perform plasma assisted CVD (see ref. 848 for an overview). It is however important to match the equipment design with the material one is trying to deposit and the precursor chemistry. Graphene should be simpler than multi-component systems, since it is a single element material. As with many other materials, graphene growth can be performed using a wide variety of precursors, liquids, gases, solids, growth chamber designs, thermal-CVD or PECVD, over a wide range of chamber pressures and substrate T.

4.4.1. Thermal CVD on metals. In 1966 Karu and Beer849 exposed Ni to methane at T = 900 °C to form graphite, to be used as sample support for electron microscopy. In 1969, May850 reported that thermal decomposition of acetylene and ethylene led to ring-like LEED patterns, as in the case of carbon on Pt, and that the material on the surface contained carbon.850 His further analysis suggested that the ring-like patterns (polycrystalline) were due to rotationally disordered graphite.850 He also discussed the growth of monolayers,850 as the first step to grow graphite, as revealed by X-ray diffraction. Thus, ref. 850 indicated the possibility of forming graphene on metals by CVD. In 1971 ref. 851 observed the formation of FLG via evaporation of C from a graphite rod.

In 1984 Kholin et al.852 grew graphene by CVD on Ir, to study the catalytic and thermionic properties of Ir in the presence of carbon.853 Since then, other groups exposed metals, such as single crystal Ir833,854 to carbon precursors and studied the formation of graphitic films in UHV systems.

The first studies of graphene growth on metals were primarily focused on the understanding of the catalytic and thermionic activities of the metal surfaces in the presence of carbon.855 After 2004, the focus shifted to the actual growth of graphene. Low pressure chemical vapour deposition (LPCVD) on Ir(111) using an ethylene precursor was found to yield graphene structurally coherent even over the Ir step edges.833 While Ir can certainly be used to grow graphene by CVD because of its low carbon solubility,815 it is difficult to transfer graphene to other substrates because of its chemical inertness. Ir is also expensive. Growth on Ni856 and Co827,857, metals compatible with Si processing since they have been used for silicides for over two decades,858–862 and less expensive than Ir, poses a different challenge, i.e. FLGs are usually formed,825–827,849,854 and SLGs are non-uniform. Therefore, while many papers claim CVD growth at high T on Ni and Co,825–827,849,854 the process is in fact carbon precipitation, not yielding uniform SLG, but rather FLG films. The shortcoming of high solubility or expensive and chemically unreactive metals motivated the search for a process and substrate that would be better suited to yield SLG.

The first CVD growth of uniform, large area (∼cm2) graphene on a metal surface was reported in 2009 by ref. 6 on polycrystalline Cu foils, exploiting thermal catalytic decomposition of methane and low carbon solubility. This process is almost self-limited, i.e. growth mostly ceases as soon as the Cu surface is fully covered with graphene, save ∼5% of the area, consisting of BLG and 3LG.6,863 Large area graphene growth was enabled principally by the low C solubility in Cu,864 and Cu mild catalytic activity.865

Growth of graphene on Cu by LPCVD was then scaled up in 2010 by ref. 7, increasing the Cu foil size (30 inches), producing films with μ ∼ 7350 cm2 V−1 s−1 at 6 K. Large grain, ∼20–500 μm, graphene on Cu with μ ranging from ∼16[thin space (1/6-em)]400 to ∼25[thin space (1/6-em)]000 cm2 V−1 s−1 at RT after transfer to SiO2 was reported in ref. 866 and from ∼27[thin space (1/6-em)]000 to ∼45[thin space (1/6-em)]000 cm2 V−1 s−1 on h-BN at 1.6 K in ref. 867. There is now a huge effort towards the optimization and the growth of high quality single crystal graphene by LPCVD. Ref. 868 synthesized large (∼4.5 mm2) single crystal hexagonal SLG domains on commercial polycrystalline Cu foils, indicating the potential for large scale at low cost. The as synthesized graphene had μ ∼11[thin space (1/6-em)]000 cm2 V−1 s−1 on a SiO2/Si substrate at RT. Ref. 869 by controlling surface oxygen, has repeatably grown centimeter-scale single-crystal graphene domains. The μ measured for these single-crystal graphene samples ranged from ∼40[thin space (1/6-em)]000 to ∼65[thin space (1/6-em)]000 cm2 V−1 s−1 at 1.7 K and from ∼15[thin space (1/6-em)]000 to ∼30[thin space (1/6-em)]000 cm2 V−1 s−1 at RT. Since it is not possible to measure the films while still on Cu, it is difficult to determine if there is degradation as a result of transfer.

The current understanding of the growth mechanism is as follows: carbon atoms, after decomposition from hydrocarbons, nucleate on Cu, and the nuclei grow into large domains.866,870 The nuclei density is principally a function of T and pressure and, at low pressure, mTorr, and T > 1000 °C, very large single crystal domains, up to ∼1 cm[thin space (1/6-em)]869 are observed. However, when the Cu surface is fully covered, the films become polycrystalline, since the nuclei are not registered,6,865,868,870i.e. they are mis-oriented or incommensurate with respect to each other, even on the same Cu grain. This could be ascribed to the low Cu–C binding energy.871 It would be desirable to have substrates with higher binding energy with C.871 However, while Ru is compatible with Si processing,872 oriented Ru films may be difficult to grow on large (300–450 mm) Si wafers, or transferred from other substrates.

Graphene nucleation needs to be investigated, to control and enhance domain size, via controlled and multi-step exposures. Basic reaction kinetics and layer-by-layer growth need to be investigated by dedicated surface science experiments. Real-time spectroscopy could in principle be of great help to monitor CVD-graphene growth and understand thickness distribution as well as optimize the growth conditions. Cu pre-treatments, electrochemical polishing,873,874 and high-pressure annealing868 are shown to be critical for suppressing graphene nucleation site density. Electrochemical polishing873,874 cleans the Cu surface and removes the impurity layer.

There are some difficult issues to deal with when growing graphene on most metal substrates, especially Cu, because of the difference in thermal expansion coefficient between Cu and graphene, of about an order of magnitude.875 The thermal mismatch gives rise to a significant wrinkle density upon cooling.863 These wrinkles are defective, as determined by Raman spectroscopy,866 and may also cause significant device degradation through defect scattering, similar to the effect of grain boundaries on μ in semiconducting materials.866 These defects however, may not be detrimental for many non-electrically-active applications, such as transparent electrodes. Perhaps one could use relatively cheaper substrates such as Cu (Cu is cheaper than Ir, Ru, Pt) and use an electrochemical process to remove graphene while reusing Cu, so that the cost is amortized over many growth runs. Because of some unattractive properties (e.g. surface roughening and sublimation) of Cu at the current thermal CVD growth T > 1000 °C, the community has been searching for new substrates that take advantage of the self-limited growth process, in addition to dielectrics. Ref. 876 reported growth of SLG on Ni(111) at lower T, 500–600 °C, using ethylene by UHV CVD, and identified the process as self-limiting, presumably due to the low C solubility in Ni at T < 650 °C.877 However, the T range within which graphene can be grown on Ni is narrow, 100 °C,876 and could result in a Ni2C phase,876 which can give rise to defects in the Ni crystal. Thus one could surmise that any graphene could be non-uniform across the Ni–Ni2C regions.

Graphene was also grown on Cu by exposing it to liquids or solid hydrocarbons.801,878Ref. 878 reported growth using benzene in the T range 300–500 °C.

The presence of the substrate generally modifies the graphene electronic properties, thus it is of paramount importance to optimize the interaction between graphene and substrate. This can be tuned by applying surface treatments that, in turn, can provide additional control on the graphene properties. E.g., substrates and interfacial dielectrics with optimized properties would enable high μ devices.

The process space for SLG-CVD growth is very wide and depends on many factors, from substrate choice, to specific growth conditions, as well as variables not under direct control. It is critical to know the material requirements for specific applications, so that one can tune the growth process/conditions to the application. Growth of graphene on single crystal substrates would be a desired route for improving electronic properties. Following the growth of graphene on Cu, ref. 879 developed a Co deposition process to form highly crystalline Co on c-plane sapphire where they grew SLG by CVD at high T. However, they did not distinguish between face centred cubic (fcc)(111)Co and hcp(0002)Co and did not comment on potential phase transformation issues at T lower than the fcc to hcp phase transition (∼400 °C). While this process may seem incompatible with Si processing, and the material cost could be high, it is important to learn how to take advantage of processes that enable growth of higher quality graphene on stable surfaces, not necessarily single crystals.

Another question is: can we controllably grow FLGs? Catalytic decomposition of CO on various metals, such as Fe, Cu, Ag, Mo, Cr, Rh, and Pd, was studied by Kehrer and Leidheiser in 1954.808 They detected graphitic carbon on Fe after exposure to CO for several hours at 550 °C, but found the other metals to be inactive. The presence of BLG and TLG on Cu6 poses the question of the growth process for these isolated regions, since at first one would like to grow uniformly SLG. Growth of controlled Bernal stacked films is not easy, but small regions have been observed.880Ref. 880 reported homogenous BLG by CVD on Cu. However, it not clear whether the films are of high enough quality for high performance electronic devices, since ref. 880 did not map the D peak, and μ was ∼580 cm2 V−1 s−1 at RT.

Another approach was proposed by ref. 881 by increasing the solubility of C in Cu via a solid solution with Ni, forming the binary alloy, Cu–Ni. By controlling Ni percentage, film thickness, solution T, and cooling rate, N was controlled, enabling BLG growth.881

Graphene films with size ∼50 cm and μ > 7000 cm2 V−1 s−1 were produced on Cu and transferred via a R2R process, see Fig. 47.7 The goal is now to grow high quality samples (with crystal size >1 mm) over large areas via R2R. Low pressure thermal CVD synthesis and a direct R2R transfer using photocurable epoxy resin was used to fabricate a 100 m-long graphene transparent conductive film (GTCF) with a Rs as low as 150 Ω □−1.882

image file: c4nr01600a-f47.tif
Fig. 47 Roll-based production of graphene. (a) Schematic of the process. (b) A Cu foil is wrapped on a 7.5 inch quartz tube, then placed into an 8-inch quartz reactor. (c) R2R transfer of graphene from to a Polyethylene terephthalate (PET) film.7

Among the many areas that still need to be explored, growth on metallic alloys need yet to be investigated, the optimisation of CH4–H2–Ar mixtures has yet to be done, and the same can be said about testing and screening of alternative precursors. So far there is also little experimental insight into the underlying GB formation mechanisms, crucial to understand and control charge propagation across/within these line-defects. On one hand, there is a need for quick and easy characterization methods, able to reveal the grain structure of the CVD samples.883 On the other hand, GBs can be highly transparent,884 as well as perfectly reflective,884 and are expected to act as molecular metallic wires885 or filter the propagating charge carriers based on the valley-index.886 In order to explore and exploit these properties, the investigation of the electronic properties of individual GBs with known atomic configuration is needed. STM is a versatile tool for investigating the structure of individual GBs at atomic resolution and their electronic (and magnetic) properties on the nm scale.307

Growth of graphene on single crystal substrates is another route towards the improvement of its electronic properties, although the high cost of such substrates makes them less suitable for large-scale applications. This requires in situ growth monitoring. TEM will be fundamental to obtain atomic images of domain boundaries, as well as macroscopic images of the relative domain orientations in a film.241

Reducing the growth T is desirable in order to cut production costs, and directly integrate graphene with standard CMOS processing. Growth of FLG at 650 °C was demonstrated on Fe.887 However, the optimization of layer control and growth T below 450 °C is required for CMOS integration. Although the growth of graphene at 325 °C was shown on MgO,888 its suitability for applications is yet to be elucidated. Graphene grown on Cu foils at T as low as 300 °C using benzene as a precursor was also reported.889

MWCVD has also been proposed as a method for graphene growth.890,891Ref. 891 reported growth at 150–300° C. However, to date SLG grown has not yet been demonstrated, with the deposited films consisting of sub-micrometer flakes.891 Nevertheless, this approach was successfully used for the production of TCs.891 With further developments it could be a viable strategy for large scale, low T graphene production.891

A key target is thus the development of low T growth on large area, by understating and optimising plasma chemistry.

Low-T, high-density pulsed micro wave plasmas (electron density close to 1012 cm−3 ) are not only usable for large area growth, but also for surface processing and work-function engineering, important for chemical functionalization, and for μ control. Systems scalable to large areas should be investigated.

The production of unsupported flakes using alcohols as carbon feedstock was also demonstrated,892,893 with the potential for up-scaling. The target is to further up-scale the CVD-growth, developing a protocol for a batch reactor, producing m2-sized graphene.

The development of new routes to achieve stable doping is another key need for commercialization. Different approaches are to be investigated, either during growth, by exploiting alternative precursors (i.e. pyridine), or post-growth, via “molecular doping” by stable hydrazil- and nitroxide- radicals and metal grids.

The growth of graphene on non-flat substrates such as, e.g., Cu wires or corrugated Cu substrates instead of flat Cu films, is an area to be investigated that could open new applications both in electronics, energy and catalysis.

Another key point is the development of reliable, fast, economic and environmentally friendly transfer techniques. Methods for the recovery of the metal substrates are also needed for cost reduction and environmental issues.

4.4.2. CVD on semiconductors and insulators. Electronic applications require graphene grown, deposited or transferred onto dielectric surfaces. Growth of high-quality graphene layers on insulating substrates, such as SiO2, SiC, sapphire, etc., would be a major step forward towards the applications in nano-electronics. The use of SiC wafers in microelectronics is becoming increasingly popular894 (see, e.g., the new SiC MOSFET developed by CREE894) which should result in lowering of their prices. SiC substrates of up to 150 mm diameter should become available in the short term.

CVD of carbon thin films on insulators has been known since 1971.895 However, thus far, it was optimized to give highly sp3 bonded diamond-like carbons.199 Recently this approach was developed to achieve graphitic films.896,897 Growth of high-quality graphene on insulating substrates, such as SiO2, high-[scr K, script letter K] dielectrics, h-BN, etc. would be ideal for electronics. There have been many attempts to grow on SiC,898 sapphire899 and Si3N4/Si,900 as well as on metal oxides such as MgO,888 and ZrO2.901 However, while graphitic regions are observed at T < 1000 °C, none of the processes yield, to date, planar SLG films covering the whole surface.899 Thus far, the best quality was achieved on sapphire899 (3000 cm2 V−1 s−1 and 10[thin space (1/6-em)]500 cm2 V−1 s−1 at RT and 2 K, respectively).

h-BN was shown to be effective as a substrate for graphene CVD,902–905 and graphene produced by this method appears comparable to that grown on transition metal catalysts833,906 with promise for hetero-epitaxial growth of heterostructures (e.g. graphene/h-BN). As well as achieving direct growth on an insulator, this approach has the additional benefit of an atomically smooth substrate, comparable with diamond-like carbon (DLC) surfaces,907 with few dangling bonds and charge traps.908 Direct growth of graphene/h-BN stacks, by both CVD and metal–organic CVD (MOCVD), is the ideal alternative to tedious successive exfoliations of rare BN single crystals (mostly one source in Japan909). CVD can also give c-BN composite layers with various topologies.910 Different B and N precursors (solid, liquid, gaseous) should be tested in a variety of environments, aiming at optimum quality and layer control, and privileging less costly and harmful ones. Wafer scale extension of BN/graphene encapsulation techniques would pave the way to transport in high-frequency electronics,788 a regime exploiting Dirac Fermion optics, with no counterpart in semiconductor electronics. Substrate tailoring may optimize these properties in a broad (and economically relevant) spectrum, from microwave to optics, including millimeter waves (THz) and IR. The aim is to produce graphene on smart substrates in a single CVD run. At present, however, only FLGs were grown by CVD on h-BN, thus improved thickness control is needed. Understanding basic growth processes is needed, with the help of both in situ and ex situ characterizations. These procedures need to be adapted to more scalable conditions.

As mentioned above, growth of single crystal graphene on dielectric surfaces is highly desirable, but to date the crystal size on dielectrics is limited to micron size.911Ref. 912 reported the growth of single crystal monolayer graphene on germanium. The authors took advantage of the low (108–1010 cm−3)913 solubility of C in Ge as in the case of C in Cu to grow single crystal graphene on hydrogen-terminated Ge(110). They observed a low wrinkle density, and assigned it to the low thermal expansion coefficient difference between graphene and Ge.912 Due to the small (∼60 meV)914 binding energy between graphene and Ge, ref. 912 were able to mechanically exfoliate graphene using an Au film carrier, minimizing the defects usually created using organic material transfer methods. This is a major advance, although nucleation and surface flatness may still require optimization.

4.4.3. Plasma enhanced CVD. Plasma-enhanced CVD (PECVD) is a scalable and cost effective large area deposition technique, with numerous applications ranging from electronics (IC, interconnects, memory and data storage devices), to flexible electronics7 and photovoltaic.915 PECVD's strength relies in the potential of synthesizing graphene at lower T than conventional CVD in a graphene-on-insulator environment. This comes useful in multiple applications, especially when considering the process for CMOS devices.

The use of plasmas to reduce T during growth/deposition was extensively exploited in the growth of nanotubes,916–923 and amorphous carbon.200–203,924–926 In 1998 ref. 316 reported SLG with a curved structure as a byproduct of PECVD of diamond-like carbon. A number of other groups later reported growth of vertical SLG925 and FLG,919,924,926–929 by microwave PECVD on several substrates, including non-catalytic, carbide forming substrates, such as SiO2. SLGs and FLGs nucleate at the surface, but then continue to grow vertically, perhaps because of the high concentration of carbon radicals,891 thus resulting in high growth rate. This material is promising for supercapacitors or other applications, such as FE, not requiring planar films.

Graphene was grown by PECVD using methane at T as low as 500 °C,930 but the films had a significant D-band, thus with quality still not equivalent to exfoliated or thermal CVD graphene.891,930 Nevertheless, ref. 891 demonstrated that growth may be carried out at low T, and the material may be used for applications without the stringent requirements of electronics. E.g., ref. 891 used PECVD at 317 °C to make TCs with Rs ∼ 2 kΩ □−1 at Tr ∼ 78%.

PECVD might address the key issue of growth at low T, in order to prepare graphene directly on substrates compatible with applications and processing technologies listed above, e.g. direct deposition on plastic. The advantage of plasma methods with respect to thermal CVD, is the production of graphitic materials without metal catalyst over a wide range of growth conditions, both surface-bound and freestanding.917

However, large domains are needed, together with a reduction of damage caused by direct plasma during growth, that might limit the quality of graphene that can be achieved with this approach. Nevertheless, this approach seems to be promising for TCs.7

Inductively coupled plasma (ICP) CVD, (where the deposition pressure and the applied power to the plasma are controlled from two radio frequency-generators; one connected to the ICP-source, the other to the substrate931) was also used to grow graphene on 150 mm Si,931 achieving uniform films and good transport properties (i.e. μ up to ∼9000 cm2 V−1 s−1). This process is still under development with insufficient data on the structure of the material.

The long term target plan (>10 years) is to achieve on-demand graphene deposition on insulator/Si and other materials on 300–450 mm wafer size, in-line with the fabrication projections in the electronic industry. The challenge is to develop an integrated ALD-PECVD process that would allow deposition of compatible insulators at the same time as synthesising graphene. This should be done without compromising the quality of the graphene layer.

4.5. Molecular beam epitaxy growth of graphene on insulating surfaces

MBE is a UHV-based technique for producing high quality epitaxial structures with monolayer control.932 Since its introduction in the 1970s932 as a tool for growing high-purity semiconductor films, MBE has evolved into one of the most widely used techniques for epitaxial layers of metals, insulators and superconductors, both at the research and the industrial level. MBE of single crystal semiconductors, e.g. GaAs, is well-established and has produced hetero-junctions with the current record μ (3.5 × 107 cm2 V−1 s−1 (ref. 933)). MBE has also produced record low threshold current density multi-quantum-well lasers.934 MBE can achieve precise control of both the chemical composition and the doping profile. MBE can use a wide variety of dopants compared to CVD epitaxial techniques.

MBE can be used to grow carbon films (see Fig. 39h) directly on Si(111),935 and is a promising approach to achieve high-purity graphene heterostructures on a variety of substrates, such as SiC, Al2O3, Mica, SiO2, Ni, etc.

MBE is more suited to grow 2–6 inch wafers rather than 30-inch ones.7 MBE graphene may find industrial applications in markets where highly specialised devices are required. Despite the conceptual simplicity, a great technological effort is required to produce systems that yield the desired quality in terms of materials purity, uniformity and interface control. The control on the vacuum environment and on the quality of the source materials should allow higher crystal quality compared to non-UHV-based techniques. Although MBE of graphene is still very much in its infancy, there are a number of groups working on it. Multi-crystalline graphene has been reported, with crystal grain size up to 20–400 nm.936In situ growth of heterostructures could produce devices based on hybrid structures, combining graphene and semiconductors. Graphene can be grown directly on a wide variety of dielectric and metallic substrates as well as h-BN. Growth on MBE-grown h-BN is a possibility.

The aim is now to develop atomic beam epitaxy techniques for high-quality large-area graphene layers on any arbitrary substrates. In particular,targeted characteristics of MBE graphene are: high-μ samples – at least as good as exfoliated graphene on h-BN, i.e. ∼ a few 105 cm2 V−1 s−1 (at small carrier densities) and precise control over N, i.e. SLG/BLG/TLG. To date the growth process gives mainly polycrystalline graphite-like films,935 with lack of N control,937 because MBE is not a self-limiting process relying on the reaction between the deposited species932, and the reported RT μ is thus far very low (∼1 cm2 V−1 s−1).938 However with future optimizations, it may be possible to produce large area single crystal sheets on a wide variety of dielectric and metallic substrates. The fine control of doping, and the growth of hybrid semiconductor/graphene heterostructures, e.g. for heat management applications, should be investigated. MBE is also interesting for semi-transparent large-area electrodes, most of all in view of integration with Si technology.

Another benefit of MBE is that it is compatible with in situ vacuum characterization. Thus, the growth can be controlled by in situ surface sensitive diagnostic techniques, such as reflection high-energy electron diffraction, STM, XPS, etc.

One might even envisage the use of chemical beam epitaxy (CBE)939 to grow graphene in a catalytic mode, taking advantage of the CBE ability to grow or deposit multiple materials, such as dielectrics940 or LMs, on the top of graphene, to form heterostructures.

4.6. Atomic layer epitaxy

Atomic layer epitaxy (ALE) has not been as successful for semiconductor materials as is MBE. ALD,941 on the other hand, has been extensively used to produce thin layers of nano-crystalline binary metal nitrides (e.g. TaN, TiN),942,943 and high-[scr K, script letter K] gate dielectrics such as HfO2.944 ALD can controllably grown very thin, less than 1 nm, films941 but, to our knowledge, single atomic layers have not been commonly deposited on large areas.

Large area graphene can be grown by thermal CVD6,7,870 and PECVD891,930 using hydrocarbon precursors. A process dealing with a specific precursor and reactant could in principle be used in the ALE mode. However, to date there are no reports, to the best of our knowledge, of ALE-growth of graphene.

4.7. Heat-driven conversion of amorphous carbon and other carbon sources

Heat-driven conversion of amorphous carbon (a-C), hydrogenated a-C (a-C:H), tetrahedral a-C (ta-C), hydrogenated (ta-C:H) and nitrogen doped (ta-C:N) ta-C (for a full classification of amorphous carbons see ref. 197, 199), to graphene could exploit the extensive know-how on amorphous carbon deposition on any kind of substrates (including dielectrics) developed over the past 40 years. The process can follow two main approaches: (1) Annealing after deposition or (2) Annealing during the deposition.

Post-deposition annealing requires vacuum (<10−4 mbar),945–948 and T depending on the type of amorphous carbon and the presence of other elements, such as nitrogen946,947 or hydrogen.945,947,948Ref. 945 demonstrated that ta-C transitions from a sp3-rich to a sp2-rich phase at 1100 °C, with a decrease in electrical resistivity of 7 orders of magnitude from 107 to 1 Ω cm. A lower T suffices for a-C:H (∼300 °C)947 and ta-C:H (∼450 °C).947 For ta-C:H a reduction of resistivity is observed from 100 °C (R ∼ 1010 Ω cm) to 900 °C (R = 10−2 Ω cm).947Ref. 949 used a current annealing process for the conversion. However, it did not report the resulting transport properties.

Annealing during deposition allows the sp3 to sp2 transition to happen at lower T than post-deposition annealing.946,947,950,951Ref. 950 reported a reduction of resistivity of ∼6 orders of magnitude (R ∼ 108 Ω cm at RT and R ∼ 102 Ω cm at ∼450 °C). As in the case of post-processing, the presence of hydrogen (ta-C:H) or nitrogen (ta-C:N) changes the transition T.946Ref. 946 reported a transition for ta-C:N at ∼200 °C, with a much larger reduction, with respect to ta-C, of resistivity (∼11 orders of magnitude, R ∼ 108 Ω cm at RT and R ∼ 10−3 Ω cm at ∼250 °C), comparable with that of RGO films.701 However, unlike post-deposition annealing, annealing during deposition tends to give graphitic domains perpendicular to the substrate.947

Heat-driven conversion can also be applied to self-assembled monolayers (SAMs), composed of aromatic carbon rings.950Ref. 950 reported that a sequence of irradiative and thermal treatments cross-links the SAMs and then converts them into nanocrystalline graphene after annealing at 900 °C. However, the graphene produced via heat-driven conversion of SAMs had defects and low μ (∼0.5 cm2 V−1 s−1 at RT).950 Thus, albeit being simple and cost effective, at the moment the quality of the obtained material is poor, and more effort is needed targeting reduction of structural defects.

The aim is to develop reliable protocols to improve and exploit this process for a cheap and industrially scalable approach.

4.8. Chemical synthesis

In principle graphene can be chemically synthesized, assembling benzene building blocks,952,953 see Fig. 39i. In such approach, small organic molecules are linked through surface-mediated reactions at relatively low T (<200 °C). The resulting materials include nanostructured graphenes, which may be porous, and may also be viewed as 2d polymers. Graphene nanostructures could be obtained after polymerization of graphene-like molecular precursors in the form of polyphenylenes.320 By designing and synthesizing appropriate precursors, one should be able to scale up the formation towards (i) micron-sized graphene islands; (ii) nano-ribbons and nano-graphene(s) with a large variety of structures.954 The aforementioned methodology can be generalized to (1) sp2-like monolayers of BN, (2) nano-porous, “graphene-like” 2d structures, and (3) “graphene-like” organometallic co-polymers lattices based on phthalocyanines, for their applications in molecular spintronics.955 The chemical approach offers opportunities to control the nano-graphenes with well-defined molecular size and shape. Thus, properties that can be tuned to match the requirements for a variety of applications, ranging from digital and RF transistors, photodetectors, solar cells, sensors, etc. GNRs with well-defined band gap and/or QDs with tuneable absorption can already be designed and produced, Fig. 48.956 Such approaches will ultimately allow control at the atomic level, while still retaining the essential scalability to large areas.
image file: c4nr01600a-f48.tif
Fig. 48 Bottom-up fabrication of synthetic graphene and GNRs starting from 10,109-dibromo-9,99-bianthryl monomers (1). Top, dehalogenation during adsorption of the precursor monomers. Middle, formation of linear polymers by covalent interlinking of dehalogenated intermediates. Bottom, formation of fully aromatic GNRs by cyclodehydrogenation.320

Chemical graphenes tend to form insoluble aggregates.662 A common strategy to solubilise conjugated systems is the lateral attachment of flexible side chains.956 This was very successful in solubilising small graphene molecules, while failing for graphenes with increasing size,956 because the inter-graphene attraction rapidly overtakes the solubilisation forces, making the current strategy less and less effective.956 A possible approach relies on supramolecular interactions that can be used to cover SLG with PAHs composed of (i) an aromatic core able to interact strongly with graphene and (ii) flexible side chains to make them soluble in most organic solvents. NGs adsorb reversibly forming ordered layers, with precise control of orientation and spacing.952,953 These interact with the graphene backbone allowing in principle to control and tune its optoelectronic properties,952 while the NG flexible side-chains makes the graphene-NG composites soluble.957

Supramolecular interactions have the advantage of keeping intact the sp2 network, without compromising the transport properties.956 Possible applications include the integration of graphene with other chemical functionalities, such as metal containing dye molecules or reactive sites for the attachment of biological molecules.

Chemical synthesis is also suited for the formation of superstructures, whose physics is very rich. E.g., the rotation angle between graphene layers controls the carrier velocity.958

There is much more to do in terms of designing and tuning the strength and type of interaction with the substrate. None of the superstructure-induced effects have been thus far harnessed in real devices. Several directions as starting points for the realization of such structures and effects can be planned. E.g., the induced growth across atomic islands of insulating materials deposited by nano-stencilling in regular patterns on catalytic, atomically flat metallic surfaces. Alternatively, exploring routes (e.g. thermally, electric field controlled or through electronic excitations) for the initiation of cascade chemical reactions and assembly from NG precursors on insulating monolayers and nanostructures to form graphene origami959 and GNRs. Such assembly has only been demonstrated so far on atomically flat metallic surfaces, but molecular self-assembly processes show considerable promise and versatility. Novel, volatile, metallo-organic and organometallic complexes could be used as the precursors for this process which will provide a route at the molecular level.

The aim is to explore synthetic graphenes starting from the compatibility with a very large range of substrates and the easy association of organic and inorganic layers. The target is to control with atomic precision the shapes and edges, in order to tune continuously the band gaps and conductivity, as well as control doping, obtaining a spatial distribution of dopants with ultimate resolution. Indeed, as the device size is pushed down, the dopant distribution needs to be precisely tuned, which is very difficult to achieve by post-treatment of large area graphene, such as hydrogenation or fluorination (or even doping during thin film growth). Precise and tailored dopant distribution may be assured by means of a hetero- (dopant) atom that is readily part of the precursor molecule. The effect of local doping needs to be investigated. This could be in principle achieved functionalising graphene using acceptor/donor molecules that would self-assemble precisely on its surface. Another challenges for direct chemical growth include growth on insulating substrates for electronic applications and development for efficient transfer methods that allow the nanoribbons to be incorporated in electronics.

4.9. Nano-ribbons and quantum dots

Ref. 296, 330 prepared GNRs by combining e-beam lithography and oxygen plasma etching. GNR down to ∼20 nm were reported, with band gap ∼30 meV, then used in FETs with ION/IOFF up to 103 at low T (<5 K) and ∼10 at RT. Ref. 329 reported much smaller GNRs, with minimum width ∼1 nm and gap ∼500 meV produced by e-beam lithography and repeated over etching. Sub-10 nm GNRs with bandgap up to 400 meV were produced via a chemical route,97 consisting in the dispersion of expanded graphite in liquid phase followed by sonication. Used as channels in FETs, they achieved ION/IOFF up to 107 at RT.303 A solution-based oxidative process was also reported,304 producing GNRs by lengthwise cutting and unravelling SWNTs and MWNTs.332 As result of the oxidative process, such GNRs show poor conductivity (∼35 S cm−1) and low μ (0.5–3 cm2 V−1 s−1) at RT.960

Patterning of SLG into sub-10 nm GNRs with predetermined crystallographic orientation was achieved by STM lithography,307 by applying a bias higher than for imaging between the STM tip and substrate, while moving the tip at constant velocity.

GNRs can also be formed without cutting. Ref. 961 demonstrated that spatial selective hydrogenation can be used to create graphene “nanorods”, i.e. conductive paths of graphene surrounded by fully hydrogenated areas. Ref. 962 fabricated encapsulated ∼35 nm GNRs by depositing a polymer mask via scanning probe lithography, followed by chemical isolation of the underlying GNR by fluorinating the uncovered graphene. These GNRs retained μ of non-patterned graphene. Also, the fluorination is reversible, enabling write-erase-rewrite. GNRs down to 12 nm were produced by local thermal reduction of GO by scanning probe.963

Sub-10 nm GNRs were fabricated via catalytic hydrogenation, using thermally activated Ni nanoparticles as “knife”.310,964 This allows cutting along specific crystallographic directions, therefore the production of GNRs with well-defined edges.

GNRs were also made via LPE of GICs328 and expanded graphite.303 Growth on controlled facets on SiC resulted in 40 nm GNRs100 and the integration of 10[thin space (1/6-em)]000 top-gated devices on a single SiC chip.783

Chemical synthesis seems to be the most promising route towards well-defined GNRs,678 see Fig. 48. Atomically precise GNRs were produced by surface assisted coupling of molecular precursors into linear polyphenylenes and subsequent cyclo-dehydrogenation.678 GNRs up to 40 nm in length and soluble in organic solvents such as toluene, dichloromethane and tetrahydrofuran were synthesized336 from polyphenylene precursors having a non-rigid kinked backbone to introduce higher solubility in comparison to that of strictly linear poly(para-phenylene).965

Another route to GNRs is the so-called nanowire lithography,966 consisting in the use of nanowires as masks for anisotropic dry etching. GNRs smaller than the wire itself can be fabricated via multiple etching.966 Also, the wire, consisting of a crystalline core surrounded by a SiO2 shell, can be used as self-aligned gate.967

Arrays of aligned GNRs were produced by growing graphene by CVD on nanostructured Cu foils and subsequently transferring on flat Si/SiO2 substrates.968 The Cu structuring results in controlled wrinkling on the transferred material,968 which allows production of aligned GNRs by plasma etching.968

Besides their semiconducting properties, GNRs show other interesting properties, such as magnetoelectric effects.969 Also, half-metallic states can be induced in zigzag GNRs subjected to an electric field,325 chemically modified zigzag GNRs970 or edge-functionalized armchair GNRs.971 Half-metals, with metallic behaviour for electrons with one spin orientation and insulating for opposite, may enable current spin-polarization.325

Another approach to tune the bandgap of graphene relies in the production of QDs.956,972–977 These GQDs have different electronic and optical properties with respect to pristine graphene1,5 due to quantum confinement and edge effects.

Graphene oxide quantum dots (GOQDs) have been produced via hydrothermal972 and solvothermal973 methods (i.e. synthesis in an autoclave using aqueous and non-aqueous precursors, respectively) having lateral size ∼10 nm972 and ∼5–25 nm,973 respectively. Another route to produce GOQDs exploits the hydrazine hydrate reduction of small GO sheets with their surface passivated by oligomeric PEG.974 These GOQDs show blue PL for 365 nm excitation, while green PL for 980 nm excitation.974 GOQDs were also produced by electrochemical oxidation of a graphene electrode in phosphate buffer solution.975 These have heights between 1 and 2 nm and lateral size ∼3–5 nm.975 A bottom-up approach was used by ref. 976 to produce GQDs by metal-catalysed cage-opening of C60.

The fragmentation of the embedded C60 molecules at T ∼ 550 °C produced carbon clusters that underwent diffusion and aggregation to form GQDs.

As reported in section 4.9, GQDs can also be chemically synthesized, assembling PAHs,952,956 through surface mediated reactions. Ref. 977 exploited chemical synthesis to produce GOQDs by using an hexa-perihexabenzocoronene (HBC) precursor. GOQDs with ordered morphology were obtained by pyrolysis and exfoliation of large PAHs.977 The HBC powder was first pyrolyzed at a high T, then oxidized and exfoliated and reduced with hydrazine.977 The GOQDs had diameter ∼60 nm and thickness ∼23 nm, showing broad PL.977

4.10. Transfer and placement

4.10.1. Transfer, placement and shaping. The deterministic placement of graphene on arbitrary substrates is pivotal for applications and characterization. The ideal approach would be to directly grow it where needed. However, to date, we are still far from this goal, especially in the case of non-metallic substrates. The development of a transfer procedure is thus needed. This would also allow the assembly of novel devices and heterostructures, with different stacked 2d crystals.
4.10.2. Graphene membranes. Graphene membranes are extremely sensitive to small electrical signals,329 forces or masses978 due to their extremely low mass and large surface-to-volume ratio, and are ideal for NEMS. Graphene membranes have also been used as support for TEM imaging979 and as biosensors,980,981 Nanopores in SLGs membranes are used for single-molecule Deoxyribonucleic acid (DNA) translocation,980 paving the way to devices for genomic screening, in particular DNA sequencing,982,983 Thanks to its atomic thickness, graphene may be able to detect variation between two bases in DNA molecules,980 unlike conventional solid state nanopores made of Si3N4.984

Freestanding graphene membranes were first reported in ref. 134. Graphene samples were deposited by MC onto Si + SiO2 substrates, and then a grid was fabricated on them by lithography and metal deposition. Si was subsequently etched by tetramethylammonium hydroxide, leaving a metal cantilever with suspended graphene. This process was originally developed to fabricate suspended SWNTs.985Ref. 986 used the same approach to fabricate graphene membrane and study them by TEM, revealing that graphene sheets are not perfectly flat and exhibit intrinsic roughening. Ref. 987 fabricated mechanical resonators from SLG and FLG by mechanically exfoliating graphite over trenches in SiO2. Ref. 988 transferred graphene exfoliated either on SiO2 or polymer on TEM grids by first adhering to the grid and subsequently etching the substrate. Ref. 989 fabricated graphene membranes up to 100 μm in diameter by exfoliating graphite on a polymer and subsequently fabricating metal scaffolds on it by e-beam lithography and metal evaporation. The polymer was then dissolved leaving graphene membranes suspended on grids.989 A similar technique was used in ref. 990 to produce suspended samples to study graphene's optical transmission. Suspended graphene was obtained by contacting it via lithography and subsequently etching a trench underneath. This approach allowed to achieve ballistic transport at low T (∼4 K)145 and high μ (106 cm2 V−1 s−1).607 Suspending graphene drastically reduces electron scattering and samples suspended between contacts allowed observation of the fractional quantum Hall effect (FQHE).144,151

Graphene membranes have also potential for applications involving liquid separations, such as desalination, water filtration, DNA sequencing, as well as lab on chip microfluidic processes.991E.g., ref. 991 demonstrated that, when SLG is immersed in an ionic solution, it becomes an ionic insulators with a very small stable conductance that depends on the ion species in solution. This small effective thickness makes graphene an ideal substrate for very high resolution, high throughput, nanopore-based single-molecule detectors.991 The sensitivity of graphene's in-plane σ to its surface environment and trans-membrane solution potentials991 will offer new insights into atomic surface processes and sensor development opportunities.

4.10.3. Transfer of individual layers. Several transfer processes have been developed so far and can be classified either as “wet” or “dry”, see Fig. 49. The first includes all procedures where graphene is in contact, at some stage, with a liquid. In the second, one face of graphene is protected from contacting any liquid, while the other is typically in contact with a polymer, eventually dissolved by solvents.
image file: c4nr01600a-f49.tif
Fig. 49 Wet (left) and dry (right) transfer of graphene. Wet transfer: (a) A graphene sample is deposited on oxidized Si by MC. (b) A PMMA film is deposited by spin coating (c) the PMMA film is detached from the substrate either via NaOH etching or water intercalation. Graphene adheres to the polymer and is removed from the Si + SiO2 substrate (d) PMMA + graphene film is “fished” using the target substrate. By sliding the PMMA + graphene film with respect to the substrate a flake of choice can be aligned with features such as electrodes, cavities, etc. (e) Once the sample has dried, PMMA is dissolved by acetone releasing the graphene on the target substrate (f). (g) A graphene flake deposited onto BN by wet transfer. Dry transfer: (h) graphene is exfoliated onto substrates covered by a polymer stack consisting of a water-dissoluble polymer (such as PVA) at the bottom and PMMA on the top. (i) The sample is left to float in a water bath in such a way that the “release” layer is dissolved from the side. (j) Graphene is on top of the stack and is never in touch with water (k) the polymer + graphene film is attached to a special holder and flipped over. (l) By means of a manipulator the flake of choice is placed in the desired position on top of the desired substrate, then the film is pressed on the target substrate (m) PMMA is dissolved leaving graphene in the desired position (n) graphene layer deposited onto a BN flake by dry transfer. Adapted from ref. 602.
4.10.4. Wet transfer of exfoliated flakes. In 2004 ref. 992 placed SWNTs onto arbitrary substrates by transfer printing using poly(dimethysiloxane) (PDMS) stamps. Ref. 993 reported transfer of various nanostructures (such as SWNTs, ZnO nanowires, gold nanosheets and polystyrene nanospheres) by a poly(methyl methacrylate) (PMMA)-mediated process. In 2008 ref. 994 adapted this process to transfer MC graphene on various target substrates. The process is based on a PMMA sacrificial layer spin-coated on graphene. The polymer-coated sample is then immersed in a NaOH solution, which partially etches the SiO2 releasing the polymer. Graphene sticks to the polymer, and can be transferred. PMMA is then dissolved by acetone, releasing graphene.

Ref. 995 reported the deterministic placement of graphene by exploiting a thin layer of water between the PMMA/graphene foil and the substrate. Ref. 996 reported transfer of nanostructures (including graphene) embedded in a hydrophobic polymer. Also in this case, intercalation of water at the polymer–substrate interface was used to detach the polymer/nanostructures film, then moved on a target substrate.402

PMMA is a positive resist widely used for high resolution e-beam lithography.403 By patterning PMMA, it is also possible to remove unwanted graphitic material surrounding MC-SLGs, while shaping and isolating the flakes of interest.

4.10.5. Dry transfer of exfoliated flakes. In order to fabricate heterostructures with clean interfaces (i.e. without trapped adsorbates), dry transfer methods have been developed. Ref. 908 reported a mechanical transfer based on stacking two polymer layers, the bottom being water dissolvable and the top being PMMA. Graphene was exfoliated onto this polymer stack and the sample floated on the surface of deionized (DI) water, resulting in the detachment of the PMMA + graphene film from the substrate. The upper graphene face was not in contact with water, thus minimizing contamination. The polymer + graphene film was then collected and the alignment achieved using a micromanipulator. Ref. 99 used a similar technique to encapsulate graphene between two h-BN layers, while ref. 997 reported an alternative technique based on a glass/tape/copolymer stack. However, ref. 998 reported that even dry transfer may not result in perfectly clean interfaces, as some adsorbates may get trapped.

The Van der Waals interaction between 2d materials was used to assemble heterostructures such as BN-SLG-BN.999Fig. 50 shows a schematic of the process where an isolated few layer BN flake is used as a stamp to successively pick up alternating layers of monolayer graphene and few-layer BN.999 An advantage of this technique compared with other transfer methods is that the active interfaces do not contact any polymer throughout the process, reducing impurities trapped between the layers.999

image file: c4nr01600a-f50.tif
Fig. 50 Layer by layer assembly process. (a) Schematic diagram showing the steps to make a BN-SLG-BN stack or a multiple-layer stack with device layer (graphene) never exposed to any polymer or solvent. (b) Optical images showing in sequence a BN flake on poly-propylene carbonate (PPC) and film picking. Adapted from ref. 999.

A big challenge is the development of a dry transfer technology for ultrahigh quality graphene up to 450 mm, avoiding the wet conditions with polymer coating, which suffer from polymer contamination. The development of dry processes for large area graphene on insulator would be extremely useful for long-term sustainability in device engineering.

Optical trapping can also be exploited to manipulate (translate, rotate) and deposit trapped graphene250,669 and 2d crystals on solid substrates in a controlled fashion.

Optical binding, the formation of extended periodic self-organized (optically bound) structures, could be used for patterning and controlled deposition on a substrate over an extended area for parallel nano-lithography. Another promising application of radiation forces is optical stamping lithography,1000 where the repulsive force exerted by a laser beam is exploited to deposit flakes at desired positions on a substrate. Exploiting this, graphene flakes could be optically stamped on substrate using holographic patterns (with a Spatial Light Modulator), and combined with any other nanomaterial.669 This paves the way for microfluidic integration, expanding the applicability for biosensing.

4.10.6. Transfer of graphene grown on metals. In 2009, ref. 824 first reported the transfer of SLG and FLG grown by precipitation on Ni, by depositing a PMMA sacrificial layer and subsequently etching the underlying Ni by an aqueous HCl solution. Ref. 865 transferred films grown by CVD on Cu, etched by iron nitrite. Ref. 865 introduced etching by aqueous FeCl3 to remove Ni without hydrogen bubbles, which may damage graphene. It also reported a technique where PDMS stamps are attached directly to the graphene surface. Ni is then chemically etched by FeCl3 leaving graphene attached to the PDMS. Graphene is then transferred to SiO2 by pressing and peeling the PDMS. Ref. 7 introduced R2R transfer of graphene grown by CVD on Cu foils as large as 30 × 30 in2: a thermal release tape was attached to the Cu + graphene foil, and then an etchant removed Cu. The tape + graphene film was then attached to a (flexible) target substrate (e.g. PET) and the supporting tape removed by heating, thus releasing graphene onto the target substrate.

To avoid Fe contamination caused by FeCl3 etching, ammonium persulfate [(NH4)2S2O8] was used.1001 To avoid mechanically defects caused by R2R transfer, a hot pressing process was developed:1002 similar to a R2R process, the Cu + graphene foil is first attached to thermal release tape and then Cu is chemically etched. The tape + graphene foil is then placed on the target substrate and they are inserted between two hot metal plates with controlled temperature and pressure. This results in the detachment of the adhesive tape with very low frictional stress, therefore less defects, than a R2R process.1002

4.10.7. Di-electrophoresis. Electrophoresis is a technique used for separating particles according to their size and electrical charge.1003 An uniform electric current is passed through a medium that contains the particles.1003 Particles travel through a medium at a different rate, depending on their electrical charge and size. Separation occurs based on these differences.1003 Di-electrophoresis (DEP) is the migration of uncharged particles towards the position of maximum field strength in a non-uniform electric field.1004 The force in DEP depends on the electrical properties of the particle and surrounding fluid, the particle geometry, and electric field frequency.1003 Particles move toward the regions of high electric field strength (positive DEP) if their polarizability is greater than the suspending medium,1003 whereas they move in the opposite direction (negative DEP) if the polarizability is less than that the suspending medium.1003 This allows fields of a particular frequency to manipulate particles,1003 at the same time assembling them on pre-defined locations.1003

In 2003 ref. 1005 reported large area deposition of SWNTs between electrode pairs by DEP.410 Subsequently, DEP was used for the separation of metallic (m-SWNTs) and semiconducting single wall nanotubes (s-SWNTs),1006 exploiting their dielectric constants difference, resulting in opposite movement of m and s-SWNTs.1006 These SWNT processes were then adapted for graphene. Ref. 1007, 1008 used DEP for the manipulation of GO soot, and single and few-layer GO flakes. In 2009 ref. 1009 placed individual FLGs between pre-patterned electrodes via DEP. Once trapped, the higher polarizability of graphene compared to the surrounding medium1009 limits the deposition to one flake per device.1009,1010 Self-limiting is one of the advantages of this method, together with the direct assembly of individual flakes at predetermined locations.

4.10.8. Applications and processing of graphene inks. Dispersions or inks can be used in a variety of placement methods, including vacuum filtration, spin and spray coating, ink-jet printing and various R2R processes. The latter are most attractive because of their manufacturing characteristics, with transfer speeds in excess of 5 m s−1 currently used in a variety of applications.1011 R2R consists in processing and printing a rapidly moving substrate.1011,1012 Generally, a flexible substrate (e.g. paper, textile, polymer) is unrolled from a source roller, coated (i.e. without patterning) or printed (i.e. with patterning), with one or more evaporated materials (e.g. dielectrics) or liquid inks (e.g. inks containing polymers or nanoparticles), simultaneously or in sequence, and treated/cured while the substrate continuously moves along the coating/printing roller, before being rolled up again, or cut into individual pieces/devices. Unlike assembly style “pick and place” strategies, the continuous fabrication process makes R2R a cheap technology,1012 ideal for high throughput coating, printing and packaging. R2R is a focus of research in plastic electronics, because of its high throughout, and low cost compared to other approaches (e.g. conventional vacuum deposition and lithography) with similar resolution.1013,1014 A standard R2R process may include evaporation, plasma etching, spray or rod-coating, gravure, flexographic, screen or inkjet printing and laser patterning.1011 In many R2R processes, e.g., rod-coating or flexographic printing, solution processing of the ink or material (e.g. polymer, nanoparticles) is required, especially when they cannot1012 be evaporated at low T.1011,1012,1015

Rod-coating employs a wire-wound bar, also known as Mayer bar (invented by Charles W. Mayer, who also founded the Mayer Coating Machines Company in 1905 in Rochester, USA).1012 This is a stainless steel rod wound with a tight wire spiral, also made of stainless steel. During coating, this creates a thin (∼tens μms) ink layer on a substrate.1013 Spray coating forms aerosols of the liquid ink, resulting in uniform thin (∼μm) films on a substrate.1012 Screen printing, on the other hand, uses a plate or screen containing the pattern to be printed on the substrate.1012 The screen is then placed onto the target substrate, while the ink is spread across the screen using a blade, thus transferring the pattern.1012 Flexo- and gravure1015 printing also use a plate to transfer images onto target substrates. Flexo uses a relief plate, usually made of flexible polymeric material, where the raised sections are coated with ink, then transferred onto the substrate by contact printing.1012 Gravure uses an engraved metallic plate, consisting of dots representing pixels.1012 The physical volume of the engraved dots defines the amount of ink stored in them,1016 thus can be used to create gray-scale patterns/images.1016 In general, different viscosities are preferred for different R2R techniques, ranging from 1 to 10[thin space (1/6-em)]000 mPa s or above,1012,1016. Rod- or spray-coating form uniform films, that may be used for larger scale devices (>several cm), the fabrication of TCs, or devices such as batteries or supercapacitors. Screen (∼50–100 μm resolution1012), flexographic (∼40 μm resolution1012) and gravure (∼15 μm resolution1012) printing can be used to print different materials with specific patterns for flexible electronics.1016 For resolutions down to ∼50 μm, inkjet printing offers a mask-less, inexpensive and scalable low-T process.1017 The resolution can be significantly enhanced (<500 nm) by pre-patterning,1017 so that the functionalized patterns can act as barriers for the deposited droplets.1017 The volume can be reduced to atto-liters/drop by pyroelectrodynamic printing.1018 The process is based on the control of local pyroelectric forces, activated by scanning a hot tip or a laser beam over a functionalized substrate (e.g. lithium niobate1018), which draw liquid droplets from the reservoir and deposit them on the underside of the substrate.1018

All of the above techniques can be applied to graphene inks/dispersions. Large scale placement of LPE graphene can be achieved via vacuum filtration,35 spin706 and dip coating,1788 Langmuir–Blodgett662 and spray coating.622 Amongst the R2R techniques, rod-coating has been demonstrated to fabricate TCs.995 Inkjet printing of pristine graphene dispersions was also demonstrated.63 Inkjet printing1019 permits selective deposition and high concentration for partially soluble compounds.1020Ref. 629 reported an inkjet printed graphene TFTs with μ up to ∼90 cm2 V−1 s−1 and Tr ∼ 80%. Inkjet printing of GO was also demonstrated,1021–1024 To minimize clustering of the graphene flakes at the nozzle edge, the flakes should be smaller than 1/50 of the nozzle diameter.629

Inkjet-printing was demonstrated a viable technique to obtain high conductivity graphene patterns.1025 The ink was prepared exploiting ethanol as solvent and ethyl cellulose as stabilizer.1025 The inkjet-printed graphene features had low resistivity of 4 mΩ cm, with uniform morphology, compatibility with flexible substrates, and tolerance to bending stresses.

4.11. Contamination and cleaning

Cleaning is a critical part of semiconductor device processing.1026 It is usually performed after patterning and etching processes leave residues.1026 Wet chemical etches are also performed to remove damage from surfaces.1026 Most applications require graphene on a dielectric surface. When graphene is grown directly on a dielectric as in the case of graphene on SiC [see section 4.2] or when graphene or GO is deposited on the dielectric substrate directly [see section 4.12.3], cleaning is required only after patterning and etch processes, as devices are fabricated. Because every atom is a surface atom, graphene is very sensitive to contaminants left by production, transfer or fabrication processes. In order to remove them, several methods have been developed.

4.11.1. Cleaning of graphene produced by MC

The amount of contamination can be assessed optically.1027 Organic contamination arising from the diffusion of tape glue used in MC changes the contrast.1027 TEM and scanning probe,1028,1029 microscopy (e.g. AFM, STM), Raman185,1030 together with transport measurements1030 are other viable techniques to detect contaminants on graphene films or flakes. Ref. 1028 cleaned MC samples from resist residuals by thermal annealing (at 400 °C, in Ar/H2), assessing the quality of the cleaning process via scanning probe techniques. Ref. 1029 introduced thermal annealing (at 280 °C) in ultra-high vacuum (<1.5 × 10−10 Torr), to remove resist residues and other contaminants. Ref. 1028 cleaned graphene by using high current (∼108 A cm−2). This allows removal of contamination in situ, and is particularly useful when graphene devices are measured in a cryostat.103 Chemical cleaning by chloroform was reported in ref. 1030. Mechanical cleaning by scanning the graphene surface with an AFM tip in contact mode was also reported.1031
4.11.2. Cleaning after transfer. Cleaning is particularly important when transferring flakes, as the processes typically involves sacrificial layers, to be chemically dissolved. Thermal annealing in H2/Ar is normally used to remove polymer residuals.99,998,1032

In graphene transfer from metals to dielectric surfaces, organic materials such as PMMA or perylene-3,4,9,10-tetracarboxylic dianhydride (PTCDA) are typically used as the carrier material, with subsequent chemical removal, e.g. by acetone.6,865Ref. 1033, 1034 detected by XPS the presence of residue on the surface of graphene grown on Cu and transferred onto SiO2. The C1s spectrum was found broader than that of graphite and the original graphene on Cu. The broadening was associated with the presence of the residue. Upon annealing in high vacuum (10−9 mbar) at T ∼ 300 °C, the C1s width decreased to a value close to the original graphene on Cu.1033,1034 The use of thermal release tape7 (i.e. a tape adhesive at RT, but that peels off when heated), instead of PMMA or PTCDA, is more problematic, since tape residues can contaminate the sample.7 There is some anecdotal evidence that the presence of residue has a beneficial effect on the nucleation of ALD dielectrics, such as Al2O3.1035 However, this approach to prepare the graphene surface for ALD is not ideal, since the residues have uncontrolled chemical nature and are not uniform. Ref. 1036 developed a modified RCA transfer method combining an effective metal cleaning process with control of the hydrophilicity of the target substrates. RCA stands for Radio Corporation of America, the company that first developed a set of wafer cleaning steps in the semiconductor industry.1026Ref. 1036 demonstrated that RCA offers a better control both on contamination and crack formation with respect to the aforementioned approaches.6,7,865

4.11.3. Removal of solvents/surfactants in LPE graphene. For graphene and GO produced via LPE, the cleaning, removal of solvents and/or surfactants, mainly depends on the target applications. For composites (both for mechanical134 and photonic625,1339,1353 applications) the presence of surfactants does not compromise the mechanical and optical properties, thus their removal is not needed, and is in fact essential to avoid agglomeration.625,703,1339,1353 Different is the situation when the applications require high conductivity (>104 S cm−1), i.e. TCFs. In this case, solvents/surfactants compromise the interflake connections, decreasing the conductivity. The solvents and the deposition strategy used for the TCFs production mostly determine the cleaning procedure. In the case of TCFs produced by vacuum filtration (e.g. on a cellulose filter membrane) of surfactant-assisted aqueous dispersions, the as-deposited graphene or RGO films are first rinsed with water to wash out the surfactants624,642 and then transferred to the target substrate. The membrane is then usually dissolved in acetone and methanol.1037 For freestanding films, the deposited flakes are peeled off from the membrane.624 The films are then annealed at T > 250 °C in Ar/N2[thin space (1/6-em)]624 or air.642 The latter process could help remove residual surfactant molecules.624,642 However, there is no “fixed” T for solvents/surfactants removal, and the different conditions/requirements are ruled by the boiling/melting points of each solvent/surfactant.

4.12. Inorganic layered compounds

4.12.1. Mechanical cleavage. As with graphene,5 individual inorganic sheets can be removed from their parent crystal by MC5,421,1038 MC can involve a single crystal,603 or a single grain,603 in the case of polycrystalline materials.1039 The local scale dynamics of the fracture process is complex603 and depends on the crystal structure.603 To date the lateral size of 2d crystals produced via MC is ∼10 μm in h-BN,1040 limited by the average crystal size of the starting material.1040 Similar size flakes (∼10 μm) were also achieved via MC of MoS2, WS2 and NbSe2.1041

This allowed the structural characterisation of BN by high resolution TEM1038 and its use as a substrate for high performance graphene devices.908 Similarly, for MoS2 a number of advances have been demonstrated including the production of sensors,1042 transistors5,379,1043 and integrated circuits,1044 the measurement of the mechanical properties of individual nanosheets409 and the observation of the evolution of the vibrational421 and electronic structure,378 with number of stacked flakes.

As in the case of MC of graphite, MC of LMs is not industrially scalable, and MC-flakes are mostly suited for fundamental studies and proof of principle devices.

4.12.2. Laser ablation. Ref. 1045 used laser pulses to ablate MoS2 down to a single-layer. Ref. 1045 generated 1L-MoS2 in arbitrary shapes and patterns with feature sizes down to 200 nm, with electronic and optical properties comparable to MC-1L-MoS2.1041Ref. 1045 reported similar PL emission between MC-1L-MoS2 and laser thinned 1L-MoS2, and μ up to 0.49 cm2 V−1 s−1 and up to 0.85 cm2 V−1 s−1 for laser thinned and MC-1L-MoS2, respectively.
4.12.3. Liquid phase exfoliation. BN,38,1046–1050 TMDs38,396,1051,1052 as well as ternary carbides and nitrides,1053,1054 can be exfoliated in liquids (solvents or aqueous surfactant solutions) by ultrasonication. The exfoliated sheets can then be stabilised against re-aggregation either by interaction with the solvent,38 or through electrostatic repulsion due to the adsorption of surfactant molecules652,1055 In the case of solvent stabilisation, good solvents are those with surface energy matching that of the exfoliated materials.38 This results in the enthalpy of mixing being very small.38,1052 Because these exfoliation methods are based on VdWs interactions between the flakes and either the solvent molecules or surfactant tail group, stabilisation does not result in any significant perturbation of the flake properties. These dispersions can easily be formed into films or composites38 and facilitate processing for a wide range of applications.

The exfoliation and dispersion of LMs were also carried out by the exploitation of co-solvents where the dispersibility of LMs can be greatly improved by using a mixture of solvents1051,1056e.g., water–ethanol,1051,1056 water–isopropyl alcohol,1056etc. By adjusting the relative concentration of the co-solvents it is possible to tune the rheological properties1057 of the mixture “on demand”. However, YM and concentration of the exfoliation process in such co-solvent mixtures is, up to date,1051,1056 much lower than in NMP629 and water-surfactant dispersions.250 Exploitation of co-solvent mixtures,1051,1056,1057 mostly based on water and alcohols, for the dispersion and exfoliation of LMs has some practical disadvantages.1058 The surface tension changes exponentially after the addition of alcohols to water,1057 thus being very sensitive to solvent evaporation.1056 The rheological properties of alcohol-based co-solvents are very temperature sensitive.1057 This is a problem both during processing (the ultrasonication causes a temperature increase of the dispersion) and for the shelf-life (i.e. the maximum time for which the inks can be stored without alteration of their properties) of the dispersions/inks.

Much work remains to be done: exfoliation techniques must be extended to a wider range of materials. Both solvent38 and surfactant-exfoliated396 TMDs tend to exist as multilayer stacks with few individual sheets, thus requiring an improvement of the exfoliation. The dispersed concentrations (up to tens of grams per litres) and the lateral flake size (up to mm) still need to be increased considerably.

In order to pursue these requirements, the development of a sorting strategy that may allow to control both lateral dimensions and N is necessary for the full exploitation of their optical and electronic properties.

Quantitative analysis for monitoring the exfoliation (yield and quality of the as-produced material) of TMDs and TMOs is based on a range of techniques such as AFM, TEM, Raman spectroscopy, etc.2337N can also be controlled via separation in centrifugal fields or by combination with DGU.602,642

The availability of dispersions opens up a range of applications in composites, thin films and inks that can be printed in a variety of ways, and mixed to create hybrids. Many applications in photonics and optoelectronics, such as TCs, third generation solar cell electrodes, and optical-grade composites benefit from LPE produced and assembled materials.

LPE can also produce ribbons with widths <10 nm,303 allowing a further in-plane confinement of the 2d crystals, thus an extra handle to tailor their properties. LPE does not require transfer techniques and the resulting material can be deposited on different substrates (rigid and flexible) following different strategies, such as dip and drop casting, spin, spray and rod coating, ink-jet printing, etc. Several LMs (including BN, MoS2, WS2, MoSe2, MoTe2, TaSe2, NbSe2, NiTe2, and Bi2Te3) have been already successfully exfoliated using this simple, yet efficient, method38 that potentially brings also a benefit in terms of chemical stability, as the layer of liquid might protect the crystallites from oxidation. Such suspensions allow, lastly, easy assembling of the materials into superstructures.

4.12.4. Synthesis by thin film techniques. A number of thin film processes can be brought to bear on the growth of 2d crystals. These range from PVD (e.g. sputtering), evaporation, vapour phase epitaxy, liquid phase epitaxy, chemical vapour epitaxy, MBE, ALE, and many more, including plasma assisted processes. The selection of the growth process depends on the materials properties needed and the application. Each material has its own challenges. Other than controlling the thickness and orientation of the films, the composition and stoichiometry is of utmost importance because this has large influence on transport. As a result, great care must be taken in controlling the point defects concentration. Low growth T (∼300 °C) techniques are usually better suited in controlling the defects arising from vacancies, since the vapour pressure of the chalcogenide elements decreases exponentially with T.1059 However, low T techniques tend to give higher extended defect densities because of the lower atomic mobility.1060 Therefore, the growth technique must be selected to match the desired target application.

To date, WS2 films have been deposited by magnetron sputtering from both WS1061 and WS2 targets,1062 sulfurization of W1063 or WO2 films,1064 ion beam mixing,1065etc. The preferred production process for tribological applications is magnetron sputtering,1066 because of its lower T than thermally activated deposition methods.1066 This is also well suited for large area deposition (∼m2).1067 CVD was used to grow h-BN,1068 and is now being developed to grow TMDs.1069 If single layers of the binary films are desired, then ALD or, more appropriately, ALE might be a better suited.

4.13. Graphene and other 2d crystal hybrids

Technological progress is determined, to a great extent, by developments in material science. The most surprising breakthroughs are attained when a new type of material, or new combinations of known materials, with different dimensionality and functionality, are created. Well-known examples are the transition from 3d semiconducting structures based on Ge and Si to 2d semiconducting heterostructures, nowadays the leading platform for microelectronics. Ultimately, the limits and boundaries of certain applications are given by the very properties of the materials naturally available to us. Thus, the band-gap of Si dictates the voltages used in computers, and the Young's modulus of steel determines the size of the construction beams. Heterostructures based on 2d crystals will decouple the performance of particular devices from the properties of naturally available materials. 2d crystals have a number of exciting properties, often unique and very different from those of their 3d counterparts. However, it is the combinations of such 2d crystals in 3d stacks that offer vast opportunities in designing the functionalities of such heterostructures. One can combine conductive, insulating, probably superconducting and magnetic 2d crystals in one stack with atomic precision, fine-tuning the performance of the resulting material. Furthermore, the functionality of such stacks is “embedded” in the design of such heterostructures.

Heterostructures have already played a crucial role in technology, giving us semiconductor lasers and high μ FETs. However, thus far the choice of materials has been limited to those which can be grown (typically by MBE) one on top of another, thus limiting the types of structures which can be prepared. Instead, 2d crystals of very different nature can be combined in one stack with atomic precision, offering unprecedented control on the properties and functionalities of the resulting 2d-based heterostructures. 2d crystals with very different properties can be combined in one 3d structure, producing novel, multi-functional materials. Most importantly, the functionality of such heterostructures will not simply be given by the combined properties of the individual layers. Interactions and transport between the layers allow one to go beyond simple incremental improvements in performance and create a truly “quantum leap” in functionality. By carefully choosing and arranging the individual components one can tune the parameters, creating materials with tailored properties, or “materials on demand”. Following this novel approach, part of the functionality is brought to the level of the design of the material itself.

Inorganic LMs can be exploited for the realization of heterostructures with graphene, to modulate/change the electronic properties, thus creating “materials on demand”: hybrid superstructures, with properties not existing in nature,99,106,1070,1078 tailored for novel applications. E.g., superstructures like those in Fig. 31 (SLG/BN/MoS2/BN/SLG) can be used for tunnel devices, such as diodes, FETs, and light emitting devices, or for energy application, such as photovoltaic cells.

To date, 3 methods can be envisaged for the production of atomically thin heterostructures: (I) growth by CVD;1071 (II) layer by layer stacking via mechanical transfer,99,908,1072 and (III) layer by layer deposition of chemically exfoliated 2d crystals. However, as the field develops, other techniques will emerge.

CVD is emerging as a promising approach to grow 2d crystals. Growth of h-BN1073 and MoS2[thin space (1/6-em)]1074 was already demonstrated. Ref. 1073 reported growth of single layer h-BN by thermal decomposition of borazine (B3N3H6) on Ni (111). Ref. 1074 demonstrated that the CVD growth of MoS2 is scalable and films of any size can be made because the lateral size of the layers is defined by the size of the substrates used. Moreover, also the thickness of the MoS2 film can be controlled being directly dependent on the thickness of the pre-deposited Mo metal on the substrate and the as-grown layers can then be transferred onto arbitrary substrates.1074

The aim is to produce a large range of 2d crystals on large scale and with control on demand of N. Low-T CVD and CMOS compatible substrate will be investigated as well all the transfer strategies already developed with graphene.

2d crystals can also be produced as nanoribbons (NRs) with tuneable electrical and magnetic properties. MoS2-NRs were made via electrochemical/chemical synthesis,1075 while zigzag few- and single-layered BN-NRs were obtained unzipping multiwall BN nanotubes through plasma etching.1076 The target is to produce, within the next 10 years, NRs with controlled electrical and optical properties.

Field effect vertical tunnelling transistors based on graphene heterostructures with atomically thin BN acting as a tunnel barrier, were reported.489 The device operation relies on the voltage tunability of the tunnel density of states in graphene and of the effective height of the tunnel barrier adjacent to the graphene electrode.489Ref. 491 used WS2 as an atomically thin barrier, allowing switching between tunnelling and thermionic transport, with much better transistor characteristics with respect to the MoS2 analogue,489 thus reaching a much higher ION/IOFF ∼ 106. A “barristor”, i.e. a graphene-Si hybrid three-terminal device that mimics a triode operation, was developed by ref. 492. The electrostatically gated graphene/Si interface induces a tunable Schottky barrier that controls charge transport across a vertically stacked structure.489,491,492

4.13.1. CVD growth of heterostructures. Chemical vapour deposition is a method suitable for mass production of heterostructures, though it requires the largest investment and effort in terms of developing the necessary expertise and machinery. There are several indications that such growth is indeed feasible:1077 H-BN has already been shown to be effective as a substrate for graphene CVD, see Fig. 51.
image file: c4nr01600a-f51.tif
Fig. 51 Schematic of Graphene/BN stacked film. (a) The Cu foil is cleaned and prepared as the growth substrate. (b) High-quality and large-area graphene film is grown via CVD with n-hexane as a liquid carbon source at 950 °C. (c) The as-grown graphene is then loaded into another furnace for the growth of h-BN film on top. Adapted from ref. 1077.
4.13.2. Mechanical transfer. Transfer of individual 2d crystals into heterostructures has been widely demonstrated99,1078 Graphene – BN heterosystems, e.g., have enabled the observation of several interesting effects, including FQHE,908 ballistic transport99 and metal–insulator transition in grapheme.1078 ‘Dry’ mechanical transfer offers the possibility of controlling/modifying each individual layer as it is being deposited, including by chemical modification, at any stage of the transfer procedure. Also, any atomic layer in the multilayer stack can be individually contacted, offering unprecedented control on the properties of the stack (one could produce a material with individual contacts to every conducting atomic plane). Furthermore, one can apply local strain to individual layers, which significantly modifies the band structure of GRMs.19,1079,1080 Important is the control of the relative orientation of the layers, which may affect the electronic properties of the stack in certain intervals of the energy spectrum.958
4.13.3. Heterostructures from dispersions and inks. Large-scale placement of LPE samples can be achieved by spin coating, Langmuir–Blodgett, spray and rod coating, dip casting (Fig. 52). Surface modifications by SAMs enable targeted large-scale deposition. High uniformity and well defined structures on flexible substrates can also be obtained. DEP can also be used to control the placement of individual crystals between pre-patterned electrodes. Inkjet printing allows to mix and print layers of different materials and is a quick and effective way of mass-production of such systems.
image file: c4nr01600a-f52.tif
Fig. 52 Superstructure assembly by dispersions and inks. Dip casting is used to deposit successive layers of different 2d crystals to create an heterostructure.

Although the quality of the resulting structures would be significantly lower than that obtained by mechanical or CVD methods, it would still be suitable for a number of photonics and optoelectronics applications, as well as for applications in thin film transistors, RF tags, solar cells, batteries and supercapacitors.

4.13.4. Bonding using polymers. 2d polymers offer large structural diversity through different possible connections between the monomers. Different synthesis routes are possible: using small fragments covalently bonded (difficult), self-assembly approaches (maximum sizes of the order of a few nanometers), knitting polymeric strands (planarity difficult to achieve), polymerization at air/liquid or liquid/liquid interfaces, SAMs, stacked (3d) multilayer polymerization, etc.

4.14. Silicene, germanene, phosphorene, MXene and other graphene like systems

Silicene sheets,113,114i.e. the Si equivalent of graphene,1081 see Fig. 53a, have been synthesized by in situ growth on silver (Ag) (111) surfaces. A honeycomb atomic structure with a Si–Si distance of 0.23 nm was revealed in STM,113,1081 with a long-range epitaxial order confirmed by sharp LEED patterns. Conical band dispersions at the corners of the silicene Brillouin zone (K and K′ points), evidenced in High-Resolution ARPES measurements, point to Dirac fermions with a Fermi velocity of 1.3 × 106 ms−1, as theoretically predicted,1082 quite the same as graphene, and four times higher than previously obtained on a 1d grating of silicene NRs.1083 GGA-DFT calculations including the Ag(111) substrate-confirm the stability of the epitaxial arrangement.
image file: c4nr01600a-f53.tif
Fig. 53 (a) STM image showing the graphene-like structure of one monolayer of silicon deposited on a closepacked silver surface, Ag (111). Adapted from ref. 114 (b) 16.2 nm × 16.2 nm STM image of the modulated honeycomb of germanene √7 × √7 superstructure with a zoom-in at the bottom left corner (the √7 × √7 unit cell is drawn in black). Adapted from ref. 1099; (c) Side view of MXene, where M (green dot) is an early transition metal, X is either carbon and/or nitrogen (grey dots) and A is an A-group (mostly IIIA and IVA, or groups 13 and 14) element. Adapted from ref. 1100.

Silicene is predicted to have non-trivial topological properties.115 Hence, it could offer the possibility, if interfaced with a s-wave superconductor, for advances in the long quest for Majorana fermions.270 Furthermore, being Si the workhorse of electronics, this synthesis could have a impact for novel devices because of the compatibility with existing Si technologies. A key issue in this direction is the transfer -or even the growth-on an insulating substrate, like, e.g., AlN.1082

The growth of silicene paves the way to the synthesis of germanene, see Fig. 53b, the equivalent of graphene for Ge, also with nontrivial band topology, and a gap induced by effective spin orbit coupling for the π orbitals at the K point.270,115

Stanene, a single layer of tin atoms,123 arranged in a manner similar to graphene, was observed in MBE experiments in the nineties.117 Stanene deposited on substrate has a gap ∼0.3 eV,123 with quantum spin Hall (QSH) states (i.e. edge states where the carriers with opposite spins move in opposite directions1084) states that can be tuned by chemical functionalization as well as by external strain.123

Phosphorene is a stable elemental 2d material that can also be mechanically exfoliated from bulk black phosphorus,116 which is the most RT stable phosphorus allotrope.1085 Black phosphorus was first synthesized from white phosphorus (also known as tetraphosphorus, a molecule made up of four atoms in a tetrahedral structure) under high pressure and high temperature in 19141086 and it has a layered structure, held together by vdW forces.1087–1089 Black phosphorus displays a sequence of structural phase transformations, superconductivity at high pressures with Tc above 10 K, and T dependent resistivity and magnetoresistivity.1090–1095 Phosphorene layers share a honeycomb lattice structure with graphene with the notable difference of non-planarity in the shape of structural ridges. The bulk lattice parameters of black phosphorus are: a1 = 3.36 Å, a2 = 4.53 Å, and a3 = 11.17 Å (ref. 116). The large value of a3 is caused by the nonplanar layer structure and the presence of two AB stacked layers in the bulk unit cell, see Fig. 26. The ridged layer structure helps to keep orientational order between adjacent phosphorene monolayers and thus maintains the in-plane anisotropy.116

Another class of 2d crystals are the MXenes, i.e. transition metal carbides and nitrides, see Fig. 53c, created by selectively removing Al (in hydrofluoric acid at RT)119 from layered ternary carbides known as MAX phases. These could add at least another 60 members to the LM family.119,120 Through this exfoliation process, the carbide layers were separated into two MXene sheets just a few atoms thick.119 MXenes can accommodate various ions and molecules between their layers by intercalation.1096,1097 MXene sheets are promising for energy applications, such as Li-ion batteries,1096 electrochemical capacitors120,1097 and hydrogen storage.1098Ref. 1097 reported the spontaneous intercalation of cations from aqueous salt solutions between 2d Ti3C2 MXene layers. A variety of cations, including Na+, K+, NH4+, Mg2+, and Al3+, can also be intercalated electrochemically, offering capacitance in excess of 300 F cm−3, a value that is much higher than that achieved by porous carbons1097 and comparable with those reported in other GRMs, see section 10.2.

4.14.1. Chemical modification of 2d crystals. Chemical modification of graphene (Fig. 54) is a powerful approach to create new GRMs (e.g., graphane,33 chlorinated graphene1101 or fluorographene34). Applying this to other 2d crystals offers a chance to have new varieties with versatile physical properties. It is only the beginning, and many other materials with very different properties are possible. Methods for chemical modification of the two sides of a 2d crystal are desirable. Chemical modifications can provide a fine control over the distance between neighbouring planes in the 2d heterostructures. This might be done by intercalation, or by placing other metallic and semiconducting nanostructures between planes.
image file: c4nr01600a-f54.tif
Fig. 54 Chemically modified graphene. One can add different species (e.g. hydrogen, purple dots or fluorine, green dots) to the graphene scaffolding.

Ref. 122 synthesized millimeter-scale crystals of a hydrogen-terminated germanane from the topochemical deintercalation of CaGe2. This layered solid is analogous to multilayer graphane.33 The resultant four-coordinate puckered lattice of Ge atoms has an analogous geometry to the sp3-hybridized graphane surface, in which every Ge atom is terminated with either H or OH above or below the layer.124 The surface layer of germanane only slowly oxidizes in air, after 5 months.122 Germanane is thermally stable up to 75 °C (ref. 122). However, above this T, dehydrogenation and amorphization begin to occur.122 Germanane is the first of a new class of covalently terminated graphane33 analogues, that could cover a variety of traditional semiconductors, and has potential for a wide range of (opto)electronic and sensing applications, because of its direct band gap of 1.53 eV (ref. 122) and an electron μ ∼5 times higher than that of bulk Ge (3900 cm2 V−1 s−1), as predicted by theory.122

4.15. Outlook and future challenges

The successful use of GRMs depends not only on the identification of the right products for new and current applications, but also on the ability to produce any of the materials in large quantities at a reasonable cost. We now know over 600 LMs exist, and probably more. The progress in developing new materials processes over the past few years has been impressive.

However, the suitability of any given process depends on the application. Nanoelectronics more than likely has the most demanding requirements, i.e. low defect density single crystals. Other applications, such as biosensors, may require defective graphene, while printable electronics can tolerate lower quality, e.g. lower μ, graphene. CVD techniques are emerging as ideal for large area graphene films for touch screen displays and other large display applications, while graphene derived from SiC maybe better suited for resistor standards. Many issues still remain to be addressed in the growth of graphene by CVD to improve the electrical and optical characteristics, including mechanical distortions, stable doping, and the development of reliable low cost transfer techniques. While transfer techniques can be developed to place graphene onto insulating substrates, it is desirables to grow graphene directly on dielectric surfaces for many device applications and progress is being made in achieving films on h-BN as well as SiO2. However, a lot more effort is required to achieve large area uniform high quality graphene films on dielectrics. In the case of graphene on SiC, among other issues related to uniformity, crystal size could be an impediment for large scale production. LPE is appealing for inks, thin films and composites, and future research is needed to control on-demand N, flake thickness and lateral size, as well as rheological properties. Synthetic graphenes, together with GNR produced by SiC,100 are promising for the production of atomically precise NRs and QDs to overcome the lack of band gap necessary for many electronic device applications. A controlled dopant distribution is also needed, and techniques such as functionalization using self-assembled acceptor/donor molecules or assembling pre-doped molecules are being studied.

The layered nature of graphite makes its integration with other LMs a natural way to create heterostructures. LMs have been around for a long time and studied and developed mostly for their tribological properties. Now these materials are being considered as new interlayer dielectrics for heterostructures with potential for new electronic devices with exotic properties. Because of this, there will be a host of new processes that will need to be developed in order to grow or deposit high quality large area monolayer films integrated with graphene with controlled thickness and transport properties.

The GRMs production timeline is shown in Fig. 55. Timescales: 23 years: Formulation of GRM inks. CVD growth high μ (6 × 104 cm2 V−1 s−1) graphene films. Homogeneous (mm2) graphene films on SiC 35 years: Production of heterostructures with on-demand (opto)electronic properties via LPE and CVD. 57 years: Production of 2d crystal by MBE and ALD. Formulation of high concentration inks (10 g L−1) with tuneable morphological (>80% 1-L) and controlled rheological properties (ν = 1–1000 mPa s). 710 years: Production of large area (mm size) 2d crystals via LPE. Large area single crystals. Surface roughness <0.2 nm. Engineered band gap NRs and QDs.

image file: c4nr01600a-f55.tif
Fig. 55 GRM production timeline.

5. Electronic devices

Graphene has already demonstrated high potential to impact most ICT areas, ranging from top-end, high-performance applications in ultrafast (>1 THz) information processing, to consumer applications using transparent or flexible electronic structures.

The great promise of graphene is testified by the increasing number of chip-makers now active in graphene research. Most importantly, graphene is considered to be amongst the candidate materials for post-Si electronics by the ITRS.12

5.1. Opening a band-gap in graphene

The target is to fully explore the performance of graphene transistors in both logic and RF applications. Graphene may be used to develop new applications based on stretchable electronics, such as conformal biosensors and rollable displays. These are required to meet the increasing needs of human-interface technology. Graphene can solve the standstill of stretchable electronics, due to the difficulty in developing semiconducting materials with the high stretchability required for such applications. Fig. 56 and Table 3 show some possible applications and an indication of when functional device prototypes could be expected.
image file: c4nr01600a-f56.tif
Fig. 56 GRM electronics’ application timeline. The rectangles in the figure indicate the timeframe when functional device prototypes could be expected.
Table 3 Drivers leading the implementation of graphene for different electronic applications and issues to be resolved with current graphene technology
Application Drivers Issue to be addressed
Touch screen Better endurance with graphene as compared to other materials Need to better control contact resistance
E-paper High Tr of SLG; Visibility Need to better control contact resistance
Foldable OLED Graphene with high electrical properties and bendability Need to improve the Rs
Efficiency improved due to graphene's work function tunability Need to control contact resistance
Atomically flat surface of graphene helps to avoid electrical shorts and leakage current. Need a conformal coverage of 3d structures
RF Transistor No manufacturable solution for InP high-electron-μ transistors (HEMT) (low noise) after 2021 according to the 2011 ITRS Need to achieve current saturation
f T = 850 GHz, fmax = 1200 GHz should be achieved
Logic Transistor High μ New structures
Need to resolve the band-gap/μ trade-off
Need ION/IOFF >106

Graphene can replace materials in several existing applications, but the combination of its unique properties should inspire completely new applications, which is the ultimate target.

The major obstacle of graphene in transistor applications, especially for ICs as potential Si replacement, is its zero band-gap. This is responsible for the low ION/IOFF in GFETs due to a non-zero off state drain current, which leads to considerable static power dissipation. E.g., the typical static drain current in graphene inverters1102 is ∼270 μA μm−1 at a supply voltage VDD = 2.5 V, in contrast to the much smaller leakage drain current ∼100nA μm−1 at VDD = 0.75 V in 22 nm node high-performance Si logic transistors.1103 Small band gap opening was observed in large area hybrid films, consisting of graphene and h-BN domains synthesized on Cu substrates by CVD.1104

Thus, opening a band gap without compromising any of its other outstanding properties, such as high-field transport and μ, is one of the most active research areas. Apart from quantum confinement (GNRs and GQDs), many other techniques have been developed for this goal. Substrate induced band-gap opening was investigated.1105 Band-gap opening in graphene, both on h-BN and h-BN/Ni(111) with band gap up to 0.5 eV was reported.1106 Theory suggests that a band gap ∼0.52 eV can be opened in graphene deposited on oxygen terminated SiO2 surfaces.1107 A bandgap is observed for BLG grown on SiC.1108

Substitutional doping is another promising route for opening a band gap. Nitrogen doping might be used to convert graphene into a p-type semiconductor.1109

A few other approaches also exist for band gap engineering. Formation of GNRs with finite band gaps is possible using conventional BC lithography.339 A band gap ∼0.7 eV was demonstrated by selective hydrogenation of graphene on Ir.358 Molecular doping and charge transfer methods could also modulate the electronic properties via paramagnetic adsorbates and impurities that can dope graphene.1110 Selective chemical functionalization can also be used for band gap engineering.33,34 Complete hydrogenation of graphene forms graphane, an insulator,33 while a similar process using Fluorine, produces fluorographene.34 The latter is optically transparent with a gap ∼3 eV.34 If an electric field is applied perpendicular to BLG, a gap opens, with size dependent on the strength of the field.127,1111–1113

The organic synthesis of GNRs seems to be a powerful tool.320 However, as reported in section 4.9, a reliable approach for on-demand bandgap engineered GNRs needs to be developed. Band-gap opening in graphene on SiC sparked a lot of interest because of the viability of the growth process. Recently, STS measurements taken on GNRs grown on the terrace structure of SiC have shown a gap of more than 1 eV (ref. 100). However, graphene grown on SiC tends to be electron doped and the Fermi level lies above the gap. To make it viable for electronics requires hole doping, or Fermi level moving by applying a gate voltage.

All these methodologies are at their infancy, and need be further developed. E.g., B substitutional doping, one of the most promising ways of opening a band gap in graphene,971 increases defects and disorder.971 Uniform doping over large areas has not been achieved yet.

Techniques need to be studied to locally functionalize graphene on an atomic length-scale employing a Scanning Catalyst Microscope (SCM).1114 A catalyst particle attached at the end of a scanning tip is positioned close to the sample and then a local chemical reaction is triggered by local heating in the presence of a reaction gas.310E.g., Ni particles preferentially cut graphene along specific crystallographic directions.310 Atomic precision is assured by the limited contact area between tip and sample.

Another aim is to achieve control over domain size and shape in graphene-BN hybrids. This is essential for tuning the gap and other electronic properties. Tuneable band gap and spintronic properties in graphene-graphane superlattices need to be addressed.

5.2. Graphene-based microelectronics and nanoelectronics

The progress in digital logic relies in downscaling of CMOS devices through the demand for low voltage, low power and high performance. This size scaling has permitted the IC complexity to double every 18 months.1115,1116 The decrease of gate lengths corresponds to an increase of the number of transistors per processor. Nowadays, processors containing two billion MOSFETs, many with gate lengths of 30 nm, are in mass production (Fig. 57).433
image file: c4nr01600a-f57.tif
Fig. 57 Evolution of MOSFET gate length integrated circuits (filled red circles). The ITRS targets a gate length of 7.4 nm in 2025 (open red circles). With the decrease of gate lengths, the number of transistors per processor increased (blue stars). New materials, like graphene, are needed to maintain these trends.433

However, CMOS scaling is approaching fundamental limits due to various factors, such as increased power density, leakage currents and production costs, with diminishing performance returns.1115,1116E.g., static (leakage) power dissipation in state-of-the-art Si microprocessors has already exceeded the dynamic (switching) power,1117 and is expected to increase further with the continuation of the aggressive scaling of CMOS technology. Faster computing systems need access to large amounts of on-chip memory and Si technology scaling limits create bottlenecks in realizing high-density memories.1118 Thus, a significant challenge for the semiconductor industry is the development of a post-Si age, with new materials, such as GRMs. However, the potential performance of graphene-based transistors is still unclear. It is not the extremely high μ of graphene, but rather the possibility of making devices with extremely thin channels that is the most forceful feature of GFETs. Indeed, these devices may be scaled to shorter channel lengths and higher speeds, avoiding the undesirable short channel effects that restrict the performance of current devices. With the continuous downscaling of devices and increasing dissipation power densities, materials that can conduct heat efficiently are of paramount importance.102 The outstanding thermal properties102 of graphene provide an extra motivation for its integration with CMOS technology, as well as beyond-CMOS, with the possibility to overcome state-of-the-art Si and III–V semiconductor based high frequency FETs at the ultimate scaling limits.1119

Apart from the high cut-off frequency fT (which is a measure of internal transistor delays, rather than its functionality in realistic electronic circuits1120), GFETs must also exhibit an over-unity voltage gain. Voltage gain Av > 1 is required in general-purpose electronic circuits, such as analogue voltage amplifiers and digital logic gates. The first graphene back-gated FET was reported in 2004.4 However, such back-gated devices, although very useful for proof-of-concept purposes, do not exhibit Av > 1, suffer from very large parasitic capacitances and cannot be integrated with other components. Consequently, practical graphene transistors need a top-gate. The first graphene top-gate GFET was reported in 2007.1121 From that important milestone, huge progresses have followed. Top-gated graphene GFETs have been made with graphene produced by MC159,705,1121–1124 carbon segregation1125 and CVD.1126 Large scale integration of GRMs in modern electronic devices requires development of ultra-thin high-[scr K, script letter K] gate dielectrics, such as SiO2,1121 Al2O3,1127 and HfO2[thin space (1/6-em)].1128 This is especially important in GFETs in which high transconductance is required to obtain Av > 1. Such dielectrics should also exhibit large scale uniformity and high reliability without having a negative impact on the electronic properties of GRMs. High-quality dielectrics are often grown by ALD, requiring a seeding layer which increases the total thickness of the gate insulator and therefore decreases the transconductance. In order to realize ultra-thin dielectrics of Al2O3 and HfO2 an organic seeding monolayer was used,1130–1132 The growth of Al2O3 on an ozone seeding layer which leaves no residual layer after growth was demonstrated.1133 As an alternative to ALD, self-assembled nanodielectrics were incorporated in GFETs.1134 Similar investigations on ALD-grown Al2O3 in MoS2 FETs were also carried out.1135,1136

The first GFETs with Av > 1 (∼6) were realized by utilizing ultra-thin AlOx gate dielectrics.1129,1137 The gate stack was fabricated by evaporation of Al followed by exposure to air. This naturally forms a very thin (<4 nm) AlOx layer at the interface between graphene and the Al layer evaporated on top.1129,1137 However, these GFETs exhibited Av > 1 only at cryogenic T,1129,1137 as strong hysteresis observed in their transfer characteristics suppressed Av at RT.1137 Hysteretic behaviour of GFETs under ambient conditions stems from water charge traps adsorbed on the substrate,1138–1140 which has a detrimental impact on their transconductance, therefore on Av. The first Av > 1 under ambient conditions1141 was obtained by deploying misoriented BLG as active material and using a solid polymer electrolyte as gate dielectric.1141 However, these devices exhibited Av > 1 only in direct current (DC) mode, as a consequence of a large overlap between the polymer gate and source/drain contacts.1141 DC gain is of no interest in realistic electronic applications, as logic gates and voltage amplifiers operate in dynamic, alternating current (AC) mode. An AC Av > 1 was firstly demonstrated at RT in a 6-finger-gate GFET configuration.1142 The obtained gain was small (Av = 1.7) and the devices were not integrated (they required external inductors and capacitors to operate). Despite this, they exhibited promisingly large bandwidth (6 GHz).1142

Soon thereafter integrated graphene voltage amplifiers were demonstrated.1143 They exhibited the highest AC voltage gain (Av = 3.7) reported until then in SLG-FETs at RT.1143 In contrast to standard GFETs in which there are ungated parts of graphene channel on either side of the gate, those in ref. 1143 did not have ungated parts, due to a self-aligned fabrication process. This also eliminated hysteresis in the transfer curve, thus far detrimental in obtaining Av > 1 at RT. Such a unique blend of transistor properties combined with the use of very thin gate insulators resulted in Av that can readily be utilized both in analogue and digital electronics. However, these devices were fabricated from MC graphene, unsuitable for mass production. The highest Av reported so far in self-aligned top-gated wafer-scale GFETs under ambient conditions is 5.3.1144 Such gain allowed graphene integrated complementary inverters to exhibit digital signal matching at RT.1144 Cascading of digital inverters in which the previous stage is capable of triggering the next stage, was also demonstrated.1144 This approach resulted in the realization of integrated graphene ring oscillators (GROs), consisting of an odd number of inverters cascaded in a loop which makes the RO unstable inducing oscillation, operating at GHz frequencies. These GROs are the fastest demonstrated in any low-dimensional nanomaterial to date.1102 The highest Av reported so far in GFETs is 35, but in exfoliated BLG subject to perpendicular electric fields.87

The channels of most top-gated transistors are made of large-area graphene, with a minimum conductivity (∼4 e2 h−1), even within the limit of nominally zero carrier concentration.74 This is too high for applications in logic elements, as it leads to high leakage in the off state and poses serious limitations for the switching of these devices. Thus, one option for graphene in digital logic relies on bandgap opening. To date, the formation of GNRs seems the most promising route and GFETs with back-gate control were demonstrated.312 Such devices operate as p-channels with ION/IOFF ∼ 106 (ref. 312). Ref. 1122 reported top-gated GNR GFETs with HfO2 top-gate dielectric, with a RT ION/IOFF ∼ 70. Proof of principle devices were demonstrated in BLG GFETs, with ION/IOFF ∼ 2000 at low T, and 100 at RT.1145 However, large ION/IOFF must be obtained together with Av > 1 for these GFETs to be usable in practical applications, i.e., large μ must be preserved while opening the band gap.

The high μ, coupled with high κ and high current density, make graphene ideal as a replacement for Cu interconnects.1146 Theoretical projections suggest that graphene with low line-widths (<8 nm) may outperform Cu.1146 Thus, although for digital electronics the entry of graphene is expected on a longer timescale, Fig. 61, the first components, such as interconnects, may be fabricated within the next few years. The long-term target plan (>20 years) is to transform graphene transistors, from being excellent tools to probe the transport properties of this material, to viable devices to compete and replace/integrate state-of-the-art Si and compound semiconductor electronics. Routes for realizing graphene-based digital electronics need to be explored and assessed to fully exploit the potential of this material to bring the semiconductor industry beyond the 7.4 nm node, which the ITRS expects to be reached in 2025 (ref. 12).

GFETs with controlled threshold voltage and both n-channel and p-channel need to be demonstrated for CMOS logic. The contact resistance between the metallic source and drain and graphene channel should be investigated deeply and more focussed research is needed to understand the contact properties. New graphene device concepts, such as tunnel FETs (TFETs) and bilayer pseudospin FETs (BISFET)1147 need to be studied, and different design options must be explored, evaluated and optimized. The BiSFET is based on the electrical properties of two layers of graphene in close proximity.1147 Electrons in one layer can pair with holes (both Fermions) in the opposite layer resulting in e–h-pairs/excitons (Bosons) which then can condense.1147 The condensation alters the quantum wavefunctions in the bilayer structure, converting states that were isolated in one of the two layers into states that are a coherent linear combination of top and bottom layer components.1148 This qualitative change effectively shorts the two layers, reducing the tunnel resistance from a large value to a value limited by contacts only.1147 The reduction in tunnel resistance applies only for small interlayer bias, however, because high current destroys the condensate.1148 The BiSFET exploits the IV nonlinearities associated with this maximum tunnel current, allowing, in principle, lower voltage, lower power operation than possible with CMOS FETs.1147

Moreover, the integration with exiting CMOS technology is a critical step in establishing a pathway for graphene electronics.

Another crucial point concerns the steady increase in power dissipation demand per unit area (despite the reduction of the supply voltage). This is becoming a major issue for the design of next-generation devices: it is mandatory to efficiently remove heat. Besides its practical importance, the investigation of heat transport in graphene and graphene-based systems offers other rewards, more closely related to fundamental physical issues like, e.g., the role of the reduced dimensionality and/or different shaping on transport features.

It is possible, through computer simulations, to understand how/to-what-extent in-plane κ is affected by structural defects, stretching and bending deformations and lateral dimensions (in GNRs). Also, by simulations, proof-of-concept studies can be executed on possible thermal rectification effects in GNRs.

5.2.1. Transistor count in graphene circuits. The development of graphene devices exhibiting Av > 1 is tightly connected to the development of graphene multi-stage (i.e. multi-transistor) circuits. The transistor count in some of the graphene circuits realized so far is shown in Fig. 58. The first functional circuits comprised only one GFET.1149,1150 Their functionality was controlled by DC input biasing, depending on which different types of logic gates1149 or a frequency multiplier1150 were realized. Although these simple circuits demonstrated that graphene can be used to realize functionalities typically found in conventional electronic circuits, they suffered from several drawbacks. Both circuits were made from MC graphene and not integrated (they required external resistors to operate). The GFETs were back-gated, therefore the circuits exhibited a very small Av, which resulted in a large attenuation of the output signal. This lead to the inability to directly couple digital logic gates (due to a mismatch between input and output voltage logic levels) or to amplify analog AC signals. Therefore the realized functionalities could not be used in realistic and more complex electronic systems.
image file: c4nr01600a-f58.tif
Fig. 58 Transistor count in graphene circuits with Av < 1 (open symbols) and Av > 1 (filled symbols). The linear fit is performed only on circuits with over-unity RT AC Av. From this, a graphene Moore's law is obtained: the number of GFETs on a chip currently doubles approximately every 8 months. However, it is unlikely that this will hold in the near future. Values taken from: image file: c4nr01600a-u3.tif,1149image file: c4nr01600a-u4.tif,1150image file: c4nr01600a-u5.tif,1140 ◊,11image file: c4nr01600a-u6.tif,1143 ■,1142image file: c4nr01600a-u7.tif,1151 ●,1141image file: c4nr01600a-u8.tif,1129 ▲,1144image file: c4nr01600a-u9.tif,1155image file: c4nr01600a-u10.tif.1102 Adapted from ref. 1208.

The first attempt to increase the transistor count was the realization of graphene complementary inverters comprised of two GFETs.1151 This was the first time a complete functionality was integrated on a single MC graphene flake, and no additional components were required. This integrated circuit also established a simple concept of complementary operation between the Dirac points of the two GFETs connected in series. However, these inverters were also back-gated, therefore exhibited Av < 1. The concept of complementary operation was later extended to top-gated inverters, exhibiting Av > 1 at cryogenic1129 and RT.1143 However, their transistor count was also 2, and there was no attempt to demonstrate multi-stage circuits, probably because of the use of MC graphene.

The first graphene multi-stage circuit comprised two inverters connected in series, thus bringing the transistor count to 4.1144 Signal matching and cascading was demonstrated under ambient conditions, with inverters fabricated from CVD graphene.1144 A good Av ∼ 5, crucial in achieving cascaded operation,1144 was obtained due to a combination of factors including full-channel gating, thin (∼4 nm) gate oxide (leading to good control of the top-gate over the channel), μ ∼ 500 cm2 V−1 s−1, low output conductance (gd ∼ 50 μS μm−1), and manageable contact resistivity (∼9 kΩ·μm). Although these inverters pave the way to realistic multi-stage circuits, they have large parasitics, limiting the clock rate to 200 kHz.1144

Multi-stage circuits must operate at high (>1 GHz) technologically-relevant frequencies to be useful. High-frequency graphene circuits were developed over the past few years, but they are all single-transistor and single-stage circuits.1153,1154 The first wafer-scale graphene integrated circuit was a 10 GHz frequency mixer, but it consisted of a single GFET integrated with two inductors.11 Similarly, high-frequency graphene voltage amplifiers were reported with Av > 1, up to 6 GHz, but with a single GFET connected to two off-chip bias tees.1142

Elimination of parasitics in inverters integrated on wafer-scale graphene paves the way to high-frequency multi-stage integrated circuits. Ref. 1102 demonstrated high-frequency GROs, consisting of 4 cascaded stages and 8 GFETs. GROs were also realized with 12 GFETs,1155 however running below 50 MHz.

5.2.2. Digital logic gates. Application of graphene in digital logic gates is limited by the zero bandgap, which prevents depletion of charge carriers. The inability to completely turn off GFETs increases static power dissipation with respect to traditional, Si-based, CMOS logic. This also limits the control of gate voltage over drain current, i.e. it reduces the transconductance, gm, with respect to conventional FETs, which can be turned off at suitable gate biases.1156 Moreover, lack of depletion leads to a weaker drain current saturation regime in GFETs, which in turn increases their output conductance gd. Hence most of the GFETs so far have intrinsic gain gm/gd smaller than unity1140,1149,1150,1151,1157–1159 which results in the inability to directly couple digital logic gates (due to a mismatch between input and output voltage logic levels).1141 Current modulation in graphene devices can be increased by patterning GNRs, which increases ION/IOFF296,1152,1160 However this significantly reduces the on current,303,1140 which in turn reduces Av. Similarly, very large ION/IOFF obtained in recently reported GTFETs106 are currently unusable in digital logic due to very low on currents.

GFETs must satisfy two additional requirements in order to be considered as building blocks of future logic gates: Av > 10 and ION/IOFF > 104. The short-term goal is thus to achieve Av sufficient for realistic applications where high-speed operation is desired, but power dissipation is not a concern, similar to the SiGe and InP emitter-coupled logic (ECL), the fastest logic family.1161 The long-term goal should be based on both requirements, as only in this way graphene could be considered as a replacement for Si CMOS in future ubiquitous logic gates (e.g., in microprocessors). This is not a far-fetched goal because the Si CMOS is also experiencing some fundamental difficulties related to short-channel effects, gate oxide thickness scaling, and channel carrier mobility.12,1162 As a consequence, static (leakage) power exceeds dynamic (switching) power in most state-of-the-art Si logic circuits.12,1163

In order to achieve the short-term goal it will be necessary to realize top-gated GFETs with ultra-thin (<4 nm) high-[scr K, script letter K] gate insulators, as Av > 1 has already been demonstrated in graphene devices with similar gate thicknesses (∼4 nm).1141–1143 The voltage gain could further be improved by using BLG. In turbostratic BLG misorientation electronically decouples the two SLGs.1164,1165 The bottom layer acts as a pseudosubstrate, which electrostatically screens the top layer from the substrate, thus giving enhanced μ within the top layer.1164,1165 In Bernal-stacked BLG a perpendicular electric field can open a moderate bandgap128 allowing large Av in dual-gate configurations.128 The next stage in the development of graphene logic should be incorporation of highly-efficient gate stacks in BLGFETs The final stage should be technology transfer to wafer-scale BLG. Once wafer-scale high-gain graphene logic gates become available, their application in ultra-fast logic circuitry should be investigated.

The long-term goal is more challenging, as no satisfactory solution has been found so far in order to open a bandgap in graphene without reducing μ. Bandgap engineering of graphene should be attempted by patterning into GNRs. However, state-of-the-art sub 10 nm GNRs have very low μ ∼ 200 cm2 V−1 s−1 (ref. 312) as a consequence of carrier scattering on disordered ribbon edges.312 In order to eliminate unwanted scattering, GNRs should have crystallographically smooth edges308,310,320 and be deposited on insulating substrates. This leads to an enormous fabrication challenge as GNR widths ∼1 nm are required in order to reach the bandgap of Si (∼1 eV), as necessary for reliable switching. GNRs, ∼40 nm wide, grown on SiC have shown huge μ (∼6 × 106 cm2 V−1 s−1) and could be a viable strategy, although the width (∼40 nm) and the SiC cost could be an issue for scaling up.100 Finally, complementary logic (Fig. 59) is currently realized through electrostatic doping1144 which imposes limits on supply voltages in logic gates. In order to lift this restriction, GNRs should be chemically doped1150,1166,1167 but this doping should not introduce additional scattering centres in order to maintain high-μ of crystallographically smooth GNRs.

image file: c4nr01600a-f59.tif
Fig. 59 Schematic of a large array of digital complementary inverters integrated on wafer-scale graphene. Each inverter consists of two FETs that share the same gate (input) and drain (output). The gate stack is comprised of metal (Al; dark red core) covered by an insulating layer (AlOx; gray shell). Source and drain contacts (Ti/Au; yellow) overlap with gate contacts to eliminate access resistances and increase voltage gain of the inverters.

Conventional GFETs cannot be turned off in either of the two logic states and a typical in/out voltage swing is 22% of the supply voltage at RT.1144 Although this is less than the voltage swing in Si CMOS (capable of rail-to-rail operation, i.e. output voltage goes from ground = 0 V–GND – to supply voltage, with the voltage swing reaching almost 100% of the supply voltage),1161 it is still more than the swing in ECL gates.1168 Similar to graphene logic gates, ECL gates also comprise overdriven transistors in order to achieve ultra-fast operation.1168 For this reason a typical swing of the ECL gates is 0.8 V at a supply of 5.2 V, i.e., only 15% of the supply voltage.1161 ECL gates are at the core of the fastest SiGe and InP bipolar-CMOS (BiCMOS) or heterojunction bipolar transistor (HBT) chips and are used for digital signal processing at ultra-high frequencies (f > 100 GHz),1169 inaccessible with conventional state-of-the-art CMOS technology. They are used in high-speed integer arithmetic units,1170 static ultra-high frequency dividers,1171 high data rate (>50 Gb s−1) serial communication systems for demultiplexing (i.e. extracting the original channels on the receiver side)1172 and phase detection for clock and data signal recovery.1173 Hence graphene logic gates could find uses in applications not suitable for traditional Si logic, such as ultra-fast logic applications where power dissipation is not a concern, or transparent circuits on flexible (plastic) substrates.

GNR FETs should be considered as a replacement for Si FETs in CMOS logic once they reach sufficiently large Av and ION/IOFF, as discussed above. However, at this stage it is not clear whether this would be sufficient to migrate from Si to graphene logic. In the very optimistic scenario in which μ in GNR FETs would exceed that in Si FETs by an order of magnitude, it would still require a FET to have 100 GNRs (W = 1 nm) connected in parallel in order to reach the same current drive of a Si FET (W = 1 μm).

The timeline is shown in Fig. 61 and the main targets for digital logic gates are: 510 years: Ultra-fast (>100 GHz) integrated digital logic gates replacing ECL gates. 510 years: Simple digital logic gates on flexible or transparent substrates. 1520 years: General-purpose low-power GNR digital logic gates replacing Si CMOS.

5.2.3. Digital non-volatile memories. Non-volatile memories are the most complex and advanced semiconductor devices following the Moore's law below the 20 nm feature size.1118 State-of-the-art non-volatile memories consist of floating-gate flash cells,1174 in which the information is stored by charging/discharging an additional floating gate embedded between the standard control gate and semiconductor channel of a MOSFET.

Aggressive scaling of CMOS technology has a negative impact on the reliability of non-volatile memories. Parasitic capacitances between the adjacent cells increase with scaling, leading to a cross-talk.1175 Diminished lateral area leads to reduced gate coupling and therefore to higher operating voltages.1176 Larger number of array cells leads to a reduced sensing current and increased access times.1177 For these reasons, alternative materials and storage concepts have been actively investigated, including implementation of graphene in non-volatile memories1140,1178–1182 see Fig. 60. Research is needed in order to assess the most important figures of merits of non-volatile memories, such as endurance (i.e. the maximum number of cycles that the memory cell can withstand) and program/erase (P/E) [a cycle comprised of writing and then erasing data in a memory cell] curves, and to correctly extrapolate the retention time (i.e. the capability to retain a programmed state over time). Similarly to logic gates, non-volatile memories also require large enough (>104) ION/IOFF for memory states to be unambiguously resolved from one another. For these reasons, the following parameters should be thoroughly investigated: P/E curves as a function of time, the available P/E window (i.e., difference in threshold voltage or current between the two logic states), the retention, and the endurance. In addition, if graphene is to be used as a conductive channel in flash FETs, bandgap engineering should be pursued, as for graphene logic gates.

image file: c4nr01600a-f60.tif
Fig. 60 Two cells graphene NOR gate flash memory. Graphene is used for conductive FET channels1140,1181 and bit line (black), control gates1180 and word lines (brown), and floating gates1179 (white).

The use of graphene in non-volatile memories is facing less challenges than in logic gates, because memory operation requires only large (>104) ION/IOFF (assuming that the on current is not too low) without need for Av > 1. Av of memory GFETs is not important, as the reliability of a memory state readout depends only on the sensitivity of the sense amplifiers connected to the bit lines. Graphene could be used in non-volatile memories as channel,1140,1181 resistive switch,1178,1182 and storage layer, i.e., replacement of floating1179 or control gates.1180

The timeline is shown in Fig. 61, and graphene non-volatile (flash) memories might be developed in 10–15 years.

image file: c4nr01600a-f61.tif
Fig. 61 GRMs “digital electronics” timeline.
5.2.4. Interconnects in integrated circuits. State-of-the-art ICs contain large number of FETs (e.g., a typical microprocessor contains >109 FETs1183) which must be interconnected in order to perform required functions. Interconnection of such large number of FETs requires a complex multi-level metallization network (e.g., 9 metal levels are used in state of the art microprocessors1184) which consumes the largest part of the patterned die volume. This network is especially large in power ICs, in which interconnects must withstand large currents (typically >10 A (ref. 1185)). State-of-the art interconnects are usually made of Cu, with maximum current density ∼1 MA cm−2, limited by electromigration.1186 Graphene can be considered as an alternative, because it has very large current-carrying capability,1187 which offers possibility for size reduction of interconnects.

Exfoliated SLG can sustain (without oxidization due to Joule heating) 1.2 mA μm−1 = 12 A cm−1 under ambient conditions.1187 Assuming that each SLG (0.33 nm) within a multilayer stack can sustain the same current density, the breakdown current density of a multilayer stack is ∼360 MA cm−2, i.e. 360 times more than Cu.1188 However, initial investigations of wafer-scale multilayer graphene stacks revealed an order of magnitude lower breakdown current density (40 MA cm−2).1189 Although this is still 40 times more than in Cu, the Rs (>500 Ω/□) of these 20 nm thick graphene stacks corresponds to σ < 0.1 × 103 S m−1, much less than that of Cu (σ = 60 × 103 S m−1). Therefore, further development in wafer-scale graphene synthesis is needed to increase the conductivity of thick graphene films, therefore reduce parasitic resistances of graphene interconnects.

Graphene interconnects should be first introduced in ICs in which FETs are also made of graphene, in order to eliminate initial problems of contact resistance between graphene interconnects and contacts in non-graphene FETs. Eventually graphene interconnects should be introduced both in power and general-purpose ICs.

The timeline is shown in Fig. 61. 58 years: Interconnects in graphene ICs. 510 years: Interconnects in power and general-purpose ICs.

5.3. High frequency electronics

High frequency electronics is a cornerstone of today's high-tech economy. The continuous downsizing of components in ICT sustained the electronics industry for more than three decades. This field was first dominated by defence applications, until the late 1980s, and then it moved into the mainstream in the 1990s owing to advances in wireless communications.1190 Not only because miniaturization reaches fundamental physical limits, not solvable with conventional Si technology, but also because emerging applications, such as THz-spectroscopy,1191 require higher and higher frequencies, hardly achievable with established technology platforms.

Thus, a radical new approach is needed. Graphene as material platform for both digital and analogue electronics might overcome many obstacles: scaling beyond the Si limits is possible because graphene is ultimately thin and the high μ allows transistors to operate at frequencies beyond 1 THz.

The first entry points in electronics industry for graphene may be in analogue high frequency electronics, as there the advantages are most distinct in comparison with established technologies. Graphene may allow higher operation frequencies for frequency doubling and mixer applications than Si or Si/Ge, avoiding disadvantages of III/V materials, such as production costs, toxicity and poor integrability into a cost efficient Si technology.

In addition, ambipolar devices can significantly reduce the number of transistors needed in these applications. Simpler circuits mean less power consumption and smaller chip area. Considering that RF circuits are much less complex than digital logic ones, makers of RF chips are more open to new device concepts. Indeed, a large variety of different transistor and materials are today used in RF electronics, such as Si n-channel MOSFETs, HEMTs based on III–V semiconductors (GaAs;InP), and various bipolar transistors.1192

Graphene transistors with a 240 nm gate length operating at frequencies up to 100 GHz were demonstrated in early 2010.788 This cut-off frequency is already higher than those achieved with the best Si MOSFET having similar gate lengths.788,433 Cut-off frequencies over 300 GHz were reported with graphene transistors with a 140 nm gate length, comparable with the very top HEMTs transistors with similar gate lengths.1197 More recently, a cut-off frequency of 427 GHz was extracted for a 67 nm channel length graphene transistor,1198 see Fig. 62. These results are impressive, comparing the young age of graphene with the longer timescales of other devices. This is also a clear indication that GFETs have the potential to pass the THz-border in the near future. Thus, graphene may offer a cost efficient platform for novel applications in a variety of fields, such spectroscopy or automotive radar in analogue high frequency electronics. Significant impact in analogue RF communication electronics in areas as diverse as low noise amplifiers, frequency multipliers, mixers and oscillators could be achieved within the next 10 years, see the timeline in Fig. 67.

image file: c4nr01600a-f62.tif
Fig. 62 Cut-off frequency versus gate length for GFETs, nanotube FETs and three types of RF-FETs [Adapted from ref. 1193]. pHEMT = pseudomorphic HEMT, mHEMT = metamorphic HEMT. HEMTs are made of heterojunctions, i.e., junctions of different materials.1194,1195 The most common example are GaAs HEMTs in which there are junctions between AlxGa(1−x)As and GaAs. Different materials have different lattice constants, hence crystal defects are expected in HEMTs. In pHEMTs this is avoided by using a very thin layer of one material1196 so that accumulated energy in this layer is not enough to relax the layer and introduce defects (energy is low because the layer is very thin).1196 In mHEMTs a graded buffer layer is used in between the layers to accommodate both lattice constants1196 and eliminate the defects. E.g., to grow Al0.3Ga0.7As on GaAs, an AlxGa(1−x)As layer is grown first in which x = 0 at the beginning (GaAs) and then x is gradually increased to 0.3 (Al0.3Ga0.7As).1196

Research efforts will not only be required for the optimization of graphene-based devices, but also for the development of circuit designs that can fully exploit the unique properties of graphene-based devices. The absence of a band-gap, therewith the un-incisive current saturation, the ambipolarity and the targeted operation frequencies going beyond 100 GHz require new concepts on system level and circuit design, which could also open the door towards novel functionalities. Thus, to take advantage of the full potential of graphene devices, the aim is to combine more basic research with improved material growth and device technology. A better understanding of parameters such as breakdown voltage, electron velocity, and saturation current is needed to allow a complete benchmark of this material and an evaluation of its potential performance. In addition, these new applications will have to overcome the limitations that arise from the lack of band-gap.

Once the growth and fabrication technology matures, the main challenges for RF applications are the integration of graphene devices in Si technology, and the increase of their trans-conductance by, e.g., introducing a band gap, and ensuring sufficient Av and output currents so that components can be integrated to circuits.

For real applications not only the cut-off frequency is an important figure of merit; Av and fmax, the frequency where the unilateral power gain reaches unity, are even more important. Both of these quantities are much more sensitive on the process technology than the cut-off frequency. In graphene transistors Av and fmax still lag behind established technologies. Therefore, much effort on developing the process technology is required. Additionally, integrated circuits of higher complexity have not been realized so far, because the required process technology is still in its embryonic stage and circuits can hardly be designed, as a parameter library for graphene-based devices is still missing.

Advanced graphene devices have the potential to transform communication systems in a broad array of new applications in the next 20 years. Graphene is therefore in an excellent position to help RF systems become even more ubiquitous and versatile than they are today.

5.3.1. Analogue voltage amplifiers. The main building block of analogue electronics is a voltage amplifier: an electronic device capable of amplifying small AC voltage signals. For the same reasons discussed in section 5.2.1 in case of digital logic gates, AC Av is usually much less than unity in graphene circuits. The use of GFETs in analogue electronics is currently limited to niche applications, such as analogue mixers,1199 but even these require voltage amplifiers for signal processing. RT operation of GFETs with a high Av is of paramount importance; otherwise, graphene circuits and detectors should be destined to rely on Si FETs for signal amplification and processing.1200 This is not favourably viewed by the semiconductor industry, which generally does not like expensive hybrid technologies.

One of the main factors contributing to a low Av is the use of back-gated Si/SiO2 devices, which also suffer from large parasitic capacitances and cannot be integrated with other components. For this reason, top-gated GFETs with thinner gate insulators were extensively investigated,1201 as in case of digital logic gates.

The future investigation of graphene voltage amplifiers partly overlaps with that of digital logic gates, as in both cases the short-term goal is the same: large Av > 10 should be obtained in wafer-scale SLG and BLG grown by CVD or on SiC substrates. In order to further increase Av, FETs should be fabricated from wafer-scale BLGs. Finally, several remaining challenges of GFETs, technological rather than fundamental in nature, should be addressed. E.g., graphene circuits remain sensitive to fabrication induced variability. Higher μ, gm and lower gd and contact resistance should increase Av for both analogue and digital applications. The long-term goal should be the integration of graphene amplifiers in more complex (i.e., multi-stage) analogue circuits.

GFETs are well suited as building blocks of low-noise amplifiers (LNAs) as they exhibit very low levels of electronic flicker noise102 (or 1/f noise, where f is the frequency, occurring in almost all electronic devices and related to a direct current: a resistance fluctuation, which is transformed to voltage or current fluctuations via Ohm's law) which dominates the noise spectrum at low frequencies.1202–1204 Such voltage amplifiers are also expected to benefit from graphene's high mechanical and chemical stability and high κ.102 Graphene LNAs are needed in high-frequency electronics,1205,1206 as their realization would allow seamless integration with graphene analogue mixers, thus eliminating need for Si FETs in these applications. As present, GFETs cannot be turned off, thus class-A amplifiers (i.e. amplifiers which conduct during the whole input cycle) with low harmonic distortions should be developed. Large Av may allow realization of electronic harmonic oscillators, combining high-Av amplifiers with passive graphene feedback networks.

The development of graphene voltage amplifiers could pave the way for graphene power amplifiers. These are usually found in the final stages of more complex amplifiers. They operate with Av = 1 and have a sole purpose to match the previously amplified signals (provided by the voltage amplifiers) to a low-impedance load, such as a loudspeaker (∼4 Ω) in high-fidelity audio systems1207 or antenna of a transmitter (∼50 Ω) in RF applications.

The timeline is shown in Fig. 67. 34 years: LNAs. 45 years: Audio and RF voltage amplifiers. 56 years: Harmonic oscillators. 510 years: Power amplifiers.

5.3.2. Graphene ring oscillators. GROs1102,1155,1208 are an extension of cascaded graphene inverters.1144 Each inverter in the loop must be identical, with Av > 1 and in/out signal matching. The two FETs in each inverter must also have very low on-state resistance to be able to quickly charge/discharge the gate capacitance of the next stage in order to reach high-frequency operation. Since the oscillation frequency fo is a direct measure of delays in realistic scenarios, ROs are the standard test-beds for evaluating ultimate limits and clock rates of digital logic families.1209 This is because a realistic electronic circuit is both driven and loaded by other electronic circuits, exactly what exists in a RO; in contrast, fT is measured on a single FET which is driven by an ideal current source and has a shorted output (zero load impedance). The ROs in Fig. 63 were fully integrated on CVD graphene.1102
image file: c4nr01600a-f63.tif
Fig. 63 Integrated GRO. (a) Circuit diagram of a three-stage RO. The RO comprises three inverters (1–3) cascaded in a loop with a fourth inverter (4) decoupling the RO from the measurement equipment connected to the output (out). (b) Optical microscope image of a RO with gate length L = 1 μm integrated on a CVD graphene channel. The drain contacts of inverters 1–3 (Au) overlap with the gate contacts (Al/Ti/Au) in order to form internal connections between the inverters. GND = ground. Adapted from ref. 1102.

Complementary operation of the inverters within a RO was obtained between the Dirac points of two GFETs1102, as in early graphene inverters.1149 These inverters have Av > 4, enough to enable oscillations under signal matching. The oscillation frequency of a RO with n stages cascaded in a loop is inversely proportional to the inverter rise/fall delays τ as fo = 1/(2) = fo1,1210 where fo1 is a fan-out of one (FO1) oscillation frequency, where fan-out is the number of digital inputs that the output of a single logic gate can feed. Since τCGD−1,1102 with GD the sum of extrinsic drain conductances of the GFETs in the inverter and C the parasitic capacitive load of the inverter, reduction of parasitics leads to high-frequency operation.

The speed of electronic circuits is typically increased by downscaling their dimensions,12 which also reduces parasitics (because parasitic capacitances are proportional to the surface area of the corresponding device parts). This is demonstrated in Fig. 64, plotting the maximum oscillation frequency of 26 GROs as a function of gate length L. The highest is fo = 1.28 GHz at L = 1 μm, corresponding to a FO1 τ ∼ 100 ps. This is similar to that of conventional Si CMOS ROs,1211 and smaller than polycrystalline Si CMOS thin-film ROs1212 for the same L. This is the first operating frequency above 1 GHz in digital circuits based on any type of novel low-dimensional materials (e.g. nanotubes, nanowires, graphene, MoS2). By comparison, the highest frequency of nanotube ROs was ∼50 MHz,1213 and 1.6 MHz in 2L-MoS2 ROs.1214

image file: c4nr01600a-f64.tif
Fig. 64 Scaling of oscillation frequency with L−1. The plot shows the values from measurements on 26 GROs (only the highest frequencies for each GRO are shown).1102

The dependence of the oscillation frequency of conventional ROs on the supply voltage is a serious problem in complex digital circuits, since increased power consumption places strong demand on the supply voltage causing it to fluctuate.1215 Such fluctuations degrade the operation of logic gates and their noise performance.1215 In contrast to other types of ROs, GROs are much less sensitive to fluctuations in the supply voltage, as a consequence of their reduced voltage swing.1102 Insensitivity to power supply noise could be a disadvantage in applications in which dynamic frequency and voltage scaling are used to slow the clock rate of digital circuits during periods of reduced workload. However, such insensitivity represents an important advantage in applications in which frequency stability is important, e.g. for clock generation and recovery in high-speed digital systems.

GROs and inverters could find applications in digital circuits operating at extremely high frequency (EHF; f > 100 GHz) in which high operating speed could be traded off against power dissipation, reduced voltage swing, and circuit complexity1170,1172,1173,1216 These ultra-high speed digital circuits were developed to perform data conversion at the transmitting/receiving side of serial EHF lines,1217 such that information carried by EHF digital signals can be processed at lower clock rates by low-power, highly integrated, and parallel Si CMOS logic.1217 The EHF digital circuits are used in wireless, fiber-optic, and space communications.1169

GROs can find applications in analog electronics. Oscillators are one of the main building blocks of analog electronics,1218e.g. RF (microwave) electronics is built on voltage amplifiers, oscillators, and mixers.1219 Graphene amplifiers,1142,1143 mixers11,1220 and oscillators1102 may allow realization of all-graphene microwave circuits.

GROs can also be used to mix analog signals. Graphene analog mixers11,1220 so far require an external local oscillator (LO) for frequency conversion.11,1220 GROs can overcome this limitation and perform both modulation and generation of oscillating signals to form stand-alone graphene mixers, i.e. mixers with a built-in LO. To this end, the GRO in Fig. 63 was modified by superimposing a RF signal over the DC supply of the buffering inverter 4 (without affecting the supplies of the other three inverters).1102 As the other inverters forming a ring are not affected by the addition of the RF signal, the buffering inverter mixes the RF signal with the unaltered AC component of the oscillating voltage. Fig. 65 shows the power spectrum of the up-converted RF signal around the LO signal. The conversion loss is 19.6 dB at an LO power of −18.5 dB m and RF power of −34.3 dB m, better than in early graphene mixers,11,1220 and comparable to recent graphene mixers.1221,1222

image file: c4nr01600a-f65.tif
Fig. 65 Power spectrum of the output signal of a stand-alone graphene mixer at VDD = 2.5 V. The signal frequencies are fLO = 292 MHz and fRF = 25 MHz. Apart from the signals discussed in the text, the output signal also contains a frequency component at fRF, from the amplification of the RF signal by inverter 4 in Fig. 63.1102

5.4. Layered materials-based devices

From the device point of view, bulk MoS2 seems to be a most interesting LM because its monolayer is a direct bandgap (1.8 eV) semiconductor,111 which allows realization of electronic379 and optoelectronic378,1223 devices. Due to a large band gap, FETs fabricated from 1L-MoS2 exhibit large ION/IOFF ∼ 108.379 The current μ in 1L-MoS2 deposited on conventional insulating substrates (e.g., SiO2) is high enough5,384,1231 for electronic applications at technologically relevant frequencies (>1 GHz).2338 The low current drive of MoS2 FETs can be improved by using 2L-MoS2 at the expense of a small reduction of ION/IOFF (∼107).1214 In this case, the highest oscillation frequency in 2L-MoS2 ROs is 1.6 MHz, limited by the high gate capacitance.1214μ of MoS2 was increased up to ∼500 cm2 V−1 s−1 in thicker layers (∼50 nm) deposited on PMMA.1224 However, such samples suffer from even smaller ION/IOFF (∼104) and are yet to be incorporated in top-gated FETs.

Theoretical predictions show that at RT, μ is expected to be limited by phonon scattering,1225 with values in the 320–410 cm2 V−1 s−1 range,1225 so further improvements in material and device processing seem possible. Provided that the charged impurity concentration could be sufficiently reduced, at low T μ would be dominated by scattering by acoustic phonons,1226 and μ > 105 cm2 V−1 s−1 for T < 10 K and carrier concentration >1011 cm−2 was predicted in ref. 1226. Other possible sources of scattering are ripples that have been observed in MoS2 membranes.1227 These consist of local height fluctuations and are also expected to reduce the conductivity of 1L-MoS2.1228

1L-MoS2 devices exhibit n-type transport behaviour. Ambipolar transport was demonstrated in thin (10 nm) MoS2 electrical double-layer transistor using an ionic liquid as the gate.1229 The demonstration of both p and n-type transport in devices based on solid-state dielectrics would be useful for applications in CMOS-type logic and pn-junctions for optoelectronic devices. The use of liquid gates has also recently allowed large charge carrier concentrations in 10 nm thick MoS2 and gate-tunable superconductivity.1230 Metal-insulator transition in 1L-MoS2 Hall-bars using classical dielectrics (HfO2) was reported.1231

The presence of a band gap in MoS2 allows strong drain current saturation, with drain-source conductance gds = dIds/dVds < 2 μS μm−1 demonstrated in 1L-MoS2,1232 together with high transconductance gm of 34 μS μm−1, which should allow high voltage gains for small signals (AV = gm/gds > 10). Such gains allowed the fabrication of digital circuits based on MoS2, such as inverters,1044 ROs1214 as well as analogue amplifiers.1233

Transistors based on MoS2 can also operate with current densities of at least 4.9 × 107 A cm−2,1232 50 times higher than Cu, comparing favourably to the breakdown current density for graphene on SiO2 (>108 A cm−2).103 This is due to intralayer Mo–S covalent bonds being much stronger than metallic bonds. Such high breakdown current density would allow aggressive downscaling of MoS2 devices. Integration with high κ substrates, such as diamond,199 could result in further increase of breakdown current density.

Device simulations of 1L-MoS2 transistors425,1234 predict that 1L-MoS2 would show superior resistance to short-channel effects due to its small thickness. These calculations show that top-gated MoS2 transistors with gate lengths of 15 nm could operate in the ballistic regime with on-current as high as 1.6 mA μm−1, subthreshold swing close to 60 mV dec−1 and ION/IOFF ∼ 1010. Although MoS2 will probably not compete with conventional III–V transistors on μ values, attractive electrical performance characteristics and high electrostatic control could make it and other TMDs viable candidates for applications in low-power electronics.

The use of 2d materials could be very advantageous for the realization of memory devices based on the floating gate transistor structure,1235,1236 used in flash memories. In this type of device, the operation is based on detecting a threshold voltage shift caused by a presence or absence of charges trapped on the floating gate. Decreasing the device size is limited by the amount of charge that can be stored on the floating gate. Using 1L-MoS2 or other 2d semiconductors as conductive channels in such devices brings the benefit of increased sensitivity to external electric charge and could allow deeper scaling. Such devices based on 1L-MoS2 and graphene as key elements were reported1235 in a geometry shown on Fig. 66. Graphene here acts as an ohmic-like contact, allowing efficient charge carrier injection into MoS2 while multilayer graphene is used as the floating gate. The use of a 2d contact in place of thicker metallic films is beneficial for several reasons. It allows one to realize devices and circuits with 2d materials using less expensive fabrication techniques, e.g. R2R printing, see section 4.11.8.

image file: c4nr01600a-f66.tif
Fig. 66 MoS2/graphene heterostructure memory layout. 3d schematic view of the memory device based on 1L-MoS2 semiconducting channel, SLG contacts and FLG floating gate. The FLG floating gate is separated from the channel by a thin tunnelling oxide and from the control gate by a thicker blocking oxide.

image file: c4nr01600a-f67.tif
Fig. 67 High frequency electronics timeline.

The capacitive coupling between leads in neighbouring devices is also reduced, allowing further scaling. Because of its band gap and 2d nature, 1L-MoS2 is highly sensitive to the presence of charges in the charge trapping layer, resulting in a factor of 104 difference between memory program and erase states.1235 This allows a large memory window and the possibility to have different intermediate levels of charge in the floating gate corresponding to multiple bits of information.

Several issues however still need to be addressed. There is at this point no control over intrinsic doping in MoS2 and methods for introducing dopant atoms as well as achieving ambipolar and p-type behaviour would be highly desirable. Understanding how to make good electrical contacts to MoS2 is also lagging and it is not clear yet which material and contacting scheme would result in the smallest possible contact resistance.

Few-layer phosphorene116,1237 has been investigated as a new 2d p-type electronic material. Few-layer phosphorene is stable, having a direct band-gap that depends on the number of layers,116 going from ∼2 eV in monolayer to ∼0.3 eV in bulk.1237 Phosphorene field-effect transistors with 1 μm channel length have shown a RT μ (hole) ∼ 286 cm2 V−1 s−1, and ION/IOFF up to 104.116Ref. 1237 reported that μ in few-layer phosphorene is thickness-dependent, with the highest achieved values of ∼1000 cm2 V−1 s−1 obtained for ∼10 nm samples at RT. Although these results116,1237 demonstrate the potential of few-layer phosphorene as a new 2d material for applications in nanoelectronic devices, these are only the first reports and more work is needed to fully assess this material.

5.5. Novel vertical and planar transistors and devices

2d heterostructures offer richer opportunities in terms of physics and transport properties than each of the individual 2d crystals. The most obvious device to develop is the TFET.106 Initial results indicate that such devices will indeed offer required parameters in terms of ION/IOFF and μ. Other directions to explore, both experimentally and theoretically, are hot electrons transistors, resonant tunnelling and formation of minibands.

Planar devices can include double quantum wells, drag in parallel 2d electron gases, Bose–Einstein condensation in such systems, etc. Also, effects of enhancement of electronic properties (improved μ) for each individual conductive layer need to be addressed.

Exotic combinations, such as superconductor/insulator/normal metal or superconductor/insulator/ferromagnet, or even more complex structures, providing one can find suitable 2d crystals, need investigation. Such devices would allow the exploration of quasi-particles spin and valley degrees of freedom. By embedding nanoclusters of conventional metals, ferromagnets, and semiconductors between the layers, it may be possible to combine the opportunities for quantum technologies offered by the QD physics and the unique electronic properties of 2d heterostructures.

5.5.1. Vertical tunnelling transistors and vertical hot electron transistors. The vertical tunnelling transistor106 is a viable alternative to the current approach to graphene-based electronics. Rapid response and ultra-small sizes can also be achieved in vertical transistors. Electron transfer through nm thick barriers can be extremely fast (possibly coherent). Ballistic tunnelling transistors may allow to overcome the current low ION/IOFF. Tunnelling devices would have a highly insulated off state with no dissipation, which should allow not only individual transistors, but integrated circuits at RT. The latter is difficult to achieve for horizontal transport and remains a distant goal.

The aim is to explore, by experiments and modelling, several architectures for tunnelling/hot electron transistors. The simplest is metal/BN/SLG/BN/SLG, where the metal contact (separated from the bottom graphene by thick, tunnelling non-transparent BN) serves as a gate and the two graphene layers (acting as emitter and collector) are separated by thin BN layers, Fig. 68. This relies on voltage tunability of the tunnelling density of states in graphene, and on the effective height of the tunnel barrier adjacent to the graphene electrode.

image file: c4nr01600a-f68.tif
Fig. 68 BN/SLG/BN/SLG/BN devices.106 (A) Optical image. (B) Electron micrograph. Two 10-terminal graphene Hall bars are shown in green and orange. The scale is given by the 2 μm Hall bar width.

Configurations with several different dielectrics and the use of other heterostructures, such as metal/BN/SLG/MoS2/SLG, will need investigation. Proof of principle devices with BN tunnelling barriers were demonstrated with RT ION/IOFF ∼ 50.106 With an improvement of the quality of the heterostructures and the use of dielectrics with thinner tunnelling barriers, ION/IOFF may be brought close to that required by modern electronics (105). Ref. 1238 reported a graphene-WS2 heterotransistor with ION/IOFF > 106.

The investigation, by experiments and modelling, of other possible geometries for vertical transistors, including hot electron transistors similar to ref. 443, will need to be carried out. Few-atom-thick transistors1238 based on a 2d tunnel barrier and graphene may allow much better quality, and become more successful in applications.

The transit time through such sandwiches is expected to be ≪1 ps, whereas there are no limits for scaling down in the lateral direction to true nm sizes. Metal/BN/SLG/BN/SLG structures with both BN barriers transparent for tunnelling, and bottom graphene as the control electrode, can be exploited. The thickness of the active part would be <10 atoms (∼3 nm) and should allow ballistic current that is controlled by the central graphene electrode.

Attempts to produce stacks of several transistors in series (metal/BN/SLG/BN/SLG)N, thus introducing a vertical integrated circuit architecture, need to be carried out. Other possible architectures and electronic components as well as different ways of integration into vertical integrated circuits should be studied. E.g., a gate-tunable p–n heterojunction diode using s-SWNTs and 1L-MoS2 as p-type and n-type semiconductors, respectively, has been demonstrated.1239 The vertical stacking of these two direct band gap semiconductors forms a heterojunction with electrical characteristics that can be tuned with an applied gate bias to achieve a wide range of charge transport behavior ranging from insulating to rectifying with forward-to-reverse bias current ratios exceeding 104.1239

It will be necessary to investigate integration of tunnelling devices in realistic circuits, which require much larger on currents than currently obtained. Large on currents are needed in order reach AV > 1. In addition, in order to fully exploit the short intrinsic transition time of tunnelling devices, large on currents (i.e., small on resistances) are needed to reduce extrinsic RC time constants associated with capacitances in electronic circuits.

5.5.2. In-plane transport in 2d heterostructures. The aforementioned devices (with atomically thin tunnel barriers, graphene and other materials) are new experimental systems and offer a range of opportunities for fundamental and applied research. E.g., in terms of fundamental research, two graphene layers separated by a thin dielectric allow one to look for excitonic condensation432 and other phenomena mediated by e–e interaction. Coulomb drag is a good tool to probe many-body interactions, hard to discern in conventional transport measurements.1240

The advent of SLG/h-BN and other heterostructures offers a new venue for investigation of interlayer interactions. First, the 2d charge carriers in graphene are confined within a single atomic plane, whereas a few atomic layers of h-BN are sufficient to isolate graphene electrically. This allows extremely small (nm) separation between the graphene layers, which favourably compares with the smallest effective separation ∼15 nm achieved in GaAlAs heterostructures.1241 Second, charge carriers in graphene can be continuously tuned between e and h from n > 1012 cm−2 all the way through the neutral state, where the inter-particle distance nominally diverges.1241 This makes it possible to access the limit of strongly interacting 2d systems. First results demonstrate very strong Coulomb drag in BN/SLG/BN/SLG/BN systems,106 see Fig. 69. Optimisation of the structures may lead to many interesting effects.

image file: c4nr01600a-f69.tif
Fig. 69 (A) Tunnelling IV s and their response to gate voltage (in 5 V steps) for a BN/SLG/(BN)4/SLG/BN device.106T = 300 K. (B) Changes in low-bias tunnelling (symbols) and the theory fit for 4 hBN layers (solid curve). Adapted from ref. 106.

On the theory side, there is need to impose rigor onto widely varying conflicting predictions for the Coulomb drag and a related issue of exciton condensation.24Ref. 432 exploiting heterostructures comprising a SLG (or BLG) carrying a fluid of massless (massive) chiral carriers,1078 and a quantum well created in GaAs 31.5 nm below the surface, supporting a high-mobility 2d electron gas, found that the Coulomb drag resistivity significantly increases for T below 5–10 K, following a logarithmic law.432 This anomalous behavior is a signature of the onset of strong inter-layer correlations, compatible with the formation of a condensate of permanent excitons.432

Additionally, there is the need to evaluate the contribution of the phonon drag, related to the emission/absorption of vibrations in the separating insulating layers.

The opportunity to generate new or to enhance the earlier discussed correlation effects in structures with a complex architecture should be explored. The formation of an excitonic insulator in the BLG systems is hindered by screening of the Coulomb interaction, which appears to be sensitive to the electron spin and valley degeneracy. The T of the excitonic insulator transition in MoS2/(BN)n/MoS2 sandwiches may be higher and fall into the experimentally accessible range. Also, the search for correlated states in NbSe2/(BN)n/NbSe2 sandwiches should be undertaken, since these, by extrapolating from bulk properties of NbSe2 (ref. 1242), may form both charge density waves and superconductivity.

5.6. Electron emission

“Vacuum microelectronics” has been of interest since the 1970s.1243,1244 Its development was initially driven by the aspiration to create new, more efficient, forms of electronic information display known as “field effect (FE) displays”1245 or “nano-emissive displays”.1245 Even though prototypes have been demonstrated,1245 the development of such displays into reliable commercial products has been hindered by a variety of industrial production problems, essentially related to individually addressable sub-pixels technology, not directly related to the source characteristics. However, after considerable time and effort, many companies are now shutting down their effort to develop this technology commercially. This is essentially connected with the huge development of both flat-panel liquid crystal and organic light emitting diode (OLED) displays. Nevertheless, in January 2010, AU Optronics acquired FE display assets from Sony continuing the development of this technology.1246 Large-area FE sources involve many other applications, ranging from microwave and X-ray generation, space-vehicle neutralization and multiple e-beam lithography to plastic electronics.

The early devices were essentially the “Spindt array”,1243 and the “Latham emitter”.1247 The first used Si-integrated-circuit fabrication techniques to create regular arrays where Mo cones were deposited in cylindrical voids in an oxide film, with the void covered by a counter-electrode (CE) with a central circular aperture. In order to avoid IC fabrication the Latham emitters were developed.1247 These comprise two different devices, the metal–insulator–metal–insulator–vacuum and the conductor–dielectric–conductor–dielectric–vacuum.1247 The latter contained conducting particulates in a dielectric film and the FE is assured by the field-enhancing properties of the micro/nanostructures.

Nowadays this research area targets the development/investigation of new nanomaterials that could be grown/deposited on suitable substrate as thin films with appropriate FE properties. In a parallel-plate arrangement, the field between the plates (FM) is given by FM = V/W, where V is the applied voltage and W is the plate distance. If a sharp object is grown/deposited on one plate, then the local field F at its apex is greater than FM and can be related to FM by: F = ζFM. The parameter ζ is called the “field enhancement factor” and is essentially determined by the object's shape.1248 Because FE characteristics are determined by F, the higher ζ, the lower FM and, for a given W, the lower V at which field emission occurs.

FE from amorphous, “diamond-like” carbon1249,1250 and nanostructured carbon films1251,1252 has been investigated for decades. The introduction of CNT-FEs,1253–1256 was a significant step forward, with extensive research carried out both for their physical characteristics and possible technological applications.1253–1256 In recent years there has also been interest in the development of nanostructures, with a sufficiently high density of individual emission sites based on other carbon forms such as “carbon nanowalls”1257 and on various forms of wide-band-gap semiconductor.1258

Graphene has atomic thickness, high aspect ratio (ratio of lateral size to thickness), excellent σ, and good mechanical properties, which qualify it as an attractive candidate as FE source.1259–1262 Carbon has one of the lowest sputter coefficients,1263 which is an advantage as an electron source is usually bombarded by positive ions. Consequently, thanks to the aforementioned properties, an enhanced local electric field and good electron-emission stability can be expected for graphene (Fig. 70).

image file: c4nr01600a-f70.tif
Fig. 70 Screen-printed graphene FE device.

The presence of edges may render graphene superior to CNTs for tunnelling of electrons.1260 Indeed, the atoms at graphene edges can have an unconventional electronic structure1264 and may form a distorted sp3-hybridized geometry,1260 instead of a planar sp2-hybridized configuration, with the formation of localized states, with possible barrier decrease for electron emission.1260 The orientation of graphene deposited on the substrate is another key factor for high rate FE. Indeed, graphene sheets deposited flat on the substrate surface show low field enhancement.1261 Both the interface contact and adhesion between graphene and substrate need be optimized to facilitate electron transport, and consequently improve FE performance.

The target is to position graphene as a material for vacuum nanoelectronics. One needs to develop new concept devices, to make graphene useful in applications such as lighting elements, high-brightness luminescent elements, FE lamps, cathode-ray lamps, X-ray-tube sources, electron sources for high-resolution e-beam instruments such as e-beam lithography machines and electron microscopes, EM guns and space applications, such as the high-precision thrusters needed for next generation of space telescopes. However, whichever is the application, FEs work best in UHV. The degradation of emission performance by adsorption of gas atoms needs to be addressed. This is a common problem with all FE devices,1245 particularly those that operate in “industrial vacuum conditions”.1247 The emitter shape and work function can in principle be modified deleteriously by a variety of unwanted secondary processes, such as bombardment by ions created by the impact of emitted electrons onto gas-phase atoms and/or onto the surface of CEs. Moreover, impurities such as oxygen, water, and organic residues are unavoidably absorbed on the graphene-emitter and the substrate during the transferring process to the FE chamber. The impurities could form dipoles and apply an additional disturbance on local electrical field near the graphene edge. The disturbance may change the local work function.137 In addition, a number of electrons emitted from the cathode could be trapped in the impurities in front of cathode, reducing its local electrical field. Therefore, the emission performance could be degraded. Emission stability, lifetime and failure mechanisms, energy spread, reduced brightness all need to be addressed.

The FE properties of graphene films may be further improved by optimizing the intrinsic structure of graphene, the deposition processing, and the morphology and thickness of the films. Reliable methods for the deposition of field emitting graphene and/or graphene/polymer composite films on different substrates need to be investigated and developed, opening up avenues for a variety of applications. Uniform morphology, high graphene density, and optimum graphene sheets orientation with respect to substrate surface (graphene edges normal to the substrate) may ensure emission uniformity and sufficient field-emission tips on the film surface in order to lower the threshold fields (<1 V μm−1). A successful strategy could rely on the growth of vertically aligned graphene sheets.

6. Spintronics

During the past decade, a huge research effort focused on providing new opportunities for beyond CMOS applications. It is now widely accepted that the emerging solution will be a hybrid of the most interesting technologies rather than a single one. New breakthroughs are expected at the boundary between different fields from chemistry to solid state physics, mechanics and optics, coupling their different degrees of freedom. There are increasing incentives to explore new avenues to follow the ever growing need for computational speed and storage capability. While some of the directions look at new types of devices (such as cross-nanowire transistors1265), others explore novel physical phenomena (using electron spin1266 instead of charge) and alternative materials such as graphene.

Devices relying on spintronics hold unique prospects for ICT, but, to date, the advent of spintronics-based logic has been impeded by the difficulty to achieve both long spin lifetimes and spin control simultaneously. Spin control is usually associated with a sizable spin–orbit coupling.1267 However, a large spin–orbit coupling tends to lead to fast spin decoherence.1268 Novel materials and approaches to circumvent this problem are crucially needed.

In this perspective, graphene offers opportunities for efficient spin manipulation and for the creation of a full spectrum of spintronic nanodevices beyond current technologies, including ultra-low energy demand devices and circuits comprising (re-) writable microchips, transistors, logic gates, and more, while being compatible with more-than-Moore CMOS and non-volatile low energy magnetoresistive random-access memories (MRAM). Graphene provides solutions to integrate several elements (active and passive) on the same platform. While graphene- or GO- hybrids have been shown to have memory effects,1269 the underlying microscopic mechanisms still need to be clarified. Non-volatile GFETs with ferroelectric gates have been demonstrated to operate as three terminal resistive memories,1270 while graphene-based memristors (non-linear passive two-terminal electrical components)162 are interesting since they may act both as memory and logic elements.

6.1. Graphene spintronics

There are two fundamental properties of the electron which are exploited in advanced technology – charge and spin – but conventionally their functions have been kept separated.1271 The charge is used in all electronic devices, particularly in logic circuits, where current flows are controlled using electric fields, encoding and processing binary (or analogue) information. The spin is used in its “collective” form of magnetic domains for long-term data storage, from ferrite core memories to modern hard drives.1271

This separation has begun to change with the advent of spintronics, a relatively young field, the name itself having been introduced only in 1996.23 This combines charge and spin of electrons in the solid state to create novel functionalities. In the simplest spintronic device, called spin-valve,1272 an electronic current flows between two ferromagnetic electrodes through a non-magnetic channel carrying a spin signal along with it, normally detected through its magnetoresistance,26 and its transport properties are influenced by magnetic ordering. The requirement for non-magnetic channels is to transport spin currents, minimizing spin relaxation due to spin-scattering events that produce spin-information losses through spin-flipping, ultimately (in most cases) caused by spin–orbit coupling.1273,1274

Graphene consists of light atoms, thus exhibits (in its pristine form) negligible spin–orbit coupling1275 and a possible absence of hyperfine interactions, due to the lack of nuclear spins in isotopically pure materials. These are the properties required for high spin-signal conservation,1276 making graphene an interesting candidate for non-magnetic channels, with long spin relaxation and coherence times, the central benchmarks for spintronic devices.

Integration between graphene and other materials, including 2d crystals, shows great potential for further improvements. This includes, e.g., substrate optimisation, as shown by the remarkable increase of spin relaxation length obtained by depositing graphene on a BN substrate1277 or on FLG.223,1278–1280 However, from Hanle experiments (i.e. magnetoresistance measurements in which the spin is forced to precess by an external magnetic field, while the magnetoresistance signal is monitored by spin diffusion length,223,1300 see Fig. 71), spin relaxation times have been reported to be in the ns range for varying quality of graphene samples,1276 whereas the spin relaxation mechanism is under debate, and a fully comprehensive picture is still lacking.

image file: c4nr01600a-f71.tif
Fig. 71 (a) Typical two-terminal local spin valve device. The width of graphene channel on SiC is 10 μm, and the distance between the two Al2O3/Co electrodes is L = 2 μm. The optical image shows the entire structure, including contact pads. (b) Atomic arrangement of SLG on SiC. (c) Large local resistance spin signals measured at 4 K. (d) Oscillating and decaying device non-local resistance as a function of the applied magnetic field (Hanle measurement). Adapted from ref. 223, 1300.

Following what is known for metals and semiconductors, two mechanisms are currently believed to take place in graphene: Elliot–Yafet (EY)232,233 and Dyakonov–Perel (DP).162 In EY the electron spin has a finite probability to flip during each scattering event off impurities or phonons.232,233 The DP mechanism is driven by the precession of electron spins along effective magnetic field orientations, which depends on the momentum, and with direction and frequency of precession changing at each scattering event.162 The most recent theoretical derivation in SLG (taking into account the Dirac cone physics) reports a spin relaxation time varying proportionally to the transport time, behaviour typical of the EY-relaxation.1273

A primary objective in the near future is to clarify the true nature of the dominant mechanisms determining the spin relaxation time and spin relaxation length in high quality graphene devices. Indeed, while spin transport in graphene has been demonstrated by several groups worldwide, the full understanding of spin dynamics and relaxation is still lacking, and theoretical predictions or generalization of the EY and DP-relaxation mechanisms seem unable to cope with the experimental reality.1276 Recent results indicate that the situation can be improved by a precise tuning of the contact resistance between graphene and magnetic electrodes,223 but it is still unclear what this implies for large scale RT devices. More effort is thus needed in order to explore the spin relaxation mechanisms from a theoretical perspective beyond the conventional perturbative treatments. Spin relaxation time and the g factor (or dimensionless magnetic moment) need to be evaluated by taking into account the relevant scattering processes close-to and away-from the Dirac point (short and long range scatterers, e–e and e–ph interaction etc.), depending on the intrinsic and extrinsic nature of scattering sources, and for the main classes of materials of relevance for RT operability.

Simultaneously, a global and comprehensive understanding of spin relaxation mechanisms needs to be established based on the existing information and more targeted and systematic experiments combining two-terminal magnetoresistance and four terminal Hanle transport measurements (see Fig. 71) and disentangling the contribution of contacts from intrinsic spin relaxation induced by defects, disorder, interaction with substrate, etc. This is a major bottleneck, which demands a concerted effort between experiment and theory.

6.2. Spin injection in graphene

The variety of favourable properties for spintronics that graphene provides include also efficient spin injection,223 due to the tunability of the Fermi level. The large electron velocity implies that graphene has unique properties for the transport of spin polarized currents to long distances.223 The analysis of scattering mechanisms is necessary in order to achieve optimal production methods for graphene-based spin-valves. The optimization of performances of graphene spin-valve devices is required, comprising a focus on the specific role of magnetic contacts, including their spin injection and detection efficiencies and the possible role of interface effects in spin relaxation and spin dephasing.

Interface engineering to optimize spin injection and detection from a ferromagnet into graphene still remains a very challenging issue and it represents one of the main bottlenecks. Systematic investigations of the role of the tunnel barrier have been slow because of the difficulties of growing uniform, ultrathin insulating layers on graphene. Any irregularity during growth would favour the formation of pinholes. Alternative barriers and spin injection and detection methods should also be explored.

While in early devices spin injection was achieved using transparent contacts (Co/SLG),1281 a great improvement on both injection efficiency and spin lifetime was obtained with tunnelling contacts (Co/Insulator/SLG).1282 One of the main issues in conventional interfaces remains, however, the matching of conductances. For this purpose, interfaces of graphene with graphite intercalated with magnetic impurities or molecules (similarly to magnetic semiconductors1283,1284) seem viable and have to be explored more deeply. The spin valve1285 is the natural bench to test the efficiency of spin injection with graphene hybrids.1286 Further optimization of the ferromagnetic contacts for both spin injection and detection is also possible. Currently, oxide tunnel barriers, such as Al2O3 and MgO, are common choices, but new barrier materials and new technologies based on (single) atomic layer deposition of BN, and related materials like MoS2 and MoSe2, to make tunnel barriers of single or few atomic layer thicknesses, should be considered. Other concepts involving the spin Hall effect (SHE),234 which is the formation of spin-polarized current driven by spin–orbit coupling introduced by impurities, should also be explored;1287,1288 especially owing to the highly efficient spin current generation observed in materials such as Ta due to SHE.1289 Spin injection is currently performed and deeply investigated in both SLG and BLG,1290 and measurements on FLG and MLG up to 40 nm thick have been reported.1291–1293 Further experiments require unambiguous assessment of the potential of all of these and identifying, in each case, the phenomena that determine the experimentally observed spin injection/detection efficiency. It is also crucial to investigate the integration between the graphene channel and different materials as substrates.

The development of new device fabrication and innovative spin manipulation protocols would advance spin injection efficiency at the ferromagnet/graphene interfaces, data recording and spin information processing. More exciting is the design of novel spintronic devices at <10 nm scale. Here, components enter in the quantum regime, and graphene may exploit the crossover between classical and quantum features of such nano-elements.

6.3. Graphene spintronic devices for sensing

The long spin diffusion length in graphene could enable its use in spintronic based magnetic sensors. It was shown1294 that graphene has a large non-local spin current effect near the Dirac point up to RT, and at small magnetic fields (0.1 T). Graphene spintronics has thus the potential to revolutionize the development of magnetic sensors with sensitivity ranging from the nT to the pT ranges (for comparison, the Earth's magnetic field is in the μT range1295). The potential of lab-on-chip spintronic sensors for magnetic nanoparticles would also impact onsite drug delivery control or tumour disease fight medicine. Additionally, breakthroughs in the fields of MRAM and reconfigurable logic have a cross-disciplinary interest for ICT.1296 Intrinsic non-volatility of spintronics technologies presents a significant advantage in terms of power consumption. The power consumption reduction will benefit as much for tightening supplies of energy, as for the digital societal revolution of ever more demanding portable electronics. The radiation hardness of spintronics metal-based technologies1297 is also an advantage, in particular in aerospace. The coupling of spin information with optics could lead to spin information transmission by optical links. Finally, spintronics already mainly relies on low-dimensional quantum limits, and can be downscaled without increasing power consumption.1298

6.4. Graphene spin gating

Graphene's intrinsic features are ideal not only for downscaling of conventional devices, but also to demonstrate radically new ideas that will allow spin manipulation without the trade-off related to the reduction of spin coherence time. New concepts to be explored include tailoring spin degrees of freedom through magnetic proximity effects (magnetic gating), torque effect, the SHE, mechanical strain or molecular/atomic functionalization, all requiring the networking of interdisciplinary research communities.

In addition to ballistic transport and micron long mean free paths145 spin relaxation lengths can reach unrivalled values1280 opening new avenues for spin manipulation in lateral graphene spin devices. Several ways can be explored to manipulate spin polarized (and eventually pure spin) currents. To that end, once long lifetime of injected spin-polarized electrons has been achieved, the next challenges lies in the engineering of tuneable magnetism or spin gating (i.e. external action on the electron spin to implement spin-based information processing technologies). Such strategy could be explored by evaluating the (spin)-transport response of gated devices to magnetic moments induced in graphene by external means, such as hydrogenation, defects, strain engineering or other methods (a schematic of intentional and external perturbations is shown in Fig. 72) or structuring the lattice as a nanomesh.1299 By tackling these issues, one anticipates the eventual demonstration of external ways to control (through gates or possibly multi-gates) the propagation of spin currents, achieving operational reliability at RT and architectural compatibility with Si technologies.

image file: c4nr01600a-f72.tif
Fig. 72 (a) CMG using molecular adsorption or nanomesh patterning. (b) Mechanical strain applied on the graphene device.

Other types of proximity effects can also be harnessed, including effects on the electronic and spin properties induced by the interface with insulating magnetic materials, magnetic molecules, materials having a large spin–orbit coupling, ferroelectric materials. We refer, e.g., to the theoretical prediction of spin polarization induced by an interface with Europium oxide,228 or Yttrium Iron Garnet (YIG) and/or a material with strong spin orbit interaction.228 Such mechanism would not require any ferromagnetic metallic contact to inject spin-polarized electrons. Thus, it could be a way to circumvent the conductivity mismatch. Similarly, the exploration of the potential of graphene where the injected spin-polarized charges flows in close proximity of, and interacts with, other extrinsic spins (in localized or more extended charged states, located below or on top of the graphene) is very appealing.1300

6.5. Graphene qubits

By trapping an odd number of electrons on a dot one can create an electrically controlled localised spin state,1301 allowing the implementation of so-called spin qubits,1301 which can be used for quantum information processing.1301 Several proposals to use GQDs for quantum information processing have already been made,1298 based on the long spin memory of electrons in graphene (in particular, due to the absence of nuclear spin environment, a major problem for the use of III–V semiconductor dots for quantum information processing1302).

An additional possibility to create GQDs is related to the unique properties of BLGs. In BLGs, transverse electric fields can be used to open a gap. One can confine electrons in small regions of a BLG using a combination of top/bottom gates, and, then, operate the charging states of such QDs electrostatically. Further studies of gap control and electron confinement in gapped BLGs are needed. An intriguing challenge is to couple GQDs with other quantum systems. Amongst these, molecular nanomagnets (e.g. single molecular magnets, spin transition compounds) have shown considerable potential, due to the control of their quantum features at the molecular level.955 Moreover, the ability to control/functionalize the external shell of such molecules, allowing one to graft them on carbon surfaces, makes the realization of molecule-graphene hybrids feasible. The choice of substrate and spin state of the deposited molecule make hybrid carbon-based molecular architectures a promising platform to design novel spintronic devices, e.g. molecular spin valves in vertical geometry made by graphene sandwiched between a magnetic substrate and a magnetic molecule. Another possibility is to investigate – at higher T – the spin split of the energy band induced by magnetic molecules deposited on top of graphene. These are just few examples of the opportunities that molecular quantum spintronics based on low dimensional carbon materials may open.

6.6. Spintronics using other 2d crystals and heterostructures

In spin transfer torque (STT) experiments the orientation of a magnetic layer in a magnetic tunnel junction or spin valve can be modified using a spin-polarized current.1303 STT-based memory technology relies on magnetization switching induced by spin currents, and stands as a foreseen direction of innovation for spintronics. Demonstrations of perpendicular STT devices,1304 switching of ferromagnets using pure spin currents and spin Hall torque devices,1305 voltage-controlled switching of magnetic tunnel junctions (MTJs),1306 and novel spin logic architectures1307 have been reported using hybridization of magnetic materials and semiconductors, although their operational capability remains insufficient for the advent of practical (RT and upscalable) spin-based memory and logic technology technologies.

The portfolio of 2d materials offers many perspectives for the design of new interfaces and hybrid heterostructures using semiconducting 2d systems (MoS2, WS2, …) and magnetic materials (ferromagnetic metals, insulators) which could support the development of STT-based novel devices, circuits and architectures.

Some theoretical studies have suggested1308 that MoS2, normally a nonmagnetic, direct band gap semiconductor, acquires a net magnetic moment upon adsorption of specific transition metal atoms (such as Mn), as well as Si and Ge atoms [see e.g.ref. 1309]. This permits the diversification of the class of materials which can be combined and engineered for the implementation spin-based properties.

2d semiconducting materials may also allow the manipulation of other quantum degrees of freedom, such as valley degeneracy. This gives rise to the possibility of valleytronics,31 relying on the fact that the conduction bands of some materials have two or more minima at equal energies, but at different momentum space positions. To engineer a valleytronic device, the number of electrons per valley must be controlled to monitor valley polarization. In this respect, 1L-MoS2 is particularly promising because conduction and valence band edges have two energy-degenerate valleys at the corners of the first Brillouin zone. Ref. 1310 reported that optical pumping with circularly polarized light could achieve a valley polarization of 30% in 1L-MoS2, pointing towards optical valley control and valley-based electronic and optoelectronic applications in 1L-MoS2.

6.7. Theory and quantum simulation

Experimental studies need to be backed by advanced spin transport simulations, based on a multiscale approach combining ab initio calculations with tight-binding models for material parameterization (material structure, defects, deposited magnetic oxides, ad-atoms and adsorbed molecules) with semi-classical or quantum transport methods following spin diffusion. First principles and tight-binding calculations of SLG and BLG covered with vacancies, hydrogen ad-atoms, magnetic molecules and deposited magnetic oxides (in the presence of external electric fields) have the potential to clarify the effects of spin–orbit coupling and magnetic interactions on induced spin relaxation and spin–orbit transport. Moreover, transport lengths scales (mean free path, spin diffusion length, etc.) and transport tunability could be established by spin-dependent transport calculations through efficient quantum transport methodologies, providing guidance505 on spin relaxation mechanisms and gating efficiency of spin polarized currents in chemically/structurally modified devices.

6.8. Outlook

The aim is to reach the ultimate potential of graphene for spintronics, targeting efficient spin injection and detection, but also spin gating and spin manipulation. By investigating spin-based information technology methods supported by concrete demonstrations as well as modelling, the development of new ideas and implementations of graphene-based systems can be foreseen. The achievement of these tasks would pave the way towards all-graphene-based integrated MRAM and Spin information processing based circuits.

High priority has to be given to the evaluation of materials and device's architectures in view of future industrial applications. An exhaustive evaluation of feasibility among materials and their properties, fabrication techniques and designs need to be undertaken, in terms of quality, optimization, upscalability, reproducibility and RT operability.

Over the next 10-years (see Fig. 73) it is imperative to design, engineer and address large scale manufacturability of these highly advanced spintronic devices and architectures.

image file: c4nr01600a-f73.tif
Fig. 73 Spintronics timeline.

Using the intrinsic molecular and 2d properties of graphene, the high-impact perspective is to combine the unique long spin lifetime and high mobilities with new paradigms to manipulate the spin information locally, towards a new generation of active, CMOS-compatible, molecularly-engineered spintronic devices, with possible low-energy operation. The explored concepts could use magnetic proximity effects, the SHE, mechanical strain and molecular/atomic functionalization, as well as spin torque induced switching. Overcoming these challenges would pave the way to all-spin-based information processing technology, with capabilities ranging from replicating conventional electronics, to quantum information processing, and advanced functionalities.

The timeline for spintronic applications of GRMs is shown in Fig. 73. Timescales: 3–4 years: acquiring a comprehensive picture of spin relaxation mechanisms at RT in graphene, fundamental for controlling detrimental effects of material structural imperfections, environmental perturbations on spin transport. 5–9 years: demonstration of a spin gating functionality to prove that spin can be manipulated in some way (again for practical reasons, this has to be achieved at RT). Several routes need to be explored, such as chemical functionalization or strain fields, with which some tunability of the spin–orbit coupling and resulting spin transfer can be envisioned. Over 10 years: device-oriented integration, including the search for all-spin based architectures and co-integration of computing and data storage using the same material platform (i.e. wafer-scale graphene), targeting RT operation.

7. Photonics and optoelectronics

Graphene is emerging as a viable alternative to conventional optoelectronic, plasmonic and nanophotonic materials.995 It has decisive advantages, such as wavelength-independent absorption,784 tunable optical properties via electrostatic doping,1311 large charge-carrier concentrations,179,752,1312 low dissipation rates and high μ,99 and the ability to confine electromagnetic energy to unprecedented small volumes.1313Fig. 74 and Table 4 show some possible applications, drivers and an indication of when functional device prototypes could be expected. We envision breakthroughs in highly-integrated and high-speed graphene optoelectronics, long-wavelength photodetection and THz operation, ultrafast pixelated photodetection, broad-band tunable lasers, high efficiency photodetection and PV, and tunable optical metamaterials and plasmonic nano-optoelectronic circuits.
image file: c4nr01600a-f74.tif
Fig. 74 Graphene photonics applications timeline. The rectangles in the figure indicate the timeframe when functional device prototypes could be expected.
Table 4 Drivers and issues for implementation of graphene in photonics
Application Drivers Issue to be addressed
Photo-detector Fast increase of bandwidth chip to chip/intra-chip. Need to increase responsivity, which might require a new structure, plasmonics and/or doping control.
Higher bandwidth per wavelength not possible with IV or III–V detector in 2020.
High photoconductive gain.
Graphene photo-detector can increase bandwidth per wavelength to 640 GHz.
Graphene photo-detector can detect VIS, Infrared and THz. Modulator bandwidth has to follow suit.
Mode-locked semiconductor laser Bandwidth increase between core to core and core to memory requires optical interconnect with over 50 wavelengths, not possible with a laser array. Competing technologies: actively mode-locked lasers or external mode-lock lasers
GSA enables passively mode-locked semiconductor and fibre lasers, candidates for dense wavelength division multiplexer (D-WDM). Interconnect architecture should consume low power.
Solid-state mode-locked laser GSA can be simpler and cheaper and easy to integrate into the laser system. Need cost effective graphene transfer technology.
Tuneable mode-locked laser Wide spectral range of graphene is suitable for widely tuneable mode-locked laser. Need a cost effective graphene transfer technology.
Optical modulator Si operation bandwidth limit ∼50 GHz. Graphene is a good candidate without using complicated III–V epitaxial growth or bonding High quality graphene with low Rs is key for increasing bandwidth over 100 GHz.
Improve ION/IOFF.
Polarization controller Current polarization controlling devices are bulky and/or difficult to integrate. Need to improve controllability.
Graphene can realize compactness and integration of these devices.
Isolator Graphene can provide integrable and compact isolators on Si substrate, otherwise only possible with bulky magneto optical devices Decreasing magnetic field and processing are important to products

The dispersion relation remains quasi-linear up to ∼±4 eV from the Fermi energy.141 It is therefore possible to generate charge carriers in graphene by optically stimulating inter-band transitions over a wide energy spectrum, unmatched by any other material. This includes commercially important telecommunication wavelengths and the FIR/THz and the SWIR (short-wavelength infrared) and MIR (mid-infrared) regimes (see Fig. 75a).

image file: c4nr01600a-f75.tif
Fig. 75 Schematics of (a) wavelength-independent absorption, (b) broadband photodetection, (c) plasmon generation by a molecule or through a plasmon resonance.

The high μ enables ultrafast conversion of photons or plasmons to electrical currents or voltages (Fig. 75b). By integration with local gates, this process is actively tuneable and allows for sub-micron detection resolution and pixelization.

Graphene electronic and photonic devices can be fabricated using standard semiconductor technology, which facilitates monolithical integration into Si-based mass-production platforms. This is a decisive advantage over most other promising nanotechnologies.

At elevated (>1013 cm−2) carrier densities, graphene supports surface plasmons with unprecedented properties:1313 (i) Extreme confinement, 1–3 orders of magnitude smaller than the wavelength,1313 much smaller than the confinement of plasmons in noble metals in the considered spectral regime. (ii) The optical response of graphene plasmons is strongly dependent on the doping level, i.e. the Fermi energy relative to the Dirac point. This can be changed electrostatically, providing a tool for ultrafast electro-optical switching and modulation. (iii) Crystallinity and defect-free structures over large distances due to the strength of the carbon bond, in contrast to plasmonic metals, in which fabrication imperfections constitute a bottleneck in the performance of nano-metallic structures. (iv) Low losses, resulting in surface plasmon lifetimes reaching hundreds of optical cycles.1313

Electrostatically controlled Pauli-blocking of optical transitions and controlled damping of plasmon propagation enables the realization of ultra-high bandwidth electro-optical modulators,1314 optical switches, and similar devices.

Graphene is an excellent candidate for high-gain photodetection by employing the photogating effect.1315 Because of its very high μ and its 2d nature, its conductance is very sensitive to electrostatic perturbation by photogenerated carriers close to the surface.

The graphene properties that may appear to hinder its development for purely electronic devices, such as the absence of a band-gap, are not critical for photonics and optoelectronics. In fact, they can be beneficial, enabling ultra-wideband accessibility provided by the linear electronic dispersion, allowing efficient, gate controllable, e–h pair generation at all wavelengths, unlike any other semiconductor.

Based on these unique optoelectronic properties, a wide range of applications can be developed. Here, we summarize a selection of those for which we see technological breakthroughs in the near-future:

Highly-integrated graphene photonics. The compatibility of graphene with standard CMOS processes at wafer scale makes it a promising candidate for high data-rate (inter- and intra-chip) optical interconnects. Graphene might allow the realization of high-speed, compact-footprint electro-optical modulators, switches and PDs, integrated with Si waveguides or plasmonic circuits. The mechanical flexibility of graphene may also enable the integration with bendable substrates and plastic waveguides.

Long-wavelength light detection. Graphene enables light detection at wavelengths beyond the current limit set by the band gap of traditional semiconductors, opens up new applications in the FIR (THz) and SWIR and MIR regimes (e.g. bolometers and cameras), and has potential for ultrafast pixelated detection with ballistic transport of generated charge carriers.

Terahertz operation. This will enable products such as portable sensors for remote detection of dangerous agents, environmental monitoring or wireless communication links with transmission rates above 100 Gbit s−1.

High-efficiency photodetection. The target is the realization of highly efficient photocurrent generation by providing a gain mechanism where multiple charge carriers are created from one incident photon.

Plasmonics and metamaterials. Graphene surface plasmons as well as tailored metal nanostructures can be exploited to enhance and control the coupling between light and graphene. This will pave the way to ultra-fast optical switching, ultra-strong light-absorption, PVs, and single bio-molecule sensing.

Photonic integrated circuits. The target is to develop active optoelectronic devices and co-integrate these with passive optical components, such as waveguides,(de-)multiplexers and filters. Device functionalities comprise the switching and routing of light at ultra-fast speeds and at nano-scale dimensions, converting broadband incident light into detectable electrical signals.

Lasers. Exploiting the broadband saturable absorption properties of graphene, the aim is to realize mode-locked lasers (e.g., fiber, semiconductor, waveguide and solid state lasers) with broad tunability and bandwidth in the telecommunications and mid-IR range, compared to existing technologies.

The vision is to establish a new field of GRM photonics, sustained by the convergence and co-integration of GRM-based electronic and photonic components such as lasers, optical waveguides, cavities, modulators, photodetectors, and solar cells.

GRMs may be employed as active optoelectronic materials to achieve light–matter interaction, convert incident light energy into detectable electrical signals, and, vice versa, electrical signals may be used to modulate light and realize optical switches. For this purpose, GRMs should be integrated with established and mature technologies, such as dielectric (Si or plastic) waveguides, optical antennas, plasmonic structures (e.g. gratings or nanoparticles), metamaterials, QDs, etc. Graphene's constant optical absorption over a spectral range covering the THz to the UV allows light detection over a wavelength range superior to any other material. Combined with its high μ and Fermi velocity, this implies that devices operating in the hundreds of GHz range are feasible.

Despite graphene's absorption of 2.3% being large once its monoatomic thickness is considered, it is still necessary to increase this value to allow more efficient light–matter interaction and realize highly efficient optoelectronic devices. For this purpose, several routes can be pursued. One is the combination of graphene with plasmonic nanostructures,446 whereby the near-field enhancement due to localized surface plasmons can significantly increase the light absorption.446 In principle, structures can be designed to achieve 100% light absorption.446 Semiconducting nanoparticles of various shapes and forms can also be used to improve the quantum efficiency.1315 Light harvesting and concentration with these nanostructures into graphene leads to increased absorption and more efficient conversion of light into electrical signals, with an increase of quantum efficiency.

Another concept involves the integration of graphene with highly confined optical waveguides, such as Si-on-insulator (SOI) waveguides, widely used in highly-integrated Si photonics. As light propagates along the waveguide, it is absorbed along the length of the propagation and 100% light absorption might be possible.1316–1318 Graphene may also be inserted between two mirrors to form microcavity-integrated optoelectronic devices.1319 The incident light is reflected by the top and bottom mirrors and passes through the graphene multiple times. At the resonance condition, constructive interference enhances the optical field in the cavity, leading to enhanced light–matter interaction and strong optical absorption.

The exploitation of plasmons in graphene itself is raising interest.252,266,273,1320–1326 The resulting strong light–matter interaction can be further utilized to enhance detector performance as well as to enable radically new light sensing concepts.

The ever increasing demand for higher-bandwidth brings along the need for higher-bandwidth devices on the transmitting, as well as the receiving side, of the communications link. As large parts of the internet traffic are already transmitted optically, the need for high-speed modulators and photodetectors in the telecommunications wavelength range (1.3–1.55 μm) is ever more pressing. Moreover, optical interconnects are currently being introduced as a way to link computers to mobile devices, as well as ultra-high bandwidth links for inter- and even intra-chip communication. It was demonstrated that graphene photodetectors are capable of supporting bandwidths up to 262 GHz,1327 a huge value, but still far below the intrinsic limit, thus far estimated in the THz range.1328 The speed limit of graphene-based PDs needs to be established. Devices need to be optimized in terms of responsivity, by enhancement with plasmonics and quantum structures, as well as integrated into optical cavities and waveguides. Electro-optical modulators need to be realized and benchmarked in terms of speed and other parameters.

The FIR (THz) and MIR regions are fairly unexploited parts of the electromagnetic spectrum, and especially light detection is difficult as the THz photon energy is below the thermal energy. However, many interesting applications can be thought of, due to the non-ionizing and low-energy characteristics of THz and MIR radiation. Ranging from medical applications, such as cancer diagnostics, to security, such as explosive detection, since all materials have characteristic fingerprints in the THz/MIR region, a very wide application range is feasible. Further, active devices working in the THz/MIR range are crucial to convert very high frequency signals (in particular, the THz/MIR part of the solar radiation spectrum) into DC voltage – a feature which could eventually lead to self-powered devices. Having zero-band gap, graphene offers huge potential to outperform all available semiconductor technologies in the THz and MIR range, to reach THz operation frequencies, and to enable future wireless THz systems.

Graphene-based sensing and imaging in the SWIR is another interesting area that lies at the heart of safety and security applications in civil and military surveillance, night vision applications, automotive vision systems for driver safety, food and pharmaceutical inspection and environmental monitoring, just to name a few. The SWIR region is so far based on the use of III–V single crystalline semiconductors for sensing. Imaging has curtailed their monolithic integration to CMOS read-out integrated circuits (ROIC) and focal plane arrays (FPAs).1329 This results in SWIR imaging sensors at a 3-order-of-magnitude greater cost compared to visible-wavelength Si-based ones,1330 and pixel resolution limited to ∼1 MP (mega pixel) range,1330 as opposed to tens of MPs offered by current CMOS imaging sensors.

The 2d nature of graphene makes it also feasible to use CMOS-compatible processing techniques with CVD grown large-area graphene, in order to achieve highly-integrated arrays consisting of numerous individual devices. This may allow the fabrication of pixelated graphene-based cameras working over an ultra-wide spectral range, enabling image capture from the visible to the more unexplored THz/MIR range. Potential application areas include medical, automotive and security, such as tissue imaging, driver supporting head-up displays, and explosive/biological species detection.

Current solar cell technologies use only a rather small part of the solar spectrum due to their intrinsic band gap limiting the maximum detectable wavelength.1331 The absence of a band-gap in graphene translates into the absence of this maximum detectable wavelength limit and, combined with its constant absorption, solar energy over a much wider spectral length might be converted to electricity. Solar cells based on graphene, as well as combined with plasmonic and quantum nanostructures, thus need to be explored.

Graphene can be combined with other nanostructures, taking advantage of the strong light absorption in QDs and the 2-dimensionality and high μ, to merge these materials into a hybrid system for photodetection with high sensitivity.1315 Further, the integration with plasmonic metamaterials will enable a new class of optical switches for displays.

Surface enhanced Raman Spectroscopy (SERS) can in principle achieve signal enhancements of up to 1015.1332 Plasmonic nanostructures enhanced sensing may lead to detection limits on the single-molecule level. The near-field enhancement resulting from the combination of graphene with plasmonic nanostructures1333 will increase the signal sufficiently.1334 Such structures will also benefit from graphene's compatibility with biological species. Combined with graphene's single-electron charge sensitivity, we expect this technology to become a new platform for medical applications, not only providing enhancement at the single-molecule sensing level, but also being bio-compatible.

Integration of graphene into a cheap, flexible sensing platform based on plastics must be explored. It is envisaged to integrate and merge the aforementioned medical and THz/MIR sensors with plastic electronics to achieve a sensing-platform for wearable electronics, as well as low-cost one-time use sensors for use in developing countries.

7.1. Graphene saturable absorbers and related devices

Materials with nonlinear optical and electro-optical properties are needed in most photonic applications. Laser sources producing nano- to sub-ps pulses are key components in the portfolio of leading laser manufacturers. Regardless of wavelength, the majority of ultrafast laser systems use a mode-locking technique, where a nonlinear optical element, called a saturable absorber (SA), turns the continuous-wave output into a train of ultrafast optical pulses.1335 The key requirements are fast response time, strong nonlinearity, broad wavelength range, low optical losses, high power handling, low power consumption, low cost and ease of integration into an optical system. Currently, the dominant technology is based on semiconductor SA mirrors (SESAMs).1335,1336 However, these have a narrow tuning range, and require complex fabrication and packaging.1335,1353 The linear dispersion of the Dirac electrons in graphene offers an ideal solution: for any excitation there is always an e–h pair in resonance. The ultrafast carrier dynamics512,513,1337,1338 combined with large absorption171,990 and Pauli blocking, make graphene an ideal ultrabroadband, fast SA.1339,1353 Unlike SESAMs and CNTs,1340–1352 graphene does not require bandgap engineering or chirality/diameter control.1339,1353

Since the first demonstration in 20091353 (Fig. 76), the performance of ultrafast lasers mode-locked by graphene has improved significantly. E.g., the average output power has increased from a few mW1353 to over 1 W.1354 LPE1339,1353,1355–1359 CVD,1360,1361 carbon segregation,1362 MC1359,1363 have been used for graphene saturable absorber (GSA) fabrication. So far, GSAs have been demonstrated for pulse generation at 1,1369 1.2,1370 1.51339,1353,1360,1363,1366 and 2 μm.1364 The most common wavelength is ∼1.5 μm, not due to GSAs wavelength restriction, but because this is the standard wavelength of optical telecommunications. Ref. 1356 reported a widely tuneable fiber laser mode-locked with a GSA. The laser produces ps pulses in a tuning range 1525–1559 nm, demonstrating its “full-band” operation performance.

image file: c4nr01600a-f76.tif
Fig. 76 Graphene fiber laser.1353 WDM, wavelength division multiplexer; PC, polarization controller; EDF, erbium-doped fiber; ISO, isolator.995

Fiber lasers are attractive due to their efficient heat dissipation and alignment-free format.1365 GSAs have been used to mode-lock fiber lasers.1339,1353,1355–1357,1360,1363,1366 For fiber lasers, the simplest and most economical approach for GSA integration relies in sandwiching directly the GSA between two fiber connectors (Fig. 76) [see ref. 1356, 1360, 1363, 1366 and 1339]. Other options (e.g. evanescent-wave based integration1367) have also been demonstrated for high-power generation. Sub-200 fs pulses were achieved using a stretched-pulse design, where the cavity dispersion is balanced to stretch the pulse for the limitation of nonlinear effects.1355

Solid-state lasers are typically used for high-power output, as alternative to fiber lasers.1368 GSAs have also been demonstrated to mode-lock solid-state lasers.1369–1372 In this case, CVD graphene (>1 cm2) was directly transferred to a quartz substrate.1370Ref. 1370 reported 94 fs pulses with 230 mW output power. Another approach for GSA fabrication relies in spin-coating LPE graphene either on quartz or high-reflectivity mirrors. GSA can then be inserted into a solid-state cavity. This was used to achieve average power up to 1 W using a solid-state Nd:YVO4 laser,1372 with ∼1 μm output wavelength and ∼14 nJ energy.

SLG grown via CVD was also used in ultrafast vertical-external-cavity surface-emitting lasers (VECSELs).1361 A VECSEL consists of an external cavity, formed by high-reflection mirrors, and an output coupler, with typical cavity lengths of a few mm up to tens of cm.1335,1373 The gain chip generally contains a highly reflective bottom section to reflect the laser and pump light, an active semiconductor gain section, and an anti-reflective top layer.1335,1373,1374 VECSELs combine the advantages of semiconductor lasers, such as compact footprint (down to ∼3 mm cavity1375), with those of diode pumped solid-state lasers, such as low timing jitter,1376 excellent beam quality,1377 high average1377 and peak power1378,1379 VECSELs are mode-locked by SESAMs, epitaxially grown on lattice-matched semiconductor substrates.1335 However, SESAMs only offer a limited operation bandwidth (to date, the broadest tuning range of VECSELs mode-locked with SESAMs is 13.7 nm1380). Ref. 1361 controlled the electric field intensity in SLG by changing its absorption on a high-reflection mirror. The resulting SLG-based saturable absorber mirrors (GSAMs) have shown an unsaturated loss adjustable from 0 up to 10% and modulation depth up to 5%.1361 These enabled to mode-lock a VECSEL with a series of different gain chips over a 46 nm wavelength range (from 935 to 981 nm) with repetition rates up to 2.48 GHz, and 466 fs pulse duration. These results can lead to novel graphene-based ultrafast light sources to meet the wavelength range, repetition rate and pulse duration requirements for various applications (e.g. metrology, spectroscopy and data-communication).

Graphene is also promising for other photonic applications, such as optical limiters1381 and optical frequency converters.1381 Optical limiters are devices that have high transmittance for low incident light intensity and low transmittance for high intensity. There is a great interest in these for optical sensors and human eye protection,1381 as retinal damage can occur when intensities exceed a certain threshold.1381 Passive optical limiters, which use a nonlinear optical material, have the potential to be simple, compact and cheap.1381 However, so far no passive optical limiters have been able to protect eyes and other common sensors over the entire visible and NIR range.1381 Typical materials include semiconductors (e.g. ZnSe, InSb), organic molecules (e.g. phthalocyanines), liquid crystals and carbon-based materials (e.g. carbon-black dispersions, CNTs and fullerenes).1382 In graphene-based optical limiters the absorbed light energy converts into heat, creating bubbles and microplasmas,1382 which results in reduced transmission. Graphene dispersions can be used as wideband optical limiters covering visible and NIR. Broad optical limiting (at 532 and 1064 nm) by LPE graphene was reported for ns pulses.1382Ref. 1383 reported that FG dispersions could outperform C60 as an optical limiter.

Optical frequency converters are used to expand the wavelength accessibility of lasers (e.g., frequency doubling, parametric amplification and oscillation, and four-wave mixing).1381 Calculations suggest that nonlinear frequency generation in graphene (e.g. harmonics of input light) should be possible for sufficiently high external electric fields (>100 V cm−1).1384 Second-harmonic generation from a 150 fs laser at 800 nm was reported.1385 In addition, four-wave mixing to generate NIR tunable light was demonstrated using SLG and FLG.1386 Graphene's third-order susceptibility |χ3| was measured to be ∼10−7 e.s.u.,1386 up to one order of magnitude larger than CNTs.1386 Other features of graphene, such as the possibility of tuning the nonlinearity by changing N, and wavelength-independent nonlinear susceptibility1386 could be used for various photonic applications (e.g. optical imaging).

The main parameters of a pulsed laser are output power (or single pulse energy), output spectral coverage (e.g. operation wavelength, wavelength tuneability), pulse duration and repetition rate. The requirements for ultrafast laser are highly application-dependent. E.g., for fiber-optical communications, the operation wavelength is ∼1.5 μm as optical fibers have low loss and low dispersion around this spectral range.1387 For medical applications (e.g. laser surgery), the required laser operation wavelengths mainly depend on the peak absorption of different tissues.1388,1389 These range from MIR for minimally invasive surgery of skin cutting,1390 to UV for athermal photoablation.1391 For high-speed fiber-optical signal transmission and processing, high-repetition rate (>GHz) allows the signal to carry more data, while for industrial material processing (e.g. micro-machining), kHz pulses are commonly used to decrease the cumulative heating caused by multiple laser pulses.1392 In general, high output power, wide spectral coverage, short pulse width, and high repetition rate are desirable, because it is easy to reduce the output power (e.g. by attenuators), narrow the spectral coverage (e.g. by optical filters), broaden the pulse duration (e.g. by dispersive fibers) and decrease the repetition rate (e.g. by optical choppers), but not vice-versa. Furthermore, stability, cost, compactness and efficiency (e.g. electrical-to-optical or optical-to-optical efficiency) also are key for applications.

7.1.1. 2d crystals-based saturable absorbers. Other 2d crystals (i.e. MoS2, Bi2Te3, Bi2Se3) have shown ultrafast carrier dynamics.1393–1401 Driven by the development of GSA, recently 2d crystals such Bi2Te3,1402,1403 Bi2Se3,1404–1406 Sb2Te3,1407 MoS2[thin space (1/6-em)]1408 have been used as SA for ultrafast pulse generation. Tunable output results have been demonstrated.28 Thus far, 2d crystals-based ultrafast lasers only have been demonstrated on erbium-doped fiber lasers (EDFLs)1402,1404–1407 and thulium/holmium co-doped fiber laser.1403 However, it is expected that other 2d crystals will soon be employed on other lasers (e.g., ytterbium-doped fiber lasers – YDFL-, thulium-doped fiber lasers TDFL and solid-state and waveguide lasers as well).1058

Ref. 1409 reported resonant nonlinear optical susceptibilities at the edges of MoS2 flakes, allowing direct optical imaging of the atomic edges and boundaries of such 2d crystal. Indeed, the structural discontinuity at the edges and boundaries of 2d crystals, e.g., graphene and TMDs, leads to complex interplay between the atomic positions and the electronic structures.1409 Subsequently, the atomic edges and boundaries reconstruct structurally and electronically due to the translational symmetry breaking.1409Ref. 1409 developed a nonlinear optical imaging technique that allows rapid and all-optical determination of the crystal orientations of the MoS2 at a large scale. Moreover, zigzag nanoribbons showed a two-photon resonance at ∼0.8 eV, originating from the subband transitions from the valence bands to the isolated edge states of the Mo-zigzag edges.1409 These results pave the way for the exploitation of other 2d crystals and their use as sub-band gap SA. In view of large scale integration, 2d crystals produced by LPE,35,250,602,624,625,629 that can maximize edges250,622 during the ultracentrifugation process, could be an useful route to create these novel SAs.

7.1.2. Output power/pulse energy. Currently, solid-state lasers and fiber lasers are the most commonly used for high output power/pulse energy applications, mainly because they allow high-power pump.1368 Solid-state lasers are advantageous in terms of high pulse energy and peak power, as fiber lasers suffer from nonlinear effects. E.g., Watt-level ultrafast Ti:sapphire lasers and their low-repetition-rate (<kHz), high energy and high peak power amplifiers are widely used for academic research. The primary limitation of solid-state lasers to achieve high average power is thermo-optic effects,1410 such as thermal lensing.1411 Compared to solid-state rod and slab lasers, solid-state thin-disk designs significantly reduce thermal effects and nonlinearities, due to the pump configuration and small thickness (∼a few hundred μm) of the gain medium. This could be a solution to high average power and high energy pulses. GSAs could be used in thin-disk designs for this purpose. The main challenge is the large non-saturable loss of these SAs, which can be addressed by further devices.

Compared to solid-state laser, fiber lasers have a greater potential for high average power because of their better heat dissipation,1412 due to large surface- to-volume ratio. However, the nonlinear effects, enhanced by strong mode confinement and the long fiber required (∼10 m)1365, may distort the pulses, and restrain the maximum peak power. Large-mode-area fiber (e.g. photonic crystal fiber, PCF) based ultrafast lasers working in a dissipative solution regime have been demonstrated1413 for high average power ultrafast pulse generation with MW peak power (e.g. 11 W average power and 1.9 MW peak power from a SESAM mode-locked Yb-doped PCF laser1413). In principle, large-mode-area fiber lasers mode-locked with GSA may deliver better performances (e.g. higher average power, higher peak power, system simplicity). E.g., coating GSAs on the fiber surfaces to achieve evanescent-wave interaction can preserve the alignment-free waveguide format by removing the free-space components, necessary for traditional SA coupling. It is also possible to put GSAs inside the fiber (e.g. holes of PCFs). These integration strategies (i.e. graphene on the surfaces or inside the devices) can be applied to various lasers: waveguide (e.g. laser inscribed waveguide and polymer waveguide) and semiconductor (e.g. VECSELs and optically pumped semiconductor disk lasers) for high power/energy pulse generation.

For applications it is not necessary to generate high average power ultrafast output only using one oscillator, as external cavity processing can increase the output power. E.g., external amplification of graphene mode-locked lasers or coherent combination of various lasers could boost output power and energy.

7.1.3. Spectral coverage. The operation wavelength is an important parameter. In particular, a range of applications (e.g. ultrafast PL) require resonant excitation, thus ultrafast lasers covering a broad wavelength range are attractive. Wavelength tunable1346 or switchable lasers are another solution to access a broadband spectral range. Combination of wide-band gain materials (e.g. Ti:sapphire) and GSAs could provide novel broadband tuneable ultrafast sources to meet the requirement for a wideband range.

The output wavelength or tuning spectral range of a traditional laser will be ultimately constrained by the gain medium. E.g., Ti:sapphire typically only works between 0.65 and 1.1 μm.1414 Nonlinear effects (e.g. optical parametric generation and Raman scattering) can be used to broaden the spectral range. They can provide gain covering from UV to THz.

Nonlinear frequency conversion (e.g. harmonic frequency generation, parametric oscillation and amplification, four-wave mixing, supercontinuum generation) is also useful to expand the wavelength accessibility after the oscillator.

7.1.4. Pulse width. Shorter optical pulses can provide better temporal resolution and high-speeds (e.g. pulse widths of 200–400 fs can enable 1.28 TB s−1 optical communications1415). In general, solid-state lasers facilitate shorter pulse generation (e.g. 4.4-fs pulses from a Ti:sapphire laser oscillator1416), as the shortest pulse that fiber lasers can generate is typically limited by enhanced nonlinearity.1416 Indeed, so far, the shortest pulse duration for GSAs (94 fs from a Cr:forsterite laser1417) was achieved with solid-state lasers.1418 These could be shortened further by using broadband solid-state gain materials (e.g. Ti:sapphire). E.g., with wideband gain media and laser design optimization (e.g. dispersion management), graphene mode-locked lasers could generate pulses as short as those produced by any other SAs, but with reduced system complexity. One of the limitations to get shorter pulses is the medium narrow gain bandwidth. Nonlinear effect-based gain has much broader bandwidth, which also supports ultrafast pulse generation. E.g., the combination of broadband Raman gain and GSAs can enable shorter pulse duration than ever before.

External-cavity methods (e.g. nonlinear compression, or coherent combining) could also be used to generate shorter pulse down to a few optical cycles (a cycle is defined as the time needed for light to travel over a distance equal to the light wavelength) (e.g. 4.3-fs).1355

7.1.5. Repetition rate. The repetition rate is inversely determined by the cavity length.1335 This means that shorter cavities permit higher repetition rates, and vice versa. Pulsed lasing with short cavities generally requires high-gain materials, a low-loss cavity, and low-loss SAs. So far, multi-GHz pulse sources have been demonstrated for mode-locked semiconductor lasers (e.g. 50 GHz from a 3 mm SESAM mode-locked semiconductor laser1419), and compact solid-state lasers (e.g. 157 GHz from a 440 μm Nd:YVO4 laser mode-locked with SESAMs1420). Waveguide lasers also allow high repetition rate (e.g. ∼400 MHz). GSAs have low non-saturable losses, which makes them suitable for short cavity lasers. Coating graphene on surfaces/facets of the cavity components (e.g. fiber, waveguide, semiconductor, monolithic solid-state materials, or mirrors) could enable compact lasers with repetition rates up to hundreds of GHz.

Another option to push the output repetition rate is to exploit harmonic mode-locking.1421 This requires a complex design to achieve precisely equidistant pulses, as the fluctuations of the temporal positions of pulses from those in a perfectly periodic pulse train (also termed timing jitter1422) is detrimental for various applications, such as fiber-optic communication and optical sampling measurements.

7.1.6. Other considerations. Solid-state lasers are superior to other lasers for high pulse quality (e.g. smooth spectral profile and low chirp) ultrafast pulse generation, since other waveguide formats (e.g. fiber lasers) are subject to dispersion and enhanced nonlinear effects and have low pulse quality (e.g. spectral side-bands, high chirp, increasing or decreasing total pulse dispersion with signal propagation). Waveguide-based ultrafast lasers also suffer from birefringence.1423,1424 This leads to challenges in applications where polarization mode dispersion or birefringence splitting is critical for the desired responses from optical devices.1425 Birefringence can be eliminated by using polarization-maintaining fibers. Unfortunately, nonlinear polarization evolution based mode-locking (i.e. mode-locking using optical intensity dependent polarization direction rotation1426) cannot be applied to polarization-maintaining fibers (PMF), as the polarization in PMFs does not change with optical intensity. GSAs have potential for polarization-maintaining fiber lasers.

Fiber lasers and other alignment-free waveguide based lasers can offer excellent beam quality even when operated at high average power, because of reduced thermal effects.1427 In addition, they are compatible with fiber delivery, which offers flexibility in system design and use. In terms of fabrication cost, it is also inexpensive to fabricate fiber lasers to meet applications with low demand on pulse energy (∼1 nJ), polarization, emission bandwidth, pulse quality, etc., as most fiber devices are economically available due to their mass-production for fiber-optical communications. GSAs are interesting for this type of fiber lasers, as they can further decrease the fabrication costs and reduce system complexity compared to traditional SA technologies (e.g. SESAMs).

In order to increase the damage threshold, graphene-based optical limiters need to be developed on two fronts. The first is to grow on-demand graphene with the desired characteristics, while the other is to design new optical geometries that maximize the range of protection. E.g., the use of two focal planes offers new possibilities for the optimization of graphene-based optical limiting.

The aim will also be to enhance the frequency conversion effect. Graphene samples as large as tens of cm for ICT applications and as small as a few mm for microchip laser applications need to be routinely produced. The combination of growth capabilities, device design and assembly, is expected to result in products with superior performance.

7.2. Photodetectors

PDs measure photon flux or optical power by converting the absorbed photon energy into electrical current. They are widely used in a range of devices,1428 such as remote controls, televisions and DVD players. Most exploit the internal photoeffect,1429,1430 in which the absorption of photons results in carriers excited from the valence to the conduction band, outputting an electric current. The spectral bandwidth is typically limited by the absorption.1428 Graphene absorbs from the UV to THz.1431,1432 As a result, graphene-based photodetectors (GPD), see Fig. 77, could work over a much broader wavelength range.
image file: c4nr01600a-f77.tif
Fig. 77 Scheme of a MGM photodetector. Adapted from ref. 995.

The response time is ruled by μ. Graphene has huge μ, so it can be ultrafast.1428 Graphene's suitability for high-speed PDs was demonstrated in a communications link at 10 Gbit s−1 (ref. 1433).

In this section we summarize some key concepts as discussed in ref. 1434.

Many of the characteristics and unique capabilities of photodetection systems based on GRMs have been studied over the past few years, and a multitude of application areas have been addressed. Some of these have already reached a level of competitiveness with existing technologies. Various photodetection schemes and architectures have been proposed to date. The simplest configuration is the metal–graphene–metal (MGM) PD in which graphene is contacted with metal electrodes as source and drain.1429,1435,1436 Further, these PDs can be enhanced with plasmonic metal nanostructures,446 intrinsic plasmons,1437 p–n junctions in the graphene channel,1438,1439 and integration with waveguides,1318,1440,1441 and microcavities.1319,1442 Hybrid approaches employ semiconducting nano-particles, molecules, or plasmonic nano-systems as light absorbing materials, with graphene as the conduction channel, resulting in photoconductive-gain.1315,1443,1444 Devices have also been fabricated for detection of THz light,1445 where an antenna coupled to source and gate of the device excites plasma waves in the channel. Most importantly, graphene is compatible with the highly mature Si-based platform for electronics and photonics, making it a strong contender for low-cost and large-scale integration into optoelectronic networks and multi-pixel CMOS read-out circuits.

7.2.1. Figures of merit. Light impinging on a device, with a photon energy Eph and power Pin, corresponds to an incoming photon flux ϕin = Pin/Eph [s−1], and an absorbed photon flux ϕabs = ϕin·Aabs [s−1], with Aabs the absorbed fraction. The external quantum efficiency (EQE) is equal to the number of e–h pairs per second collected to produce the photocurrent Iph [C s−1], divided by the number of incident photons per second: EQE = (Iph/q)/ϕin, with q the electron charge. The internal quantum efficiency (IQE) is calculated in a similar way except that in this case the absorbed photon flux is considered: IQE = (Iph/q)/ϕabs. The responsivity of a PD is the photocurrent Iph divided by the incident power: Rph = Iph/Pin [A W−1; can also be given in V W−1 in the case of photovoltage], or in case the photo-induced voltage Vph is measured: Rv = Vph/Pin. E.g., in a PD with 100% EQE, Rph = 1 A W−1 for Eph = 1 eV. One figure of merit conventionally employed to compare mm and sub-mm detector performances is the noise equivalent power (NEP) [W Hz−1/2].1446 This is a function of noise and responsivity and is defined as the value of the root mean square (rms) input radiant signal power required to produce an rms output signal which is equal to a rms noise value with a signal to noise ratio of 1, usually expressed in units W per √Hz. Another typical figure of merit used to characterize the performance of a detector is the specific detectivity D* [cm Hz1/2 W−1, this unit is named after R. C. Jones, whereby 1 cm Hz1/2 W−1 = 1 Jones, in recognition of his work on sensitivity of radiation sensors1447], which is given by image file: c4nr01600a-t1.tif. Here, A is the area of the photosensitive region, and BW is the frequency bandwidth of the detector. For sensitized photoconductors the external quantum efficiency is defined as QE = ηtransηabs, where ηtrans is the charge transfer efficiency and ηabs is the light absorption efficiency. A second important parameter is the lifetime, τtr, of the charge residing in the particles, as the light-induced change in carrier density Δn scales linearly with it: Δn = τtr × QE × ϕin. Photoconductive detectors are often benchmarked by their photoconductive gain, Gph = (Iph/q)/(ϕinQE), i.e. the number of detected charge carriers per single incident photon. This can be quantified by the ratio of the lifetime of the trapped carriers over the drift transit time, τtransit, of the SLG charge carriers from source to drain: Gph = τtr/τtransit. τtransit is governed by the applied field, thus it is shorter for higher bias and mobility: τtransit = L2/(μVbias), with L the source–drain separation distance.
7.2.2. Physical mechanisms enabling photodetection. The conversion of absorbed photons into an electrical signal is the physical principle on which photodetection and other optoelectronic applications are based on. Photovoltaic (Fig. 78a), photo-thermoelectric (Fig. 78b), bolometric (Fig. 78c), and photogating effects, together with the Dyakonov–Shur (DS) (Fig. 78d) or plasma-wave-assisted mechanism are the mechanisms by which photodetection can be accomplished in graphene, see Fig. 78. Here we briefly discuss these mechanisms.
image file: c4nr01600a-f78.tif
Fig. 78 Schematic representation of the four different photocurrent generation mechanisms, (a) photovoltaic, (b) photothermoelectric, (c) bolometric and (d) plasma wave-assisted (Dyakonov–Shur). Adapted from ref. 1434. Photovoltaic effect. The photovoltaic photocurrent generation is based on the separation of photo-generated e–h pairs by built-in electric fields at junctions between negatively (n-type) and positively (p-type) doped regions of graphene or between differently-doped sections in general1430,1448,1449 (Fig. 78a). The same effect can be achieved by applying a source-drain bias voltage, producing an external electric field. However, this is generally avoided in the case of graphene, as it is a semi-metal, therefore generating a large dark current.

The doping that generates the built-in field can be introduced either by local chemical doping,1166 electrostatically, by the use of two (split) gates,1438,1448 or by taking advantage of the work-function difference between graphene and a contacting metal.1433,1448–1450 In the case of split gates, the doping can be tuned to be p or n, depending on the applied gate voltages, while in the case of graphene–metal junctions the doping in the contacted area is fixed. This is typically p-type for metals with a work function higher than the work function of intrinsic graphene (4.45 eV),1451 while the graphene channel can be p or n. The PV photocurrent direction depends only on the direction of the electric field, not on the overall doping level. Thus, it switches sign, when going from p–n to n–p, or from p–p+ to p+–p, where p+ means stronger p-type doping compared to p.

A lower bound on the intrinsic response time of SLG based MGM PDs was measured using an ultrafast optical correlation technique to be about 2 ps.1327 However, while the photogenerated carriers in graphene can have very high μ, the photodetection speed is not limited by the transit time of these carriers, but by the RC (resistance multiplied by capacitance) characteristics of the detector.23 Both response time and photo-detection efficiency depend on the ultrafast scattering processes. We note that e–e scattering513,1452,1453 can lead to the conversion of one high energy e–h pair into multiple e–h pairs of lower energy.512,1453,1454 This process, also denoted as carrier multiplication, can potentially enhance the overall photo-detection efficiency. Electron-phonon scattering,134,135,1455,1456 on the other hand, transfers electron energy to the phonons, which may lead to bolometric effects, as discussed in more detail below. Photo-thermoelectric effect. Hot-carrier-assisted transport can play an important role in graphene1438,1439 (Fig. 78b). Due to the large optical phonon energy scale in this material208,1456 (∼200 meV), hot carriers created by the radiation field can remain at a temperature Te (and thus energy kBTe) higher than that of the lattice for many ps. Final equilibration of the hot electrons and the lattice occurs via the slower scattering between charge carriers and acoustic phonons.1457,1458 These processes take place on a ns timescale,1457 although they experience a substantial speed-up attributed to disorder-assisted collisions.1459–1461

The photo-generated hot electrons can produce a photovoltage VPTE by the photo-thermoelectric (PTE) effect (Seebeck effect): VPTE = (S2S1TeVPTE = (S2S1T, where S1,2 [V/K] is the thermoelectric power (Seebeck coefficient) in the two graphene regions with different doping, and ΔTe is the Te difference between the regions. More generically, the photovoltage VPTE can be calculated by integrating the local electric field which is generated by an optically-generated temperature gradient, dTe/dx, together with a spatially varying Seebeck coefficient: EPTE = ∫S·∇Tdx. The PTE effect has been shown to dominate in graphene p–n junctions1438,1439 or in suspended graphene.1462 Because hot electrons, rather than lattice heating, generate the electronic response under these conditions, PTE graphene detectors can achieve high bandwidths, as in the case of PV detectors. The thermoelectric power S (also called thermopower) is related to the electrical conductivity σ by the Mott formula:1463

image file: c4nr01600a-t2.tif(1)
where kB is the Boltzmann constant, q is the electron charge, and the derivative of the electrical conductivity σ with respect to energy ε must be evaluated at the Fermi energy, i.e. at ε = εF = ħνFκF, with νF the Fermi velocity, image file: c4nr01600a-t4.tif the Fermi wave vector, and n the carrier density. The Mott formula was derived utilizing the Sommerfeld expansion,1463 thus it is valid only for kBTεF. For this condition, S can be calculated from the transport characteristics of the device, i.e. from the dependence of conductance on gate voltage.

The sensitivity of PDs based on the PTE effect might be enhanced due to efficient intraband e–e scattering, leading to higher Te.1464,1465 A typical Rph ∼ 10−3 A W−1 was reported for PTE graphene PDs,31 similar to those of PV graphene PDs. However, higher Rph ∼ 10−2 A W−1, was observed in suspended TLG p–n junctions.1462 In this case, the interaction of substrate phonons with graphene electrons is inhibited, removing an important electron-phonon decay channel. This shows that electron/substrate polar phonon scattering can play a role, especially at RT, and removing it by suspending the graphene sheet can increase the hot electron temperature, and thus the photocurrent. Bolometric effect. The bolometric effect is associated with the change in the transport conductance produced by heating associated with the incident photons (Fig. 78c). A bolometer measures the power of electromagnetic radiation by absorbing the incident radiation (dP) and reading out the resulting temperature increase dT.1446 Today, bolometers are mainly made of semiconductor1466 or superconductor absorptive materials,1446 and are widely used in the sub-millimeter (THz) wavelength range, where they are among the most sensitive detectors. The key parameters of a bolometer are the thermal resistance Rh = dT/dP, which ultimately defines its sensitivity, and the heat capacity Ch, which determines its response time τ = RhCh.1446 Graphene has small volume and low density of states, which results in low heat capacity Ch, thus a fast response. The cooling of electrons by acoustic phonons is inefficient, due to the small Fermi surface, and cooling by optical phonons requires very high Te (kBTe > 0.2 eV). Thus Rh is high giving rise to high bolometric sensitivity.

As this photodetection mechanism is based on a light-induced change in conductance instead of direct photocurrent generation, it requires an externally applied bias and can operate on homogeneous graphene, without the need to introduce a p–n junction. The conductance change induced by the incident light can be due to two mechanisms: (1) a change in the number of carriers contributing to the current; (2) a change in carrier mobility due to the associated temperature change. We note that (1) coincides with the photovoltaic effect, with the electric field generated by the external bias. Photogating effect. The photogating process is associated to a light-induced change of the carrier density Δn of a conductor, therefore of its σ. For graphene, this change in σ is given by the relation: Δσ = Δn·e·μ.1434 Typically, the photogating effect starts with light absorption inside sensitizing centers in the vicinity of the GRM, such as nano-particles or molecules.1434 This absorption process leads to the generation of e–h pairs. Next, one type of charge carrier is transferred to the GRM (e.g. by an intrinsic electric field) while the other type of charge carrier resides in the particles, molecules or traps. As long as these charges are trapped, the other type of charges in the GRM sheet are re-circulated, due to the charge conservation condition. We note that the fundamental difference to the bolometric effect, which is based on the change in μ due to heating, is that the photogating effect is based on a light-induced change in n.

G ph can be strongly enhanced by using a high-μ conductor (such as graphene) and by a long τtr. At the same time, a long τtr reduces the operation speed. Therefore these detectors can be used for lower temporal bandwidth, such as video imaging applications. Photoconductive detectors may exhibit a high dark current, so a proper assessment of the detector performance is not just provided by the responsivity, but rather by measuring the NEP or D*. Plasma-wave-assisted mechanism. Dyakonov and Shur (DS)1468,1469 proposed a photo-detection scheme, whereby a finite dc voltage is generated in response to an oscillating radiation field (Fig. 78d). This is based on the fact that a FET hosting a 2d electron gas can act as a cavity for plasma waves.1470 When these are weakly damped, i.e. when a plasma wave launched at the source can reach the drain in a time shorter than the momentum relaxation time τ, the detection of radiation exploits constructive interference of the plasma waves in the cavity, which results in a resonantly enhanced response. This is the so-called resonant regime of plasma-wave photodetection,1469 and can give rise to a signal that is 5–20 times stronger than the broadband non-resonant one. Broadband detection occurs when plasma waves are overdamped, i.e. when a plasma wave launched at the source decays before reaching the drain.1469

DS showed1469 that the photovoltage response of a 2d electron system in a FET, i.e. the electric potential difference between drain and source, contains a dc component even if the incoming field is ac, and thus provides rectification of the signal. This is particularly useful for the detection of THz radiation. Rectification occurs due to the non-linear response of the 2d electron gas in the FET channel and is unrelated to extrinsic rectification mechanisms due to e.g. Schottky barriers at contacts or other circuital elements that respond in a non-Ohmic manner. For the resonant regime, the dc photoresponse is characterized by peaks at odd multiples of the lowest plasma-wave frequency.1469 Resonant detection of THz radiation in graphene FET was theoretically discussed in ref. 1471 where the impact of hydrodynamic non-linearities strictly pertaining to the flow of the 2d electron gas in a graphene sheet was quantitatively addressed. Using the dispersion of plasma waves in graphene, it was concluded1471 that the typical linear device size required to operate in the THz spectral range varies in the interval ∼ 1–10 μm. RT THz detectors based on antenna-coupled GFETs, exploiting the DS mechanism have been demonstrated.1445,1472 In ref. 1445 the plasma waves excited by THz radiation were overdamped, thus the detectors did not operate in the resonant regime.77 The dependence of the photovoltage on carrier density in the FET channel displays also PTE contributions.1445,1472

7.2.3. Photoelectrical response in different devices. Metal–graphene–metal (MGM) PDs were the first class of graphene-based devices to be investigated.1429,1435,1436 The gapless nature of graphene allows them to operate over an unrivaled wavelength range, from MIR to UV, see Table 5. In early reports,1429,1435,1436 photocurrent was generated by local illumination of one of the MG interfaces of a back-gated GFET. The resulting current was attributed to the PV effect.1429,1435,1436 It was shown, that the field arises from charge transfer from the respective contact metal to graphene,109,1473 and can thus be adjusted by proper choice of the metal.1429,1474 It can be further enhanced by graphene doping via electrostatic gating.1429,1435,1436 Metal electrodes were also replaced by FeCl3-intercalated graphene-multi-layers to realize an all-graphene-based PD.1475 In addition to the PV effect, the PTE effect may also contribute.1476 Both mechanisms were disentangled in specifically designed experiments. Ref. 1477 reported wavelength and polarization dependent measurements of MGM-PDs. This allows to quantify and deterministically control the relative contributions of both PV and PTE effects, both contributing to the overall photoresponse. Ref. 1477 found that photovoltage maps at short and long wavelengths are very different, showing that for long wavelengths the photovoltage is mostly generated close to the contact edge where the p–n junction is located and vanishes where the flake ends underneath the contact. At the corners of the metal contact a hot spot of enhanced responsivity occurs due to curvature induced electric-field enhancement at the small radius of the corners.1477 On the other hand, at shorter wavelengths, the whole contact area contributes, with maxima at the contact edges.1477 Even far away from the pn-junction located at the edge of the metal contact, a photovoltage is produced.1477 This persists in the metal contact even when graphene is absent underneath.1477 This is due an increased absorption of the incident light at shorter wavelengths on the Au contact used in ref. 1477. This leads to a T rise on the metal contact, thus heating the pn-junction at the contact edge, producing a thermoelectric contribution to the photovoltage.1477 Photovoltage maps acquired at different polarization angles of the incident light, for a given location at the contact edge showed that the photovoltage consists of two contributions: one polarization dependent, and another polarization independent.1477 The polarization dependent contribution is attributed to the PV effect, due to polarization dependent interband optical excitations.1477 The photovoltage maximum occurs for perpendicular orientation of light polarization with respect to the potential barrier when the majority of photoexcited electrons are moving parallel to the junction. The photovoltage is higher for longer wavelengths, because light with longer wavelength, but same radiation power, can excite more electrons into the conduction band, resulting in a stronger photoresponse.1477
Table 5 PD Key performance parameters
Description Responsivity Detector type Bandwidth Wavelength IQE EQE Ref.
Graphene-metal junction 6.1 mA W−1 Photocurrent (PV/PTE) >40 GHz VIS, NIR 10% 0.5% 1433,1479
Graphene p–n junction 10 mA W−1 Photocurrent (PTE) VIS 35% 2.5% 1438,1439,1462
Graphene coupled to waveguide 0.13 A W−1 Photocurrent (PV/PTE) >20 GHz 1.3–2.75 μm 10% 10% 1318,1440,1441
Graphene-silicon heterojunction 0.435 A W−1 Schottky photodiode 1 kHz 0.2–1 μm 65% 1528
Biased graphene at RT 0.2 mA W−1 Bolometric VIS, IR 1450
Dual-gated BLG at low temperature 105 V W−1 Bolometric >1 GHz 10 μm 1492
Hybrid graphene-QD 108 A W−1 Phototransistor 100 Hz 0.3–2 μm ∼50% 25% 1315
Graphene with THz-antenna 1.2 V W−1 Overdamped plasma waves 1000 μm 1472
Graphene interdigitated THz antenna 5 nA W−1 Photovoltaic + photo-induced bolometric ∼20 GHz 2.5 THz 1507
SLG-TMD-SLG heterostructure 0.1 A W−1 Vertical photodiode <650 nm 30% 1529,1530
Biased MoS2 880 A W−1 Photoconductor 0.1 Hz <700 nm 1512
Graphene double-layer heterostructure >1 A W−1 Phototransistor 1 Hz 0.5–3.2 μm 1532
WSe2 p–n junction 16 mA W−1 P–n photodiode <750 nm 60% 3% 1533–1535
GaS nanosheet 19.1 A W−1 Photoconductor >10 Hz 0.25–0.5 μm 1536

A p–n junction is usually required to enhance the photoresponse. Such p–n junctions are often created close to the contacts, because of the difference in the work functions of metal and graphene,109,1478 or they can be created by local gates.1439 Responses at wavelengths of 0.514, 0.633, 1.5 and 2.4 μm have been reported.1433 Much broader spectral detection is expected because of the graphene ultra-wideband absorption. A GPD with a photoresponse of up to 40 GHz was reported.1479 The operating bandwidth is mainly limited by the time constant resulting from the device resistance, R, and capacitance, C. An RC-limited bandwidth ∼640 GHz was reported for graphene,1479 comparable to traditional photodetectors.1480 However, the maximum possible operating bandwidth is typically restricted by their transit time, the finite duration of the photogenerated current.1428 The transit-time-limited bandwidth could be >1500 GHz,1479 surpassing state-of-the-art PDs.

Although an external electric field can produce efficient photocurrent generation with an e–h separation efficiency >30%,1435 zero source–drain bias and dark current operations could be achieved by using the internal electric field formed near the metal electrode–graphene interfaces.1433,1480 However, the small effective area of the internal electric field could decrease the detection efficiency,1433,1480 as most of the generated e–h pairs would be out of the electric field, thus recombining, rather than being separated. The internal photocurrent efficiencies (15–30%1435) and external responsivities (generated electric current for a given input optical power) of ∼6 mA W−1 reported for basic GPDs with two contacts1433 are relatively low compared with existing PDs.1428 This is mainly due to limited optical absorption when only one SLG is used, short photocarrier lifetimes512 and small effective photodetection areas (∼200 nm (ref. 1479)). To date, responsivities ∼0.1 A W−1 for ultrafast waveguide integrated GPDs have been reported,1316–1318 while high-gain hybrid graphene/QD PDs have revealed responsivities up to 108 A W−1.1315 In order to achieve these high responsivities, it is important to optimize light absorption within the narrow regions around the interfaces or to absorb light in a different material.

One target is to increase the light absorption of graphene, and address the difficulty of extracting photoelectrons (only a small area of the p–n junction contributes to current generation), as well as addressing the absence of a photocurrent for the condition of uniform flood illumination on both contacts of the device. Unless the contacts are made of different materials, the voltage/current produced at both contacts will be of opposite polarity for symmetry reasons, resulting in zero net signal.1429,1433,1435

The optimization of the contacts needs to be pursued both theoretically and experimentally. Other possible ways of overcoming these restrictions comprise the use of plasmonic nanostructures placed near the contacts.446 Incident light, absorbed by such nanostructures, can be converted into plasmonic oscillations, with an enhancement of the local electric field.446 Such field enhancement, exactly in the area of the p–n junction formed in graphene, can result in a significant performance improvement. The role of the plasmonic nanostructures is to guide the incident electromagnetic energy directly to the p–n junction.

Nanostructures with geometries resonant at desired wavelengths need to be investigated for selective amplification, potentially allowing light filtering and detection, as well as polarization determination in a single device at high frequencies. The frequency performance can even be improved in comparison with traditional devices, as the plasmonic structures add only negligible contribution to the capacitance (fractions of fFs), but can significantly reduce contact resistance. Further optimization (e.g., making use of coupled or cascaded plasmon resonances1481,1482) could increase the photovoltage enhancement.

The advances in the fabrication of sandwich structures can help elucidate the process of light absorption and charge separation in vertical stacks, optimizing efficient light conversion into electricity. Composite 3d structures could be realized with some layers nanostructured, so to manipulate the working range of the device.

The low optical absorption and the narrow photocurrent generation region do not impose serious restrictions on the usability of graphene for photodetection. A number of concepts can be pursued to overcome these issues, and responsivities comparable to those of semiconductor-based devices can be reached. However, graphene's constant optical absorption over a broad spectral range allows light detection over a wavelength range superior to any other material.

The photothermoelectric effect, which exploits the conversion of photon energy into heat and then electric signal,1428 may play an important part in photocurrent generation in graphene devices.1435,1476 In particular, photocurrents assisted by hot-carriers are enhanced in the vicinity of pn-junctions leading to strongly enhanced responsivities.1439 The efficiency of hot-carrier generation depends on the competition of different relaxation pathways and can be greatly enhanced when photoexcited carriers do not lose energy as heat, but instead transfer their excess energy into the production of additional hot carriers or e–h pairs via carrier–carrier scattering. Pump-probe measurements have shown that in graphene the latter process is very efficient.512,513,1464 These results indicate that the production of secondary e–h pairs dominates the ultrafast energy relaxation of photoexcited carriers, prevailing over optical phonon emission in a wide range of photon wavelengths.1464 This leads to the production of secondary hot electrons, originating from the conduction band. As hot electrons in graphene can drive currents, multiple hot-carrier generation makes graphene promising for highly efficient broadband extraction of light energy into electronic degrees of freedom, enabling high-efficiency optoelectronic applications. Employing these unique properties, the development of novel designs for further optimization of thermoelectric light harvesting and detection may be possible. In particular, the efficient conversion from photons to hot carriers, a unique aspect of graphene due to the absence of a bandgap and strong carrier–carrier interactions, should be exploited.

Another approach to increase the responsivity is to integrate a GPD a highly confined optical waveguide and increase the optical absorption, Fig. 79.1316–1318 NIR light is coupled to a Si waveguide, which is embedded in SiO2. A graphene sheet is located on top of the Si waveguide and there is a thin isolation oxide layer in-between. The fundamental TM (transverse magnetic) mode would get almost completely absorbed in the graphene, as the light propagates along the waveguide. The local field gradient at the metal/graphene interface drives a photocurrent towards the ground leads. The metal electrodes may also be replaced by poly-Si to reduce optical mode damping. The expected foot-print of such device is ∼30 μm2, comparable to state-of-the-art Ge PDs, nowadays used in highly-integrated Si photonics for on-chip interconnects. The simple device geometry, less complex processing and compatibility with Si technology could make graphene an alternative to other semiconductors such as Ge, even for large-scale integrated photonics. Apart from its superior physical properties, graphene's strength is its 2d character, which makes it compatible with standard semiconductor technology and allows for monolithic or hybrid integration with other materials. Graphene sheets can be transferred to arbitrary substrates (including flexible), etched into any shape, integrated with waveguides or cavities, electrically contacted, etc.

image file: c4nr01600a-f79.tif
Fig. 79 Schematic of an integrated waveguide GPD.

All these aspects lead us to consider graphene as a promising new material for photonic integrated circuits,1483 particularly for high data-rate optical interconnects. Graphene will allow the realization of high-speed, compact-footprint PDs integrated with Si waveguides or plasmonic circuits. The mechanical flexibility of graphene may also enable the integration with bendable substrates and low-cost plastic waveguides. Photonic integrated circuits are based on the convergence and co-integration of passive optical components (waveguides, (de-)multiplexers, filters, etc.) with modulators, switches and PDs. State-of-the-art monolithic PDs are based on Ge or III–V semiconductors, but their responsivity ranges are spectrally limited. Graphene, without such a limit due to the gapless dispersion of the Dirac electrons, could thus pave the way for broadband photodetection, from the O-band (1260–1360 nm) to the U-band (1625–1675 nm), and beyond (e.g. 850 nm). The expected foot-prints of the devices are even smaller than those of state-of-the-art Ge devices used in Si photonics. The simple device geometry, less complex processing, compatibility with CMOS technology, and broadband and ultra-fast operation could make graphene an alternative to other materials, even for large-scale integrated photonics.

The 2d character of graphene and the possibility of large-area fabrication allows up-scaling from single devices into arrays. This could enable a pixelated, camera-like structure, where each pixel may be read out individually. Co-integration with Si transistors as amplifying elements is envisaged. The goal is to achieve gated PD arrays with strongly enhanced responsivity. The energy spectrum and optical properties of graphene can be modified through an electrostatic field. This can be utilized to realize a waveguide-integrated electro-optical modulators at 1550 nm, based on optical absorption for photon energies 2EF < Eph (Eph is the photon energy, EF denotes the Fermi energy) and absorption suppression for 2EF > Eph due to Pauli blocking.1339 Graphene-based modulators can be realized on a much smaller foot-print than devices using semiconductor materials for electro-absorption or electro-refraction by the Pockels, Kerr1484 and Franz–Keldysh effects.1485,1486

Another target will be the integration of graphene electro-absorption modulators into Si waveguide ring resonators and Fabry–Perot cavities with grating mirrors, to further decrease the foot-print. This will lead to higher modulation depth and speed than previously reported.

When it comes to high-demand applications, requiring photon detection at very low levels, even approaching single photon detection, photodiodes cannot be chosen, because their quantum efficiency is limited to unity, accounting for 1 carrier per photon. Therefore, the electrical signal suffers from read-out electronic noise, which overwhelms the photocurrent at low intensities. In view of this constraint, gain has been sought in PDs: a mechanism that can provide multiple electrical carriers per single incident photon.

Photoconductive detectors are based on the effect of carrier recirculation,1315 providing Ohmic contacts to the semiconductor and appropriate sensitizing centres that can prolong the carrier lifetime. Photoconductive gain is then possible, with a value given by the ratio of the carrier lifetime over the transit time. Photoconductive detectors have regained attention with the advent of colloidal QDs PDs.1487 Photoconductive gain ∼103–105 was reported for PbS, ZnO or CdSe QDs,1487–1489 limited, however, by the low μ of QD films. Phototransistors have also been proposed to provide for photoconductive gain.1488 A FET-like phototransistor was reported with single-photon counting potential, based on a 2DEG channel and QDs as the photosensitive gate, in an epitaxially grown structure. The 2DEG was formed in an AlGaAs substrate by cooling the device to 4 K and it was critical in achieving high sensitivity due to the high μ that can be achieved in a 2DEG.

Graphene is an excellent candidate for high-gain PDs because of the high μ, and since it is 2d. Thus, its conductance is very sensitive to electrostatic perturbation by photogenerated carriers close to the surface. This makes it a particularly promising material for high gain PDs by photogating. Additionally, graphene is a very thin, flexible and durable, and can be fabricated in a large scale and easily deposited on Si, offering integration into standard ICs. Thus, the demonstration of photodetection gain with graphene would be the basis for a plethora of applications, such as integrated optoelectronic circuits, biomedical imaging, remote sensing, optical communications, and quantum information technology.

Ultra-sensitive PDs and image sensor arrays for visible and IR imaging have been developed,1315 based on sensitized graphene: a film of semiconducting particles or QDs is deposited on the graphene sheet,1315 see Fig. 80. Colloidal QDs offer high absorption and bandgap tuneability from UV to SWIR,1488 and could be employed as a vehicle to demonstrate the potential of this technology. Through sensitization, strong absorption ∼50–100% can be achieved.1315 The detection mechanism is based on photogating,1490 where light induced trapped charges in the QD change the resistance of graphene. The photoconductive gain associated to this mechanism can be quantified by the ratio of the lifetime of the trapped carriers in the QD over the drift time of the charge carriers in graphene. A photoconductive gain ∼108 and responsivity ∼107 A W−1 can be achieved,1315 thanks to the high μ of the graphene channel, and the long carrier lifetime of the photogenerated carriers in the QD layer.1315 In combination with the high quantum efficiency >25% and low noise-equivalent-power ∼10−17 W, this hybrid QD-GPD is an excellent platform for large-scale sensitive SWIR detection.

image file: c4nr01600a-f80.tif
Fig. 80 Schematic of GQD hybrid phototransistor, in which a graphene flake is deposited onto Si/SiO2 coated with PbS QDs. Incident photons create e–h pairs in the PbS QDs. [Adapted from ref. 1315].

Research needs to be extended to the mid-IR, targeting broadband detection in the wavelength range 3–30 μm. Surface plasmon nanostructures functioning as optical antennas1491 could be used to couple the long-wavelength radiation to the graphene sheets. In the simplest case, a bow-tie antenna concentrates the radiation energy to a confined region of sub-wavelength dimension.1445 Asymmetric contacts1433 or, alternatively, a potential gradient produced by band profile engineering can be used to separate the carriers. Antenna arrays will allow scaling the device dimensions to large areas. Due to the long wavelength of the radiation, the lithography requirements are relaxed, which will allow cm2-sized detector arrays/cameras by optical lithography. More advanced plasmonic geometries may be used to further concentrate the radiation at the metal/graphene interfaces. Bolometers. Bolometric responses have been measured on a biased graphene sheet at RT.1437,1492Ref. 1437 reported that two mechanisms result in bolometric photocurrents with opposite sign. Namely, the T dependence of μ leads to a reduction of the conductance, while photo-induced excess carriers induce an enhancement of the conductance. By changing EF in graphene, one can control which mechanism dominates.1437 Near the Dirac point, where the carrier density is lowest, PV effects dominate, while far away PTE effects dominate.1437

The relative contributions of the two mechanisms depend on the nature of the device, such as the presence and nature of the substrate (i.e. polar versus non-polar), the quality of the sample, as expressed by μ, and the experimental conditions, such as the magnitude of applied bias and T. E.g., it was shown1450 that in a partially supported (on SiO2) and partially suspended sample, thermal effects dominate in the suspended part, and photovoltaic effects the supported one.

The weak T dependence of the electrical resistance in graphene,1493 however, imposes a major challenge for reading dT through electrical transport measurements, and for this reason a small responsivity of 0.2 mA W−1 was measured for RT graphene bolometric detectors.1437 To solve this problem, a dual-gated BLG device with optically transparent top gate was used1492 to open a band gap at the Dirac point127,1145 and obtain Te-dependent resistance.127 The device was measured under MIR illumination (10.6 μm) using a four-terminal configuration, and the photoresponse was identified to be bolometric.1492 The thermal resistance had a RT−3.45 dependence, which approximately agrees with the theoretically expected T−3 dependence for (disorder-free) phonon cooling.1494,1495 The detector exhibits excellent performance: NEP ∼33 fW per √Hz at T = 5 K (several times lower than commercial Si or superconducting bolometers) and an intrinsic bandwidth >1 GHz (3–5 orders of magnitude higher). Another approach for implementing a T-dependent graphene resistance is to drive the electronic system into the strong localization regime by adding disorder. This was achieved at low T by using defective graphene films.1496

Ref. 1497 reported a graphene-superconductor tunnel junction bolometer. When biased within the superconducting gap, the junction resistance is dominated by tunnelling of the thermally excited quasi-particles. The increase of Te in graphene under illumination decreases the tunnelling resistance, which can be probed by a small DC current. For DC signals, the contacts are highly resistive due to the superconducting gap, whereas for high-frequency (RF to optical frequencies) signals, the contact has low impedance due to the contact capacitance, therefore the graphene resistance can be matched with that of an antenna. Although the device in ref. 1497 was measured with RF waves, the concept can also be applied for optical detection, see ref. 1498 for a recent review. Long-wavelength photodetectors. Applications of long-wavelength radiation are manifold. The term “long-wavelength” is used to refer to the FIR (THz) and MIR, i.e. the wavelength range 3–1000 μm. Many chemical agents, explosives or narcotics feature spectral fingerprints in this range.1499 Optoelectronic devices operating in the THz and MIR may be employed in homeland-security-related applications, e.g. in security systems at airports monitoring dangerous substances, but also in environmental-related applications, e.g. gas spectrometers for air-quality control, whereby traces of threat chemicals have to be detected against a spectral background. At the same time, the MIR spectral range is as well fairly unexploited, with many interesting imaging and sensing applications in medicine and security.1500

The most sensitive detectors of THz radiation require deep cryogenic cooling: superconducting hot-electron bolometers1501 reach NEPs as low as a few fw/Hz1/2 (ref. 1501) with modulation bandwidths up to the GHz.1501 On the other hand, RT THz detectors are either not very sensitive,1501 or extremely slow,1501 or operate well only below 1 THz.1501

The situation becomes even more complex when technologies that can be produced into matrix arrays are considered. Commercial THz or MIR FPAs are based on microbolometer elements, and provide moderate sensitivities and response speeds (NEPs of few pW/Hz1/2 per pixel and ∼1 kHz bandwidth near 1 THz1501). Recently, Fast electronic devices, based on the gate-modulation of the conductance channel by the incoming radiation (often called plasma-wave detectors1469,1502,1503), have been realized in both HEMT1504 and CMOS1505 architectures and show NEPs already competitive with microbolometers and response times up to the MHz range.1554 This technology is also scalable to even larger arrays (a 1024 pixel camera was demonstrated at 0.7 THz1506). For what concerns the mid-IR, narrow-gap semiconductors, currently used in photodiode detectors, still suffer from serious problems in controlling the material quality and poor integration capabilities.

GFETs can be exploited to considerably advance the present state-of-the-art of plasma-wave detectors, see Fig. 81 (ref. 1445), taking advantage of the high μ.1445 This would allow improved performance in terms of responsivity and bandwidth, and operation limits well above 1 THz. Even more appealing is the possibility to reach a regime in which stationary plasma modes are excited in the transistor channel, leading to resonant, gate tuneable, enhancement of the responsivity. This also means that the detected radiation wavelength can be controlled by the gate voltage, hence leading to a novel passive spectroscopy system, which is presently not available in the THz.1472

image file: c4nr01600a-f81.tif
Fig. 81 Antenna-coupled GFET-terahertz detector. The terahertz radiation is focused by off-axis parabolic mirrors. Adapted from ref. 1445.

Graphene-based broadband THz detectors working at RT were shown1507 with Rv ∼ 5 nA W−1 and fast response rise times (∼50 ps) at 2.5 THz, which makes them already suitable for timing applications. The device employed a log-periodic THz antenna,1507 with a central interdigitated comb providing the electrical contact to the SLG.

PTE graphene THz detectors, operating at 2.5 THz and based on exfoliated SLG with a pattern of asymmetric metal contacts were also reported.1508 In ref. 1508, the generated photoexcited carriers rapidly thermalized due to the strong electron–electron interactions, while losing energy to the lattice more slowly. The Te gradient drives electron diffusion, and asymmetry, due to dissimilar contact metals, produces a net current via the PTE.

Finally, with the development of large-area growth techniques, multi-pixel camera-like arrays are feasible and potentially relatively low-cost. The long-term objective is the realization of a graphene THz camera capable of performing passive imaging and providing spectroscopic information. 2d crystals and hybrids. Although the bulk properties of layered TMDs have been investigated for many decades,111 their 2d forms have recently attracted much attention due to their potential applications in optoelectronics.1509 The properties of TMDs depend on N. E.g., MoS2 has a bulk indirect band gap of 1.3 eV, which becomes a direct band gap of ∼1.8 eV in 1L-MoS2.1510 This changes the optical properties, such as absorption spectrum and PL. Ref. 378 showed a 1000-fold enhancement of the luminescence quantum yield for 1L compared to bulk MoS2. Valley polarization controlled by optical pumping was also demonstrated in 1L-MoS2,1310 and polarization was maintained for longer than 1 ns.1511

Compared to classical direct band gap semiconductors, TMDs can also offer additional advantages in the context of optoelectronics because of their transparency, mechanical flexibility and easy processing. This suggests the possibility of photovoltaic and photodetection applications in MoS2 and other semiconducting TMDs. PDs made of 1L-MoS2 displayed high external photoresponsivity, i.e. the measure of the electrical output per optical input, (over 880 A W−1 at 561 nm1512) and a stable photoswitching time of 50 ms.1513 Moreover, the ability to tune the bandgap by varying N allows the detection of light at different wavelengths.1514 The strong photothermoelectric effect measured at metal-MoS2 junctions1515 can also be useful for applications in thermopower generation.1515

One important use of layered semiconductors such as MoS2 could be in making novel heterojunction devices. E.g., a metal–semiconductor–metal heterojunction photodetector, made of MoS2 and amorphous silicon (a-Si), with rise and fall times ∼0.3 ms was demonstrated.1516 The transient response of MoS2/a-Si hybrid PD does not show persistent (residual) photoconductivity, thus making this heterojunction roughly 10 times faster with respect to a-Si devices.1516,1517 A photoresponsivity ∼210 mA W−1 was measured for green light, the wavelength used in commercial imaging systems, which is 2–4 times larger than that of a-Si.

MoS2 is a good illustration of the wide range of optical properties offered by 2d crystals. These complement those of graphene in many ways: graphene is useful for broadband, ultrafast technologies, whereas semiconducting TMDs are advantageous for applications requiring strong light absorption1529 and electroluminescence.1518 A promising approach to combine these properties and create multi-functional, high performance materials consists of assembling graphene with other 2d crystals in heterogeneous stacks and hybrid devices.1519 This new device architecture has lead to novel transistor concepts based on vertical transport between two graphene electrodes, separated by h-BN, MoS2 or WS2 layers.106,1520,1529 The same hybrid devices have shown potential for photovoltaic applications with a photoresponsivity above 0.1 A W−1 and EQE >30%.