Novel charm of 2D materials engineering in memristor: when electronics encounter layered morphology

Abstract

The family of two-dimensional (2D) materials composed of atomically thin layers connected via van der Waals interactions has attracted much curiosity due to a variety of intriguing physical, optical, and electrical characteristics. The significance of analyzing statistics on electrical devices and circuits based on 2D materials is seldom underestimated. Certain requirements must be met to deliver scientific knowledge that is beneficial in the field of 2D electronics: synthesis and fabrication must occur at the wafer level, variations in morphology and lattice alterations must be visible and statistically verified, and device dimensions must be appropriate. The authors discussed the most recent significant concerns of 2D materials in the provided prose and attempted to highlight the prerequisites for synthesis, yield, and mechanism behind device-to-device variability, reliability, and durability benchmarking under memristors characteristics; they also indexed some useful approaches that have already been reported to be advantageous in large-scale production. Commercial applications, on the other hand, will necessitate further effort.

---

Ye Zhou

Prof. Ye Zhou is a Fellow of the Royal Society of Chemistry (FRSC) and group leader at the Institute for Advanced Study, Shenzhen University. His research interests include flexible and printed electronics, organic/inorganic semiconductors, surface and interface physics, and nano-scale devices for technological applications, such as logic circuits, data storage, photonics and sensors.

---

1. Introduction

High carrier mobility, an adequate bandgap, strong light–matter interaction with a big light absorption coefficient, extremely large exciton binding energy, significant spin splitting, and polarised light emission are intriguing features of the two-dimensional (2D) materials. Using the remarkable properties of 2D layered morphology in addition to metal-oxide-based semiconductor technologies to develop enhanced solid-state electronic devices is a novel idea that could help Moore's law be extended.^1,2 For future applications, researchers in the area of 2D nano- and microelectronics must meet the following criteria. All synthesis and device fabrication process must be on the wafer level, the morphology, thickness fluctuations, and lattice distortions in 2D materials should be accurately defined, and actual statistics must be provided.³ Moreover, the device size should be compact enough to meet the implementation criteria. Yield, variability (cycle to cycle and/or device-to-device), reliability, and stability must all be covered, as well as explanations of the important failure configurations of every particular device should be given.

Memristors can perform neuromorphic computing to mimic the functions of the human brain. The human brain is a highly efficient and perfect information processing system, including around 10¹¹ neurons and 10¹⁵ synapses.⁴ Memristors can mimic biological brain networks in operation.⁵ Even the most sophisticated organizations face significant challenges in nanostructured device production of 2D materials on a wide scale while implementing sustainable methods and achieving remarkable performance.^6–9 North America is the most stable market for memristors. The annual global market development growth rate of memristors will be approximately 80% in 2024 and USD 13.5 billion in 2027. It all comes down to the huge investment in R&D by local market enterprises, many of which are headquartered in the United States.¹⁰

In 1971, Leon Chua realized from symmetry arguments that there should be a fourth fundamental element, which he called a memristor that could be predicted as the fourth important element in the world and resistor, capacitor, and inductor, which he named memristor (short for memory resistor).¹¹ Later hypothetical demonstrations of the presence of tremendous energies in a memristor have been directed to their implementation in an electrical device.¹² Following that, Williams and his colleagues (2008) at Hewlett-Packard Laboratories initially asserted that the performance of a memristive could be empirically achieved in a real device using the Pt/TiO₂/Pt structure and presented the solid-state memristor.¹³ The synaptic connection determines signal transmission efficiency in biological systems from one neuron to another. It can be modified through purposeful stimuli or learning process. Neuroplasticity leads to variations in synaptic weight (pre-synaptic and post-synaptic cells), and synaptic plasticity also includes the following functions (i) short-term plasticity (STP) which has been proven to be linked to cognitive functions connected with spatiotemporal information, and (ii) long term plasticity (LTP) which on the other hand is the brain system for memory and learning. (iii) Moreover, STDP is the Hebbian theory's learning rule, which includes asymmetric and symmetric STDP demonstrating Hebbian and anti-Hebbian learning. Lu and his colleagues presented a hybrid nanostructured (metal oxide semiconductor memristor) device in 2010, a neuromorphic computing synapse. Yan and his colleagues (2019) used memristors to mimic several synaptic functions, e.g., (i) LTD, (ii) LTP, (iii) PPF, (iv) and STDP.^14,15 In certain ways, memristors are making progress in the field of brain neurobionics.

There are many advanced non-volatile memories like resistive random-access memory (RRAM),^3,19–21 phase-change memory (PCM),^22–24 and magnetic random-access memory (MRAM).^25,26 These huge and permanent data storage memories showed exceptional rewards.^27–29 Among the various reference devices above, we discuss some special characteristics of memristors. Memristors can switch along with various resistance states. During resistive switching, the speeds can be on the sub nanoscale dimension with high endurance, which leads to low energy consumption.^28,30 The memristor's resistive switching performance is determined by its previous resistive switching states. This property can be used in neuromorphic synaptic system applications for artificial intelligence. Furthermore, the activity of biological synapses to allow the programming of time contingent signals among neural networks is crucial for the validation of spiking neural networks intended for consistent processes.³¹ The synapses can be realized utilizing several designs, including the complementary metal-oxide-semiconductor transistor³² and memristors.³³ Of these, memristors are considered the most promising applicants in neuromorphic engineering due to their inherent competence to recall the historical information previously sent in the form of electrical signals.

2D resources have a layered morphology, and the robust covalent bond atomic level connects the thin layers. Moreover, weak van der Waals forces are present among layers. These forces offer a perfect chance for intercalation,³⁴ which can induce potential application in diverse technologies, as their atomic thickness might permit scaled devices.^35,36 By taking appealing advantage of the above-mentioned significant characteristics of 2D materials for the progressive manufacture of solid-state electronic devices, it is a pretty good tactic that would offer us a key solution to continue Moore's law.^1,37 Woong Huh and his co-workers³⁸ have reviewed and described a variety of two-dimensional materials with various memristive resistive switching mechanisms. In the field of new materials which have been used are graphene, molybdenum disulfide (MoS₂), tungsten disulfide (WS₂), tantalum disulfide (TaS₂), h-BN, etc. According to their properties, we classified them into three categories, metallic segment (graphene), semiconducting segment (MoTe₂, WSe₂, MoS₂, and WS₂), and insulating phase (h-BN), as shown in Fig. 1a.^39,40 Architecturally, three-atom-thick layered morphological structures involving WS₂, WSe₂, MoS₂, MoTe₂, and TaS₂ in these transition metal atom-like Mo, W, and Ta atoms packed between usually two chalcogen atoms of S, Se, and Te. In the interim, layered structures of one atom have hexagonal lattice plans of B–N and C–C bonds, correspondingly graphene and h-BN.^41–44 2D materials can deform via in-plane elongating or bending out-of-plane. Consequently, we can consider the example of graphene being bowled into carbon nanotubes that have characteristics of 2D materials in-plane and bending moduli with well-defined stretching and bending.^45,46 In the meantime, van der Waals heterostructures established by three significant materials, graphene, chalcogenides, and boron nitride, display unusual optoelectronic characteristics.⁴⁷ The high-quality layered boron nitride displays the 5.97 eV bandgap, which can be used as an insulator. Fig. 1b illustrates the overall band synchronization of 3 main types of heterojunctions. Flexible devices rely primarily on Young's modulus and the splitting strain constraint of two-dimensional semiconductors. Their various architectures heavily influence 2D semiconductors' Young's modulus as well as fracture strain constrain. Single-layer graphene with planar structures produced by mechanical exfoliation is recognized as that of the best material with the greatest possible Young's modulus (Fig. 1c) and the maximum breaking strain limit due to the strong sp² hybridization C–C covalent bonds. In recent years, wafer-scale mass production of microelectronic devices based on 2D materials has achieved significant results which overcome key challenges in materials synthesis and device fabrication.⁵⁰ Transition-metal dichalcogenides (TMDs) have gained substantial curiosity as auspicious applicants for developing electronic, photonics⁵¹ and valleytronics device applications.^51,52 Owing to the layered morphology, it has enormously high anisotropic electrical and ionic transport features.^53–56 Additionally, these materials display polymorphs of different structural phases, which propose unique physical characteristics and are acknowledged as the hot topic of interest as a substantial evolving platform to introduce low-power and high-performance memristor devices. On the other hand, TMD-based RS layers have also been studied for memristors. Among them, Sangwan and his colleagues⁵⁷ examined the defect kinetics of MoS₂. Recently, Yan and his colleagues⁵⁸ testified that Ti₃C₂T_x 2D transition metal carbide memristors were used as an RS layer for mimicking biological synapses. Simultaneously, the group promoted single-layer WS₂ sheets for application in the two-terminal microelectronic device, which consumes a low power at the femtosecond level. Li and colleagues⁶ mainly discussed two major intrinsic memristive mechanisms: polarization switching and resistive switching. Current memristor devices can efficiently mimic biological synapses and empower the assembly of neuromorphic systems as a promising non-volatile memory storage device. Low power density and high-temperature stability significantly improve the performance of memristors. Rehman and his colleagues⁵⁹ presented the device structures, conduction mechanisms, resistive switching properties, and fabrication technologies of 2D-material-based RRAMs.⁵⁹ In this article, we review the performance of the 2D material in terms of various memristor engineering techniques, their physical resistive switching mechanisms, and variability. The recent challenges that 2D material-based devices are facing are discussed, and reliable strategies are presented. These materials cover a complete range of physical properties from electrochemically conductive (graphene) to semiconducting (dichalcogenides) and then highly insulator materials (hexagonal h-BN). The resistive switching mechanism should be considered on a comprehensive spatiotemporal series. It needs a thoroughly cooperative implementation of in situ techniques side by side with device demonstrations. Commonly used in situ techniques are presented in Fig. 1d, for instance, scanning tunnelling microscopy, transmission electron microscopy, optical microscopy, scanning transmission X-ray microscopy, conductive atomic force microscopy, electron beam-induced current imaging, and cyclic voltammetry.¹⁸

Fig. 1 (a) 2D materials family. (b) Three different types of heterojunctions of 2D materials have different energy band arrangements. (c) Heterostructure of various material layers.¹⁶ Reproduced from ref. 16 with permission from IOPScience, copyright [2017]. (c) Young's modulus,¹⁷ reproduced from ref. 17 with permission from Wiley, copyright [2017]. (d) Spatiotemporal resolution of several in situ systems and device demonstrating methods.¹⁸ Reproduced from ref. 13 with permission from Springer Nature, copyright [2017].

In this review, we looked at memristor devices made of 2D materials in detail, which appeared among the most intriguing electronic devices in recent years. We have discussed the properties, synthesis approaches and neuromorphic functionalities observed in 2D material-based synapses and also their artificial neural applications in depth. Because of the exceptional properties, such as high carrier mobility, an adequate bandgap, and a large light absorption coefficient, these materials have been regarded as promising applicants for use in artificial synaptic devices to encourage energy-efficient neuromorphic computing. Statistical analyses on digital devices and circuits are often undervalued. The authors have addressed the most pressing integration problems which these devices encounter e.g., stressing the importance of appropriate synthesis approach, yield, variability between devices, consistency, and stability performance analysis, and laying out feasible approaches in research papers that are helpful to the industry. Nonetheless, neuromorphic applications based on 2D synaptic devices are still in their early stage. A schematic diagram of the main emphases points of this review is presented in Fig. 2.

Fig. 2 The main points of this review are depicted in a diagram, together with the various types of 2D materials, their synthesis techniques, different structures, RS mechanisms, and application in 2 terminal memristor devices.^48,49 Reproduced from ref. 48 with permission from Springer Nature copyright [2017]. Reproduced from ref. 49 with permission from Elsevier, copyright [2019].

2. Synthesis strategies for 2D materials

The surface-to-volume proportion of the two-dimensional materials is high, and the bandgap is appropriate (Fig. 3a). For practical applications in semiconductor technology and optoelectronic devices, the ability to synthesize 2D materials reliably is required. There are various synthesis approaches for 2D material-based functional, active layers. They are commonly classified into the following categories: (i) top-down and (ii) bottom-up strategies (Fig. 3b). The first top-down technique includes exfoliation (Fig. 3c) (mechanical, chemical, solution-based, and liquid-phase exfoliation). The bottom-up technique includes following the synthesis process of 2D materials, chemical vapor deposition (CVD), physical vapor deposition (PVD) (Fig. 3d), pulsed laser deposition (PLD), and wet chemical methods. Two-dimensional materials are commonly made using three techniques: mechanical cleavage, physical vapor deposition (PVD), and chemical vapor deposition (CVD), as shown in Fig. 4. Micromechanical cleavage, also known as the Scotch tape approach, is being used to exfoliate single- and few-layer sheets. This approach produces crystalline samples appropriate for micrometer-scale electrical and optical devices. Researchers have selectively extracted, shaped, and stacked sheets with regulated thickness and orientation because of recent breakthroughs in exfoliation processes.⁷⁷ Physical vapor deposition (PVD) is a technique for depositing a material on a substrate using a heated atomic source. In most cases, PVD requires an ultrahigh vacuum (UHV) and the very best purity atomic source materials.

Fig. 3 (a) 2D material properties. Within each layer of TMC nanostructures, metal and chalcogen atomic nuclei are tightly linked. (b) Top-down and bottom-up techniques to solution-based production of two-dimensional layered materials nanosheets.⁶⁰ Reproduced from ref. 60 with permission from Royal Society of Chemistry, copyright [2018]. (c) Intercalation exfoliation and (d) pulse laser deposition.⁶¹ Reproduced from ref. 61 with permission from Elsevier, copyright [2019].

Fig. 4 Basic 2D material synthesis techniques. Microchemical exfoliation (a), physical vapor deposition (b), and chemical vapor deposition (c).⁴⁸ Reproduced from ref. 48 with permission from Springer Nature, copyright [2017].

Surface-sensitive methods can be used to study samples created under UHV conditions in situ, preserving the material's pristine state.⁷⁸ CVD is a technique for growing 2D materials that involves the breakdown or reaction between gases, liquid materials, or solid species in a precise environment.⁷⁹ The most effective substrates for CVD are usually those that are nucleophilic. The synthesis could be done at a variety of pressures, from barometric pressure to ultrahigh. The growth process differs depending on the substrate: at the same time as those with the poor precursor, solubility could catalyze the self-limiting production of single-crystal thin layers although others incorporate huge quantities of the precursors separate into the single-layer upon cooling. Solid-state precursors are injected further into the growing medium to achieve surface breakdown or separation. When the substrate is suitably heated to commence sublimation or diffusion operations, a surface enriched with the precursor's component develops. This layer then condenses as a 2D phase. Memristor device fabrication via a 2D material-based active layer for brain-inspired neuromorphic applications has attracted much attention.^80,81 Considerable attention has been devoted to enlightening the synthesis of the active layer material in fundamental memristive devices.^82,83 The stochasticity of the physical resistance switching mechanism in the devices is a key problem. This uncertainty initiates due to device-to-device variations between several devices and the cycle-to-cycle dissimilarities in single devices among various cycles.^84,85

2.1. Top-down approach

The production of ultrathin nanosheets in top-down nanofabrication is accomplished by a physical or/and chemical technique. The top-down physical approach exfoliates layered van der Waals solids into single or/and few-layer materials using the mechanical strength of ultrasonic vibrational forces. On the other hand, the top-down chemical technique primarily focuses on chemical reactions triggered by electrostatic interaction, heating, as well as other means. To understand 2D materials synthesis techniques in a more convenient way a flow sheet diagram has been drawn in Fig. 5.

Fig. 5 Flow sheet diagram lists the 2D material synthesis techniques.

2.1.1 Mechanical exfoliation. The mechanical exfoliation technique is a primitive one, and was first introduced in 2004. It is an effective approach to yield atomically thin graphene, also nominated as micromechanical cleavage.^48,87–89 By applying this technique with the help of scotch tape, single layers of graphene can be effectively sliced. The graphene layers scatter by constantly reaffirming this ordinary force until it converts into a single layer from the layered bulk.⁹⁰ This technique has likewise been extensively active for 2D materials synthesis.^21,91 The 2D materials have impressive characteristics of significant crystal quality and appropriate defects. Mechanical exfoliation is a modest and feasible technique. However, the non-scalability due to inadequate productivity/yield and sluggish output rate is a problem for future applications. On the other hand, accuracy and replicability in the form of morphology, extent, and width is also an important issue. Numerous scientists have experimented with altering the micromechanical process in this regard. Shmavonyan and co-workers successfully prepared single-layered graphene with an expanded surface area.⁹² Furthermore, additional alterations were also proposed by Peng and colleagues.⁹³ and Zhang and colleagues.⁹⁴ Various exfoliation methods to prepare 2D material-based nanosheets presented in Fig. 6a presented intercalation-assisted expansion and exfoliation significantly used for graphene and graphene-based heterostructures. Exfoliation of layer materials with mechanical force presented in Fig. 6b. Exfoliation along with a layered material using ions and molecules (Fig. 6c).

Fig. 6 Various methods to prepare 2D material-based nanosheets via exfoliation. (a) Intercalation assisted expansion and exfoliation significantly used for graphene. (b) Exfoliation aided by mechanical force, and (c) ion/molecule exfoliation of 2D layered species.⁸⁶ Reproduced from ref. 86 with permission from the Royal Society of Chemistry, copyright [2018].

2.1.2. Liquid-phase exfoliation (LPE). The three phases of the Liquid-Phase Exfoliation (LPE) technique are dispersion, exfoliation, and purification. In 2008, Coleman and his colleagues used the LPE technique to create 2D materials.⁹⁵ The materials are first disseminated in a suitable solvent, then ultrasonication and exfoliation in the LPE technique. Finally, during the purification process, the unexfoliated flakes were removed. Solvents must match the surface energy of the material to provide a well-organized exfoliation process because optimal solvents, such as N-methyl-pyrrolidone and other isopropanol, have the ability to reduce the energy needed for exfoliation.⁹⁶ As a result, it turns out to be a matter of stimulating that achievement with the right solvent. The LPE to harvest 2D materials is influenced by sonication time, temperature, and type of solvent; meanwhile, the surfactants and additives significantly impact the thickness of the layers and the concentration of 2D materials.

Solution Based Exfoliation (LPE), chemically processed ‘wet’ exfoliation methods offer a scalable way to separate single-crystal thin sheets with precise thickness control. On the other hand, synthetic elemental 2D materials frequently need particular preparation processes due to their instabilities in the air. Varieties of materials that do not have bulk counterparts, germanene, for example, could be recovered from bulk crystals if the structure is correct, for instance, CaGe₂, which is made up of covalent bonding layers placed among layers of metal cations.⁹⁷ When CaGe₂ is treated with HCl (Fig. 7a), hydrogenated germanene is formed, which is stable under ambient settings. CaGe₂ is treated with CH₃I to produce methylated germanane Fig. 7a, which results in more stable layers.⁹⁸ This functionalization improves the layers' stability and their light-emitting characteristics in the case of methylated layers. However, even though organic solvents can be used to exfoliate black phosphorus, approaches involving deoxygenated water have recently been as seen in transmission electron microscopy, and has recently been shown to produce high-quality sheets while maintaining the properties and composition of immaculate black phosphorus (Fig. 7b). Transmission electron microscopy (TEM) images presented in Fig. 7c demonstrate atomic-scale and large-scale brilliant field images of black phosphorus flakes.^98,99

Fig. 7 Solution-based exfoliation synthesis, (a) when CaGe₂ is treated with HCl and CH₃I, (b) exfoliation steps, and (c) TEM of black phosphorus flakes.⁹⁸ Reproduced from ref. 98 with permission from Springer Nature, copyright [2014].

2.2. Bottom-up approach

For the synthesis of ultrathin and good-quality nanocomposites, we used a bottom-up approach. The materials production on the nanoscale is made possible via bottom-up processes, by using atomic or/and molecular resources. Then 2D materials could be produced with efficiency and effectiveness. The cost of large-scale production by using this approach is generally more expensive than the top-down approach.

2.2.1. Chemical vapor deposition (CVD). The CVD approach is promising in producing high-quality single to few-layer sheets of 2D materials with scalable sizes, adjustable thickness, and outstanding electrical characteristics. Consider graphene production via the CVD process: at extreme temperatures of up to 1000 °C, carbon-containing gaseous oxide species interact with each other via thin metal foils, resulting in the breakdown of carbonaceous species and the formation of a few-layered graphene frame (shown in Fig. 8). Li and his colleagues in 2009 used methane and hydrogen as carbon sources, used copper film as the substrate, and successfully synthesized single layers of graphene.¹⁰⁰

Fig. 8 (a) A four configuration arrangement for synthesizing WS₂via a CVD technique and (b) single layers of WS₂ on the wafer level. (c) The multi-zone tube furnace.¹²¹ Reproduced from ref. 121 with permission from Springer, copyright [2018]. (d) WS₂ monolayer synthesis. (e) WS₂ optical image.¹²² Reproduced from ref. 122 with permission from Wiley, copyright [2016].

CVD is thought to be a useful method to grow layered materials on the surface of a metal substance. Several issues influence the resulting materials, e.g., type of precursor, the solubility percentage of the carbonaceous material of the metal component, interlayer lattice spacing fringes, crystal purity, a ratio of strain, and temperature of the whole assembly.^101,102 After fine-tuning the environment of the CVD technique, we can skillfully adjust the thickness of the layers' crystal size and its purity.¹⁰³ Somani and colleagues have used the thermal CVD technique and successfully designed planer structured thin graphene.¹⁰⁴ The substrate was cost-effective (camphor on nickel) to synthesize a few graphene layers.

3. 2D materials

Two-dimensional materials make up a novel family of nanostructured materials with promising features that can be used as emerging materials for extremely thin semiconductor devices in the future.¹⁰⁵ Meanwhile, after discovering graphene,^21,106,107 various 2D materials, for instance TMDs, h-BN, and as well as black phosphorus (BP), have all been extensively investigated.^108–110 The reference collection of nanostructured 2D materials, ranging from metal to insulators, provides us with a variety of options for designing diverse devices. Nowadays, 2D materials are famous for their exclusive properties, such as low dimensional assemblies and special physical and chemical characteristics.

3.1. Characteristics of 2D materials

High carrier mobility, an adequate bandgap, strong light–matter interaction with a big light absorption coefficient, an extremely large exciting binding energy, significant spin splitting, and polarized light emission are all desirable features of TMDs. All of these intriguing characteristics make the 2D materials an excellent candidate for use in memristors, optical modulators, light-emitting devices, and a variety of other applications.

The interlayer spacings in van der Waals materials are covalently bonded via chemical connections. Furthermore, the mobility is extremely great as it is up to 120 000 cm² V⁻¹ s⁻¹111,112 and the mobility is also high up to 1000 cm² V⁻¹ s⁻¹112 for thte 10 nm thin BP film. The above factors can influence the switching speed of 2D structured memristors and regulate the on/off ratio, power consumption during RS, and operating voltage.^112–114 We can synthesize 2D materials on any substrate, regardless of lattice disparity, using simple synthesis approaches such as solution-based methods, a mechanical exfoliating method which is very well-known, liquid-phase exfoliation, and also chemical vapor deposition. Furthermore, we may change the physical properties of 2D materials using various methods, such as defect induction, ion intercalation (Li-ion), and doping.^115–117 Because of their atomic thickness, neuromorphic synaptic applications can benefit from them. As a result, synaptic devices made of a 2D material may respond to various stimuli, including strain field, electric field, and Joule heating. Mechanical faults, including such intrinsic defects and grain structure, have a significant impact on the intrinsic characteristics of underlying atomic layers in 2D materials.^118–120 We could introduce point defects and grain boundaries using various techniques, including electron beam irradiation, changing growth conditions, and post-plasma treatment.^123–125 As a result, we may change the electrochemical and photonic characteristics using the above approaches.¹²³

Sulfur vacancies, for example, are found in layered MoS₂ and act as electron contributors, allowing the identification of contributor locations on the inner side of the bandgap.^126,127 These openings present potential networks for ion transport, which might assist the development of memristors and synapses. In 2D metal/h-BN/metal memristors, the boron vacancies result from the flow of boron ions in the direction of the anode, yielding a track that metal ions may fill.^128,129 Hypocritical heating could insert the oxygen atoms inside MoS₂ at room temperature, and layered MoS_2−xO_x was achieved.¹³⁰ The S vacancies are molded due to the thermophoresis effect. These S positions act as pouches for the movable O ions. Finally, the oxygen ions fill and leave the new S vacancies, and this phenomenon chiefly helps to modify the resistance.¹³⁰ These defects are also responsible for creating and distorting conductive filaments in several 2D-based memristors via oxidation, migration, and reduction progressions. The relocation of ions typically happens in the out of plane route with help from structural defects. For example, the defect-free single-layered MoS₂ sheets need 35.01 eV of energy to transport Ag⁺ ions.

On the other hand, single-layered MoS₂ with defects needs 1.81 eV of energy.^124,131 Defects in graphene layers have similar effects, and metal ion traveling is just possible due to defects. Due to defects, the place/dimensions of conductive filaments might be precisely controlled.^132,133 The 2D materials interact significantly with environmental stimuli because of their thin atomic layers and decreased screening efficiency. The polymorph property is a significant feature of TMDs.¹³⁴ Commonly, TMDs have five categories of phases together with 2H, 3R 1T, 1T′, and Td stages.¹¹⁷ MoTe₂ presented the various phase change between its monolayers of 2H and 1T′ following ionic liquid gating. On the other hand, MoS₂ and MoSe₂ practice the 2H phase (trigonal prismatic) to distort the 1T phase (octahedral transition) via alkali intercalation.^135–139 The polymorph transition leads to the application of 2D materials as non-volatile memristor and synapse devices.^3,140,141 When we consider the character of the crystalline phase of 2D materials, several crystalline phases have already been exposed e.g., α, β, γ, κ and δ phases.^142,143 A single-crystal few-layered material has three phases (α, β, and γ phases) altogether steady at ambient temperature. For instance, β-In₂Se₃ has low resistivity with an insufficient sequence of magnitude. Still, it is also lower than other phases (α and γ), and this phase transition is irreversible. Most importantly, during SET, the β phase is at a low resistance state. During RESET, the γ phase is at a higher resistance state.²⁹ Photoresponsivity for 2D metal chalcogenides can be from the infrared to ultraviolet range.^29,146 The monolayer of TMDs such as MoS₂, MoSe₂, WS₂, and WSe₂ can grasp sunlight optical density up to 15% inside the visible light spectrum.¹⁴⁷ Furthermore, the 2D material based on the van der Waals heterostructure can result in functional devices for enabling research in artificial intelligence.

Fig. 9a demonstrates a graphic diagram of the Pd/WS₂/Pt memristor device with TEM micrographs of the WS₂, which specify that the WS₂ contained a layered physical morphology (Fig. 9b). The WS₂ has 0.645 nm d spacing and (002) crystal orientation (Fig. 9c). The multiple dots (red and blue) presented a beautiful hexagon that specified compact hexagonal morphology, a good presentation of a 2H-WS₂ unit cell.⁶²Fig. 9d presented the schematic design of the MoS₂-UCNPs film with a device structure, and Fig. 9e presents the side view through SEM. AFM images of the active layer fabricated via a spin coating technique are shown in Fig. 9f.¹⁴⁴ Wang and his colleagues designed a robust graphene/MoS_2−xO_x/graphene memristor (Fig. 9g). The multilayer graphene was ∼8 nm thick, and the MoS₂ has been automatically used exfoliated, and positioned at the SiO₂/Si wafer substrate. The optical image of a representative GMG memristor device showed a stacked vdW heterostructure (Fig. 9h). The high angle annular dark-field (HAADF) TEM image displays layered morphology (Fig. 9i).

Fig. 9 (a) Schematic diagram of the Pd/WS₂/Pt memristor device, (b) TEM micrographs, and (c) HRTEM structure showing 0.654 nm surface spacing which conforms to the 002-crystal alignment.⁶² Reproduced from ref. 62 with permission from Wiley, copyright [2019]. (d) The schematic drawing of the MoS₂-UCNPs fabrication, and the (e) SEM, and (f) AFM of the surface structure of the active layer.¹⁴⁴ Reproduced from ref. 144 with permission from Wiley, copyright [2018]. (g and h) The schematic drawing, optical microscope image, and dimension arrangement with four-probe sizes of a GMG device and (i) the cross-section HAADF image of the MoS_2−xOx upholds the layered crystal.¹⁴⁵ Reproduced from ref. 145 with permission from Springer Nature, copyright [2018].

4. Structure and resistive switching mechanism in 2D material-based memristors

The previously reported famous two-dimensional materials and their proposed structures and resistive switching mechanism routes are summarized in Fig. 10 and 11. Various 2D materials were used as an active layer to fabricate artificial synaptic devices, the initial stage in creating a highly energy-efficient neural network model (Fig. 10a).

Fig. 10 Review of the 2D material-based memristor devices with diverse materials¹⁴⁹ and structural differences (lateral, vertical). Reproduced from ref. 149 with permission from Springer Nature, copyright [2020].

Fig. 11 Mechanisms of resistive switching in the 2D material-based memristor devices. (a) conductive filament formation, (b) phase transition, (c) vacancy migration, (d) defects formation, (e) charge trapping and (f) Schottky emission (SE) and direct tunneling (DT).

The structural geometry of 2D material-based memristor devices would be categorized into two groups entitled lateral and vertical devices (Fig. 10b and c). The working mechanism of each device depends upon the structure and active material. Zhang et al.¹⁴⁸ discussed the fundamental mechanisms of RS behaviors in different types of 2D materials compared with solid-state memristors and also the fabrication of the device. Firstly, the electrodes of the lateral-memristor horizontally arranged on the dielectric surface are presented in Fig. 10a. The charge moves from the in orientation, and analog-like resistance changes occur as a result of defect mobility towards the grain interface or structural phase change within the 2D materials.¹⁵⁰ The atomically thin layer is exposed in lateral devices, so it is quite easy to carry out in situ measurements. The mechanism behind resistive switching can be confirmed via in situ characterization.^55,57,151 The flow of charges in memristor devices is out-of-plane in vertical-structure devices (Fig. 10c). The 2D material-based active layer is inserted along with the upper and lower electrodes.^{145,151–153}

The resistive switching mechanisms in vertical-structure memristors¹⁴⁸ include filament formation¹⁵⁴ phase transition,¹⁵⁵ vacancy migration^156,157 (Fig. 11(a–c)), defects formation, and charge trapping^158,159 as presented in Fig. 11(d and e). The Schottky emission (SE) and direct tunneling (DT) are also significantly mentioned in Fig. 11f.^160–163 Depending on the suggested structures and materials, these devices use a variety of mechanisms. Furthermore, in situ characterizations are extremely straightforward to examine the morphology of the active layers made of 2D materials such as MoS₂ and WS₂ due to the nanostructured atomic layer morphology. As a result, the comparator resistive switching mechanism and various synaptic functions can be the authority to ensure those presented in Fig. 10d (including LTP, LTD, and STDP).

Table 1 classifies memristive switching mechanisms based on the active layer materials, fabrication approach, RS mechanism, electrode materials, switching voltage, ON/OFF ratio, endurance, and retention time. In particular, lateral and vertical operate at low power and have various features and functionality that assist in complex learning. Bulk materials fabricated by memristors have a high rate of reliability, stability, endurance, and retention as compared to layered materials, which can be improved by high-quality 2D materials and changing their morphology via heterostructure fabrication.

Table 1 2D material-based memristor devices with their performance

Active layer material	Fabrication method	RS mechanism	Electrode's materials	Switching voltage	On/off ratio	Endurance	Retention time	Ref.
WS2	Solution	Vacancies	Pd/Pt	0.6	—	75	1.8 × 10^4	62
WSe2	Mechanical ex-foliation	Vacancies	Ag/Graphene	0.2–0.5	10^2	—	—	63
ZrO2/WS2	Large-area RF sputtering process	Filament Formation	Ag/Pt	0.1–1.4	10^3	10^9	—	64
WO-x/WSe2		Vacancy Migration	Ag/Graphene	0.2	10^7	10^3	10^3	63
MoS2	LPE	ECM, CFs	Ag/Au	3.5	1.6 × 10^2	—	—	65
MoS2	CVD	SE-DT	Au/Au	1.0	10^2	1.5 × 10^2	10^6	66
Monolayer MoS2	CVD	Grain boundary	Au/Au	3.5	10^3	—	>120	67
MoS2 nanofibers	Solution	SCLC	Rgo/rGO	—	10^2	50	—	68
MoOx/MoS2	LPE	Charge trapping	Ag/Ag	0.1–0.2	10^6	10^4	10^4	69
MoS2 double-layer	MO CVD	ECM	Cu/Au	0.1–0.2	—	—	—	70
MoSe2	Solution	Charge trapping	Au/ITO	3.8	10^3	—	10^3	71
MoTe2	Mechanical exfoliation	Lattice distortion/phase transition	Ti/Ni Ti/Au	2.3	10^2	—	10^3	72
h-BN	CVD	Grain Boundary	Ti/Cu	0.4	10^6	—	10 h	73
h-BN	CVD	ECM	ITO/Ag	0.7	10^2	550	10^3	74
h-BN-PVOH	Solution	SCLC	ITO/Ag	0.78	10^2	>1000	—	75
TaS2	Chemical vapor transport method	Phase transition	Ti/Ti	3.0	10^3	—	—	76

---

Resistive switching mechanism in low dimensional systems

Several evolving memories, including magnetic memory, phase change memory, ferroelectric tunnel junctions, and resistive switching memory, have been studied in recent decades.¹⁶⁴ Within these developing devices, memristors have several characteristics including non-volatility, small size, random data access, and are easy to fabricate with very simple structural morphology. Due to the above mentioned properties, various materials are being investigated for memristive properties. There are numerous materials which can exhibit memristive attributes. Binary oxides,¹⁶⁵ 2D materials, perovskites, bioinspired materials, polymers, halide perovskites, and low dimensional materials (nano-composites) are examples of these.¹⁶⁶ Each kind of material seems to have economic benefits in its working principles and/or attributes, that also helps in terms of a better resistive switching device and artificial synapses performance. There are two main types of memristor working mechanisms (i) filamentary resistive switching, and (ii) interface type switching.¹⁶⁷ As per the electrode and dielectric materials, bipolar filament resistive switching of memristor is classified as electrochemical metallization (ECM) and valence change memory (VCM).¹⁶⁸ ECM is common in various oxides, organic materials and perovskites with electrodes of electrochemically active metals such as Ag and Cu.

In most cases, the VCM mechanism is established in an active material layer and inert metal electrodes. The resistive switching layer is typically composed of a transition metal oxide, such as TiOx, HfOx, TaOx and AlOx, as well as perovskites such as BiFeO₃ and SrTiO₃.^176,177 Here, the conductive filament is made up of ionic vacancies. To create oxygen vacancies, the electrode should act as an oxygen reservoir. A physical or chemical phenomenon at the dielectric and electrode interface causes interface type switching.¹⁷⁸

The polymer-based active layer in the memristor device changes the conductance via various resistive switching mechanisms by modifying their intrinsic properties e.g., charge transferring, phase change, redox reactions and conformation changes. Halide perovskites have impressive on/off ratios, high switching speeds, low operation voltages, and synaptic behaviours, yet their air stabilisation needs improvement before they can be used in practical memristors and artificial synapses.

Low-dimensional materials like nanoparticles can be used in memristors. Evenly distributed Ag NPs promote Ag filament emergence while inhibiting random filament formation and rupture.^179,180 Wang et al. published a paper in 2016 about TaN/Al₂O₃/ZnO/ITO devices with dispersed Ag NPs. In addition, MoS₂ could produce ion vacancies using an oxidation process, resulting in synaptic behaviours with changes in device conductance. It has been noted that GBs in MoS₂ can be easily modulated to regulate the change in resistance.

Neuromorphic functionalities observed in 2D memristive devices

The human brain is a highly efficient and perfect information processing system, including around 10¹¹ neurons and 10¹⁵ synapses.⁴ Nerve cells and synapses seem to be the brain's basic memory and cognitive units. They are also the ones that have received the most attention in neuromorphic computing systems. The synaptic connection determines the signal transmission efficiency in biological systems from one neuron to another. It is very well through purposeful stimuli or learning. Therefore, an artificial intelligence system based on neuromorphic computation offers a promising way to solve the von Neumann bottleneck.^182,183 Neurons consist of somatic cells, dendrites, and axons. Generally, the amplitude of the action potential produced by a neuron is about 100 mV and the duration is 1–10 ms.¹⁸⁴ According to the type of synapses, neural signals can be converted into chemical or electrical forms.¹⁸⁵ Neurotransmitters are able to diffuse freely in the synaptic cleft and then integrate into receptors in the post-neuron. Two-terminal synaptic devices were studied by Strukov et al.¹³ based on the memristor introduced by Chua.¹⁸⁶ In the devices, two types of configuration are often adopted: horizontal and vertical. Following that, there has been remarkable advancement in 2D material-based memristor neuromorphic devices. The ability of different 2D material candidates to mimic essential synaptic functions such as short-term plasticity (STP), long-term plasticity (LTP), short-term depression (STD), long-term depression (LTD), paired-pulse facilitation (PPF), and spike-time-dependent plasticity (STDP) learning has been studied. These functions are important for the construction of SNNs. STP, STD, LTP, and LTD are types of neuroplasticity related to short-term and long-term memory that indicate temporal or long-term enhancement/decay of synaptic connections. PPF is an example of short-term plasticity. In recent years, the memristor, a combination of memory and resistor, has emerged as one of the most essential electronic components for brain-inspired neuromorphic computing. This device can manage resistance with various states by remembering the history of past electrical stimuli, allowing it to simulate a biological synapse in the human brain's neural network. Among many candidates for memristive materials, such as metal oxides, organic materials, and low-dimensional nanomaterials, 2D layered materials have received a lot of attention owing to their excellent physical properties, electrical processability, and low-power switching capability. As a result, a great number of experimental demonstrations on 2D material-based memristors have been recorded, demonstrating their specific memristive properties and novel synaptic capabilities in comparison to classic bulk-material-based systems. This paper provides an overview of the most recent breakthroughs in the architectures, processes, and memristive properties of 2D material-based memristors. To achieve these features with low power consumption and multilevel resistance control with high reliability, a breakthrough in the discovery of new materials and the design of novel device architectures is required.

5. Applications for neuromorphic computing

Artificial intelligence began as a field to replicate intelligence in a machine.¹⁸⁸ The human brain is a highly efficient and perfect information processing system. One of the key techniques for efficiently managing data-intensive applications is brain-inspired neuromorphic computing. It is a cost-effective approach of incorporating analogue memristors into artificial neural networks or combining memristors with artificial neurons to build hardware neural networks.¹⁴¹ Despite the fact that such memristor-based systems give novel computer architectures, huge quantities of supplementary analogue and logic circuits are still required for the construction of artificial perceptual intelligence systems.¹⁸⁹ Luckily, the merging of synaptic devices and biometric sensors opens up new possibilities for constructing functional neuromorphic sensory stimuli, also known as near sensor technology or edge computing.¹⁹⁰ Current improved diagnostic strategies have demonstrated several more relevant and fascinating pieces of information about cognition inside mammals’ nervous systems, which must be precisely taken into account in solid-state devices in terms of developing quite plausible and effective human-like machines and human–machine integrations. These artificial devices could also assist scientists and therapists in mimicking and grasping various brain illnesses and problems. It would also lead to the formation of artificial sensory audio, visual, and motion systems, to investigate the progression of diseases and disorders, as well as the effects of medications, and artificially intelligent robotics which might ultimately simplify the complex surgical procedures.

Yan et al.⁶² designed a Pd/WS₂/Pt device, and the device showed the vacancy migration mechanism as presented in Fig. 11c.

The device performed admirably and consumed very little power (in the order of femtojoules). The results of trap-assisted tunnelling conduction were in accord with the fitting of the I–V curve (Fig. 12a), which determines that the defect-trap energies below the conduction band and the tunnelling distance at the HRS (LRS) are 0.62 eV (0.37 eV) and 0.94 nm (0.29 nm), respectively. Furthermore, with the 1 A program current in the ON state, it had an extremely fast ON/OFF switching times ratio (13/14 ns) as presented in Fig. 12b and c. The Pd/WS₂/Pt device showed the VS and VW vacancies, and also the electron hopping effect. Fig. 12d compares the Pd/WS₂/Pt device to other 2D material-based devices; obviously, this device consumes the least amount of current and power. Additionally, it effectively mimicked the significant synaptic functions including such inhibitory and excitatory under the influence of positive and negative stimulus trains. Moreover, memory and learning above the influence of PPF, STDP functions and also follow the STP transition to LTP. Significantly, it is predicted that sulfur and tungsten vacancies and electron hopping between vacancies determines the resistance switching performance. Miao and his colleagues¹⁴⁵ fabricated a graphene/MoS_2−xO_x/graphene-based memristor device (Fig. 12e) that displays reproducible bipolar resistive switching with an endurance higher than or equal to 107 stable cycles with a high specific heat and a working temperature of approximately 340 °C. The in situ high-resolution transmission electron microscope (HRTEM) results at extreme temperatures indicated that the 2D material-based active layer is responsible for the devices' high thermal stability. Furthermore, in situ micrograph (STEM) investigations on a functional device revealed a heat transfer route and a resistive switching based on oxygen ion relocation. The pristine graphene/MoS_2−xO_x/graphene layer is shown in Fig. 12g. The on/off state of the layered atomic structure of the electrode has been discovered to be very well maintained even during long resistive switching endurance. The stability in the structure plays a significant role in establishing the device's consistency (Fig. 12h and i). Thomas Tuma et al.¹⁸¹ reported on a nanoscale memristive device-based biological neuron with an integrate-and-fire functionality, which could lead to significant advancements in the development of integrated circuits (Fig. 13).

Fig. 12 (a) Characteristic I–V arch of the Pd/WS₂/Pt device. (b and c) ON/OFF states used the 13/14 ns voltage pulse,⁶² which was produced from ref. 62 with permission from Wiley, copyright [2018]. (d) comparison of the Pd/WS₂/Pt device to other 2D material-based devices.^{130,169–175} (e) Graphene/MoS_2−xO_x/graphene optical image.¹⁴⁵ (f) Typical I–V curve of the graphene/MoS_2−xO_x/graphene device. Schematic drawings of the RS mechanisms (g) pristine state, (h) ON state, and (i) OFF state. Reproduced from ref. 145 with permission from Wiley, copyright [2018].

Fig. 13 Artificial neuron based on a phase-change device, with an array of plastic synapses at its input.¹⁸¹ Reproduced from ref. 181 with permission from Springer Nature, copyright [2016].

Kumar and colleagues⁶⁴ developed a ZnO/vertically aligned WS₂ layer memristor (Ag/ZnO/WS₂/Al) architecture (Fig. 14a) that addresses the majority of the significant difficulties in artificial synapses design to adjust the conformal development of ZnO across the WS₂ layers, and cross-sectional SEM imaging was used as presented in Fig. 14b. The device demonstrates very low variability values by showing not only cycle-to-cycle stability but also device-to-device stability in performance. Temperature-dependent I–V measurements (Fig. 14c and d) and depth-dependent XPS metrics are demonstrated by the oxygen vacancies that are distributed randomly across the device. Fig. 14e depicts a typical configuration of an Ar⁺ sputtering and XPS measurement setup. Ar⁺ sputtering with a 5 keV energy was used, resulting in losing the device's layers. The XPS measurement spot diameter was set at 100 m. According to the spatial X-ray photoelectron spectroscopy (XPS), vertically oriented WS₂ layers provide an excellent porous medium for the formation of defective ZnO (Fig. 14f). The generation/redistribution of oxygen vacancies was confined by a vertically aligned 2D layered design (Fig. 14g-h), ensuring high cycle-to-cycle consistency for 104 successive cycles and very low device-to-device variability. The device also meets significant requirements for neuromorphic computing, such as digital signal hysteresis loop opening (I–V characteristics), extended recognition durability, a very high current (on/off) ratio, as well as a wide range of electrical conductivity associated with various states. The Ag/ZnO/WS₂/Al memristor can measure all essential properties of an artificial synapse e.g., short-to-long-term plasticity, spiking time-dependent plasticity, and also spike-timing-dependent plasticity. For read and write various dynamic functions were replicated using a memristive synapse array. This research will pave a way for developing very robust large-area neuromorphic devices with practical applications.

Fig. 14 (a) The Ag/ZnO/WS₂/Al memristor.⁶⁴ (b) The SEM image of a memristor device demonstrated the existence of multilayers of ZnO₂/WS₂ and (c) I–V characteristics at various scanning rates in normal and (d) semilogarithmic scales. (e) Schematic representation of the Ar⁺ sputtering setup. (f) XPS measurements present the stoichiometric ZnO and WS₂ layers with appropriate charge deceiving locations. (g) Without the charge trapping device structure, and (h) with the clear charge trapping location. Reproduced from ref. 64 with permission from Wiley, copyright [2019]. (i) Presents the single MoS₂/graphene device, and (j) the results validate that the MoS₂/graphene device demonstrates a stable change in conductance. (k) 15 pulse pulsing schematic trains, 100 identical positive/negative pulses. (l) The synaptic weight change due to pre-spike and post-spike signal timing differences.¹⁸⁷ Reproduced from ref. 187 with permission from AIP, copyright [2019].

Krishnaprasad and his colleagues¹⁸⁷ fabricated monolayer MoS₂ based synapses via chemical vapor deposition and a graphene electrode was deposited via photolithography as presented in Fig. 14i. The non-volatile devices showed a range of conductance levels which was a function like synapses. The SET process may be achieved at varying current compliance (1 nA to 1 A) as presented in Fig. 14j. These findings show that the MoS₂/graphene device exhibits a steady shift in conductance between low and high resistive switching states which is required for outdistancing synaptic potentiation and depression (Fig. 14k). They also show key synaptic features such as short-term potentiation and long-term potentiation (LTP) in the same device, and the pulse mechanism is set up in a way in which the pre-spike and post-spike transmitters just intersect once in a while for spike-timing-dependent plasticity (STDP) in Fig. 14l.

Zhu and colleagues¹⁹¹ fabricated Au/Ti/G/h-BN/G/Au devices (5 m × 5 m) which presented various constant resistive states at several current limits (CL) (Fig. 15a and c), which were very reliable in response to the performance. The CL was raised to control the transformation to more conductive levels while the reset voltage was used to control the transformation to less conductive levels. For high CC = 5 mA (Fig. 15c) the reset voltage was around 1 V, and for low CC = 0.5 mA it was around 3.5 V (Fig. 15a). The set voltage remained fairly constant at around 3.5 V. When utilizing CC = 1 mA (Fig. 15b) the metastable state displayed several resistive switching states (H-RS, S-LRS, and L-RS) which is presented in Fig. 15d. This operation has not been detected in graphene-free devices 5 m × 5 m (Au/Ti/h-BN/Au) or analogous devices (Au/Ti/G/h-BN/G/Au) with bigger sizes (100 m × 100 m). These findings suggest that 2D material-based memristors must be scaled up to detect an authentic device performance (e.g., tristate operation). Because graphene could indeed restrict the ionic interaction between both the metallic electrode and the h-BN resistive switching, an additional S-LRS can be formed. Variations in thermal conductivity and switching voltage have indeed been statistically examined. It has been established that all resistive states may be consistently discriminated against. This discovery is a significant step forward in creating variable memristors based on 2D layered materials and contributes to a deeper understanding of the out-of-plane RS process.

Fig. 15 (a) In an Au/Ti/G/h-BN/G/Au device, I–V curves show two-state RS between the HRS and SLRS using CL 0.5 mA, (b) using CL 1 mA, and (c) using CL 5 mA. (d) The mechanism behind filament formation at various stages.¹⁹¹ Reproduced from ref. 191 with permission from the American Chemical Society, copyright [2019]. (e) Representative I–V features of the Cu/MXene/SiO₂/W memristor. (f) The CFs in Cu/MXene/SiO₂/W memristors are much more likely to grow in lock-step with a position of MXene, lowering the variations in the CFs. (g) DFT calculations showed unrelaxed and relaxed geometries, and the supercell of the hybrid structure of MXene (Ti₃C₂)/SiO₂.¹⁹² Reproduced from ref. 192 with permission from AIP, copyright [2019].

Lian and his colleagues¹⁹² fabricated MXene based devices (Cu/MXene/SiO₂/W) (Fig. 15e). Incorporating the desired features of silicon with that of an MXene (Ti₃C₂) material, a memristor-based on an MXene/SiO₂ structure is fabricated. Under the same circumstances, the authors compared the performance of both devices and the Cu/Ti₃C₂/SiO₂/W memristive devices outperform typical Cu/SiO₂/W devices in switching performance. The material characterizations and DFT first-principles calculation results of the MXene/SiO₂ structure presented the geometry of their structure in Fig. 15g after MXene coupling with SiO₂. The findings show that conductive filaments (CFs) are much more able to constitute at the sites of MXene nanostructured materials, reducing CF variation and improving the performance of the device in Cu/MXene/SiO₂/W memristors (Fig. 15f). In the meantime, the MXene/SiO₂ device significantly reduces Cu ion mobility across the RS area. It increases the SiO₂-based memristor performance while keeping the operating voltages low.

Arnald et al.¹⁹³ used hysteresis programming in the MoS₂ based device to simulate biological synapses via a solid-state device. They proved that by varying the frequency, amplitude, and polarity of gate voltage pulses, they were able to properly capture the quantal, stochastic, excitatory or inhibitory nature of neurotransmitter release in chemical synapses. They also described a potential method for replicating and engineering long-term potentiation in the human brain, which serves as the foundation for learning and memory. As shown in Fig. 16, these are structures in which the plasma membrane of the signal-carrying neuron as a presynaptic neuron is in close proximity to the plasma membrane of the target neuron as a postsynaptic neuron.

Fig. 16 Biological synapse illustrations: the inflated assessment of a synapse with the description that the postsynaptic current expands in accordance with the presynaptic potential number.¹⁹³ Reproduced with permission¹⁹³ copyright 2017, American Chemical Society.

The memtransistor is a multi-terminal experimental electrical component that could be used to build artificial neural networks. It's a hybrid of a memristor and a transistor. Sangwan et al.¹⁵⁰ prepared multi-terminal memtransistors from polycrystalline monolayer MoS₂. In comparison to the original brain learning behavior of long-term potentiation/depression, six-terminal MoS₂ memtransistors include gate-tunable heterosynaptic capability, which two-terminal memristors do not have. Long-term potentiation and depression in MoS₂ memtransistors imitated excitatory and inhibitory synapses in organisms. Farronato et al.¹⁹⁶ presented a memtransistor device based on multilayer MoS₂ with an ultrashort channel length. The device had transistor characteristics with a high resistance ratio between LRS and HRS and a memristive switching characteristic at source–drain voltages around 0.8 V. Finally, the potential of the memtransistor device was exhibited with a chain-type memory array design displaying select/unselect properties during the program and read operations. These findings pave the way for high-density, 3D memtransistor arrays based on 2D semiconductors.

Chen et al.¹⁹⁴ developed a memory concept by designing a MoS₂ quantum dot-based device, and the transmission electron microscope (TEM and STEM) images are presented in Fig. 17a and b. The device displayed nonvolatile bipolar resistive switching (on/off ratio of 10⁴). MoS₂ QD-based artificial synapse can function as a bandpass filter for low-frequency implementations. Moreover, in their MoS₂ QD synapse, they illustrated short term plasticity. The charge trapping and de-trapping in quantum dots, rather than the conventional filament-based conduction in memristors, is used to achieve synaptic behaviour. The synaptic weight varied with the action potential sequence. As the number of pulses increases, Fig. 17c shows a clear increasing depression. They measured important functions of the artificial synapse including PPF and PPD (Fig. 17c). They discovered that the MoS₂ QD-based synapse can function as a bandpass filter for low-frequency applications as presented in Fig. 17(d–g). MoS₂ QDs have potential applications in prospective storage devices and neuromorphic technology due to their ease of fabrication and excellent memory characteristics.

Fig. 17 MoS₂ quantum dot-based device TEM and STEM representations presented in (a) I–V characteristics, (b) Excitatory Post-Synaptic Current (EPSC), and (c) PPD and PPF variation for the amplitude of the voltage pulse. (d) Dependence of pulse time variations in PPF. (e) Dependence of time variations in PPF. (f) Dependence of time/frequency in PPF.¹⁹⁴ Reproduced from ref. 194 with permission from Springer Nature, copyright [2020].

Conventional memristors based on bulk materials (for example, transition metal oxides) rely on top-down lithographic patterning that has limited control and tunability over neuromorphic functionality due to the strong dependence of the device response on the atomic-scale structure of defects. In contrast, emerging low dimensional nanomaterials present unprecedented control over stoichiometry, defect engineering and interfacial chemistry, which underlie synaptic behaviour. Low-dimensional nanomaterials also possess quantum phase transitions, new state variables, electrostatic tunability and atomic-level control for reconfigurable switches and neuromorphic functions. Within this class of materials, 2D nanomaterials provide an open architecture for multiple electrodes, in situ probing, and spatiotemporal responses, while zero-dimensional (0D) and one-dimensional (1D) nanomaterials enable highly sensitive covalent functionalization and redox chemistry for multi-bit states. Promising results are also emerging from hybrid systems that integrate nanomaterials of different dimensionalities. Due to their mechanical flexibility, low-dimensional nanomaterials are also facilitating the development of wearable neuromorphic applications such as afferent nerves and smart prosthetics. Here, we review recent progress in the emerging field of neuromorphic nanoelectronic materials with a focus on how the properties of 0D, 1D and 2D nanomaterials and their van der Waals heterostructures enable unique neuromorphic responses compared to inorganic and organic materials. Some important properties of 0D, 1D and 2D materials have been presented in Table 2.

Table 2 Comparisons of the properties of functional low-dimensional materials

Materials	Morphology	Characteristics	Synthesis approaches	Improvements	Shortcomings
0D	Nanoparticle	Charge trapping	Spin-coating	Easy fabrication of large-area devices	Aggregation
	Size controllable	Light responsible	Blending	High surface to volume ratio	Temperature susceptible
	Self-assembled QD-based film	Light responsible flexibility	Langmuir–Blodgett electro deposition	Better solubility	Spatial uniform challenge
		Transparency		Better amenability to hybridize with other nanomaterials
		Piezoelectricity		Easy to be doped and functionalized
1D	Nanowire	Directivity	Spin-coating	Easy to form a network	Difficult to fabricate
	Nano-rod	Light responsible flexibility	Spraying	High carrier transport	High cost
	Nano-fiber	Transparency	Molding	Directivity	Low surface to volume ratio
	Complex network	Piezoelectricity	Electro-spinning	Tuning electronic properties through p–n junction
2D	Nano-sheet	High carrier transport	Mechanical exfoliation	Low roughness	Difficult to fabricate
		Uniform crystallinity	Chemical exfoliation	High temperature tolerance	Specific modifications required
		Ultrathin	Solution based exfoliation	High carrier transport
		Light responsible	Chemical vapour deposition
		Flexibility	Physical vapour deposition
		Transparency

---

6. Variability, reliability, and stability

The significance of statistical analysis on electrical devices and circuits is frequently underestimated, especially for two-dimensional materials. Previous studies have highlighted the most significant integration difficulties for such devices and emphasized the precondition to evaluate variability.¹⁹⁷ Device-to-device variability can be defined as a deviation from the principal parameters and statistics of the devices (such as switching voltages, high and low resistance states in memristors, carrier ion mobility, leakage current, threshold voltage, and swing for FET).

Several elements directly impact the yield and device-to-device variability of these devices. Device-to-device variability in 2D materials is connected with the insertion of various sorts and quantities of defects throughout the device fabrication process, material synthesis, materials transfer procedure, aging, modeling stages, and the deposit of additional coupling of diverse materials.¹⁹⁸ Material defects have a varied effect on the variability of current flow via 2D materials in memristors. In such vertical devices, RS is a random process that still happens at the potentially weak position of an active device component when subjected to an electric current. The most common intrinsic defects in 2D material-based devices are vacancies, cracks, wrinkles, strain or stressed bonding, impurities, and, most important, variations in the thickness of a 2D material. While the most frequently identified defects are related to shifting interactions with the atmosphere in changeable bonding and interaction to adjoining components.^201,202 When the defect is minor, it can influence the device's properties, increasing the (device-to-device) variability yet keeping it under an appropriate working range. Pan and colleagues created an h-BN-based memristor employing a variety of electrode materials. They concluded that inserting metal cations into the 2D materials (h-BN) layers aided the RS and that B vacancies may improve. They devised the following ways to tailor the attributes of the devices and address the issue of device-to-device variability: (i) manipulating thickness, (ii) modifying the density of grain boundaries using a doped substrate approach and (iii) interfacial electrodes, which are composed of graphene are inserted. Therefore, the storage cells exhibited free restive switching, with a low set and reset voltage ratio (0.4 and 0.8 V, respectively) and a very high ON/OFF current ratio of up to six times greater with more than ten hours of capacity retention and also negligible (device-to-device) variability.¹⁷²

Villena and colleagues²⁰³ measured the variability of a metal/h-BN/metal memristor produced on various substrates (e.g., Pt, Cu, and Fe) by studying their electrical properties and resistive switching. They have utilized the CVD synthesis approach and formed 5-layer h-BN and >10-layer h-BN materials. This work focused on the problem of RS variability of device performance.²⁰⁴ Au/Ti/h-BN/Pt devices have been tested with identical Pt grain. These two devices had an extremely low level of variability. On the other hand, the electrical response of these two devices was different when constructed from different Pt grains. This fluctuation in electrical response was caused by differences in h-BN layers thickness on one Pt grain to the next, as shown. Whenever a comparable experiment was carried out on Cu grain-based (Au/Ti/h-BN/Cu) devices, the measured parameters revealed significantly reduced variability in six Au/Ti/h-BN/Cu devices, with low variability for (both set and reset) processes. The variability of iron-based (Au/Ag/h-BN/Fe) memristors has also been examined by applying a blend of experimental and simulation methods. Several RS cycles were performed on an Au/Ag/h-BN/Fe device (Fig. 18a). Fig. 18b demonstrates the probability distribution and cumulative probability of V SET and V RESET. When the Au/Ag/h-BN/Fe memristor data is compared to Au/Ti/h-BN/Cu memristor data, it can be shown that Au/Ag/h-BN/Fe devices have a somewhat larger variability. The calculations match well with the experimentally observed I–V sweepings in both RS (high and low), as illustrated in Fig. 18d and e. This consistency between experimental and simulation-based results shows that the h-BN stack has many conductive filaments. According to the calculations, quantum conduction (as predicted by QPC theory) is insignificant in this scenario, effective for high electrically conductive filaments. This finding supports the concept of the penetration of metallic ions from electrodes into h-BN to create the conductive filaments by confirming the Ohmic conduction of the conductive filaments. Inside the reduced voltage domain (about 0.6 V of L-RS), Fig. 14f shows Au/Ag/h-BN/Fe memristors (150 I–V curves).

Fig. 18 (a) The red curve presented as the first cycle about 150 sweeps of the Au/Ag/h-BN/Fe device, the set/reset voltages statistical analysis presented in (b and c). (d and e) I–V sweeps under HRS and LRS. (f) The I–V curves in the 0 V–0.6 V voltage range in the Au/Ag/h-BN/Fe device for linear regression fitting.¹⁹⁵ Reproduced from ref. 195 with permission from Elsevier, copyright [2019].

Chen and colleagues¹⁹⁹ fabricated an Au/h-BN/Au memristor device through the chemical vapor deposition process and realized its application as an artificial neural network aimed at image recognition. The bipolar RS characteristics were dominated in the fabricated devices as presented in Fig. 19, demonstrating 98% yield, 1.53% cyclic variability, and 5.74% device variability. The stochastic filamentary RS of one device is credited to the cycle-to-cycle variability. 1500 cycle to cycle statistical analyses for 48 memristor devices is shown in Fig. 19a. The set and reset voltage variability through the coefficient of variation (CV) was counted Fig. 19b. The lack of homogeneity in the device causes device-to-device variability, e.g., area, creases, and width variations of the active layer during assembling.^205,206 The lowest variability of cycles during set and reset was 2.02%, and the maximum was 9.61%. On the other hand, the variability value among devices (device to device) increased from 6.06% to 29.07% for 48 memristor devices. For CC = 1 μA, the V-SET and RESET values decreased to 1.53% and 6.21%, respectively, for cycle-to-cycle variability. Increasing the distribution of VSET and reset VREST for 16 samples at 1 mA CC (Fig. 19c). Guo and colleagues²⁰⁰ claimed the fabrication and testing of 80 Ag/SnO_x/SnSe memristor devices using an extremely consistent fabrication approach, with 100% yield, small set variability, and reset voltages shown in Fig. 19d.

Fig. 19 (a) Au/h-BN/Au memristive Cumulative distribution plot for 48 memristors (1500 cycles). (b) Increasing dispersal of device VSET and reset VREST for 48 samples at 1 mA. (c) Increasing distribution of VSET and reset VREST for 16 samples at 1 mA CC.¹⁹⁹ Reproduced from ref. 199 with permission from Springer Nature, copyright [2020]. (d) Ag/SnO_x/SnSe memristor statistical operation voltage of 80 as-fabricated devices.²⁰⁰ Reproduced from ref. 200 with permission from Elsevier, copyright [2020].

7. Challenges, possible approaches, and attitudes for 2D materials

Primarily, large-scale engineering of 2D material-based nanostructured memristor devices using low-cost approaches while achieving excellent results would be a significant challenge for the most mature businesses. Nonetheless, synthesis on a wafer scale with precise thickness is significant. With various CVD techniques, substantial surface area monolayer 2D material (such as WS₂ and MoS₂) films may be easily produced. However, the quality of these CVD-produced 2D material sheets is insufficient for industrial use. It is desirable to have wafer-scale single-layer 2D materials combined from huge single domain flakes with an aligned orientation. As a result, a thorough understanding of 2D materials growth behaviors is critical, especially if a severe layer-by-layer growth strategy with a high single-crystal domain size is to be developed. Immediately, a comprehensive assessment of critical structural characteristics and wafer-scale manufacturing techniques will be conducted.

Second, in an open environment, the reaction between a 2D material and a substrate material generates redox reactions and the formation of new materials, affecting the original characteristics of the materials. Consistency, variability, and stability are the most common issues. Memristors are non-volatile memory storage devices, and memristor devices employed as electric synapses must show altered gigabytes. What is the correct parameter for evaluating the high yield, consistency, variability, and stability of memristor devices that may not be evident. In general, there are a few easy rules to follow: (i) instead of a single device, statistical evidence concerning several important limits and statistics of performance (figure of merit) define the consistency among a number of the devices.; (ii) statistical data about numerous significant limits and figures of merit defining the consistency of the devices. To minimize scientific swings that go unreported in hypothetical papers, the importance of statistical studies must be addressed in a thorough way on 2D material-based devices. (iii) The effective contact between metal electrodes and material films has not been fully established, despite the need for dependable electrical contact in all device applications. To date, various contact methods for decreasing the contact resistance between metal electrodes and 2D material-based channels have been investigated. Device contact that is dependable and Ohmic contact are difficult to create. It is critical to provide a versatile and universal electrical contact design. Furthermore, despite the high light–matter interaction of 2D materials, light absorption is unavoidably poor due to their atomic thickness, limiting their photodetection effectiveness. Plasmonic and photonic structures can increase the device's light absorption and, therefore, its photodetection performance, leading to functional optoelectronic memristors and synapses.

8. Summary and outlook

The characteristics of the 2D material, its production, and applications in various types of memristors for neuromorphic applications are described in this review article. The performance of the memristor is primarily concerned when a specific active layer made of 2D materials is utilized. The switching voltage, ON/OFF current ratio during RS, cycle to cycle, and device to device variability, retention time, power consumption, and other properties of memristors should be considered carefully. It is concluded that we need to boost this to meet the authentic storage requirements significantly. As a result, to satisfy current requirements, we must concentrate our efforts on the synthetic approach for 2D materials and large-area integration. During the literate review, we have gone through several synthesis strategies for 2D material-based active layers. Various synthesis techniques for 2D material-based functionals and active layers include mechanical exfoliation, liquid-phase exfoliation, and CVD. The CVD can be used to make high-quality and single-sheet films, and this is an area of study that will be pursued in the future.

Second, understanding the operating process of the memristor's switching mechanism is critical to the research field. The structural characteristics of 2D materials have been addressed. According to the physical shape of the device, 2D materials offer an effective resistive switching mechanism (lateral or vertical). Typical switching processes are defect migration, phase transition, vacancy migration, filament creation, charge trapping, atomic vacancy, and trapping and de-trapping of electronic species. When a conductive filament develops inside a memristor device, it may be identified with atomic resolution under various switching conditions using an in situ probing approach and electron energy-loss spectroscopy with high 3-D resolution. A three-dimensional description of a conductive filament provides comprehensive information on its shape. The creation and extinction of conductive filaments should ideally be done at the atomic level. Aberration-corrected high-resolution TEM would be useful to detect the produced grain boundary structure. TEM can also be used to characterize atomic vacancy defects.

Third, a pioneering research topic is the 2D material-based memristor device, which is designed to store and supply data in the form of visual and electric signals. Several hybrid heterostructures have been developed to increase the number of data storage levels and improve the device's functionality. As a result, light can be introduced to switch the memristor device to determine the effectiveness of light sensitivity on performance. The optical and electrical characteristics of 2D materials can be affected by both infrared and ultraviolet light.¹⁴⁸

In summary, we concluded our understanding in this review that even for the most advanced companies fabricating 2D material-based nanosized devices at the wafer-level by utilizing scalable methods for accomplishing outstanding performance is a considerable challenge. When we compared device fabrication with the exfoliation approach, all of the mentioned performance metrics for device fabrication via synthetic approaches are still severely hampered. While research on mechanically exfoliated layered materials may be useful as a comparison to evaluate material performance, statistical analyses into the implementations of two-dimensional layered materials-based microelectronic devices are urgently needed. It is important to keep in mind that the variables and metrics of each device's merit are decided by its practical applications in the industry. For instance, the FET device used during logic gates must operate and encounter reliability criteria different from those used to control the power flow along with a memristor device. Correspondingly, memristors for use as non-volatile memory and memristors as electronic synaptic devices perform differently. In other words, while the literature contains extensive guidance on how to evaluate the synthesis, yield, fabrication, variability, and stability, we need to adopt the following basic criteria: (i) an appropriate synthesis approach, (ii) careful fabrication and characterization should be on multiple devices, (iii) offering statistical data including all metrics and figures of merit that impact the device's reliability, and (iv) moreover, instead of relying exclusively on intuition focused schematics, analyze and evaluate failures depending on nanometric accuracy and experimental measurements. To allow a technological change, the significance of statistics on two-dimensional devices must be emphasized on a larger scale, even though crucial technical challenges risk being neglected in scholarly papers.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (Grant No. 61974093), the Science and Technology Innovation Commission of Shenzhen (Grant No. RCYX20200714114524157, JCYJ20180507182042530 and JCYJ20180507182000722), and the NTUT-SZU Joint Research Program.

Notes and references

1. L. M. Yang, S. K. Su, H. P. Wong and L. J. Li, Nature, 2019, 567, 169–170.
2. D. Akinwande, C. Huyghebaert, C. H. Wang, M. I. Serna, S. Goossens, L. J. Li, H. S. Wong and F. H. Koppens, Nature, 2019, 573, 507–518.
3. Q. Zhao, Z. Xie, Y.-P. Peng, K. Wang, H. Wang, X. Li, H. Wang, J. Chen, H. Zhang and X. Yan, Mater. Horiz., 2020, 7, 1495–1518.
4. K. Wang, J. Chen and X. Yan, Nano Energy, 2021, 79, 105453.
5. G. Ding, Y. Wang, G. Zhang, K. Zhou, K. Zeng, Z. Li, Y. Zhou, C. Zhang, X. Chen and S. T. Han, Adv. Funct. Mater., 2019, 29, 1806637.
6. H. Li, L. Tao and J.-B. Xu, J. Appl. Phys., 2021, 129, 050902.
7. Z. Wang, S. Joshi, S. Savelev, W. Song, R. Midya, Y. Li, M. Rao, P. Yan, S. Asapu and Y. J. Zhuo, Nat. Mater., 2018, 1, 137–145.
8. R. Wang, J.-Q. Yang, J.-Y. Mao, Z.-P. Wang, S. Wu, M. Zhou, T. Chen, Y. Zhou and S.-T. Han, Adv. Intell. Syst., 2020, 2, 2000055.
9. Z. Wang, S. Joshi, S. E. Savelev, H. Jiang, R. Midya, P. Lin, M. Hu, N. Ge, J. P. Strachan and Z. J. Li, Nat. Mater., 2017, 16, 101–108.
10. A. I. Vlasov, I. V. Gudoshnikov, V. P. Zhalnin, A. T. Kadyr and V. A. Shakhnov, Entrepreneurship Sustainability Issues, 2020, 8, 98.
11. K. Sun, J. Chen and X. J. A. F. M. Yan, Adv. Funct. Mater., 2020, 8, 2006773.
12. L. Liu, C. Bing, B. Gao, F. Zhang, Y. Chen, X. Liu, W. Yi, R. Han and J. Kang, Appl. Phys. A, 2011, 102, 991–996.
13. D. B. Strukov, G. S. Snider, D. R. Stewart and W. J. Duncan, Nature, 2009, 459, 1154.
14. X. Yan, Y. Pei, H. Chen, J. Zhao, Z. Zhou, H. Wang, L. Zhang, J. Wang, X. Li and C. J. Qin, Adv. Funct. Mater., 2019, 31, 18052841–18052849.
15. Y. Zhan, Y. Chen, Q. Zhang, J. Zhuang, M. Tian, H. Chen, L. Zhang, H. Zhang, J. He and W. Wang, Nat. Chem. Biol., 2012, 8, 897–904.
16. U. Wurstbauer, B. Miller and E. Parzinger, J. Phys. D: Appl. Phys., 2017, 17, 173001.
17. J. Du, H. Yu, B. Liu, M. Hong, Q. Liao, Z. Zhang and Y. J. Zhang, Small Methods, 2021, 5, 2000919.
18. W. Sun, B. Gao, M. Chi, Q. Xia, J. J. Yang, H. Qian and H. J. Wu, Nat. Commun., 2019, 10, 1–13.
19. Y. Yang, G. Yuan, Z. Yan and W. Xubing, Adv. Mater., 2017, 26, 1700425.
20. J. Lindner, D. Burgler and S. J. Springer, Magnat. Nanostruct. Nanostruct., 2013, 44, 1–35.
21. K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. Grigorieva, S. Dubonos and A. J. Firsov, Nature, 2005, 438, 197–200.
22. H. Hamann, M. O. Boyle, Y. Martin, M. Rooks and M. Wickramasinghe, Nat. Mater., 2006, 5, 383–387.
23. S. B. Furber, D. R. Lester, L. A. Plana, J. D. Garside, E. Painkras, S. Temple and A. D. Brown, IEEE Trans. Comput., 2012, 62, 2454–2467.
24. E. M. Vinod, K. Ramesh, R. Ganesan and K. S. Sangunni, Appl. Phys. Lett., 2014, 104, 383.
25. J. H. Chang, H. H. Chen and C. R. J. Chang, J. Phys. D: Appl. Phys., 2012, 46, 035002.
26. C. A. Downing, A. R. Pearce, R. J. Churchill and M. E. Portnoi, Phys. Rev. B, 2015, 92, 165401.
27. H. M. Vo, S. N. Truong, S. Shin and K. S. Min, J. IKEEE, 2014, 18, 228–233.
28. T. W. H. Oates and S. J. Noda, Appl. Phys. Lett., 2009, 94, 053106.
29. M. J. Hollander, Y. Liu, W.-J. Lu, L.-J. Li, Y.-P. Sun, J. A. Robinson and S. J. Datta, ACS Nano Lett., 2015, 15, 1861–1866.
30. A. C. Torrezan, J. P. Strachan, G. Medeiros-Ribeiro and R. S. J. N. Williams, Nanotechnology, 2011, 22, 485203.
31. Y. Zhao, L. Chai, X. Yan, W. Huang, T. Fan, O. A. Al-Hartomy, A. Al-Ghamdi, S. Wageh, A. G. Al-Sehemi, Z. Xie and H. Zhang, Mater. Chem. Front., 2022 10.1039/D1QM01342D.
32. J. Mcmanus, T. Ito, V. Ferreira, T. Baumgartner, J. Mey, T. Weber, J. Crusius, C. Tems, W. Berelson and R. J. Thunell, Science, 2014, 345, 665–668.
33. M. J. Lee, B. L. Chang, D. Lee, S. R. Lee, C. Man, H. H. Ji, Y. B. Kim, C. J. Kim, D. H. Seo and S. J. Seo, Nat. Mater., 2011, 10, 625–630.
34. R. Chen, T. Zhao, W. Wu, F. Wu, L. Li, J. Qian, R. Xu, H. Wu, H. M. Albishri and A. Al-Bogami, Nano Lett., 2014, 14, 5899–5904.
35. A. D. Franklin, Science, 2015, 349, 2750.
36. T. Skotnicki, J. A. Hutchby, T.-J. King, H.-S. Wong and F. Boeuf, IEEE Circ. Devices Mag., 2005, 21, 16–26.
37. D. Akinwande, C. Huyghebaert, C.-H. Wang, M. I. Serna, S. Goossens, L.-J. Li, H.-S. P. Wong and F. H. Koppens, Nature, 2019, 573, 507–518.
38. W. Huh, D. Lee and C. H. Lee, Adv. Mater., 2020, 51, 2002092.
39. M. Chhowalla, H. S. Shin, G. Eda, L.-J. Li, K. P. Loh and H. Zhang, Nat. Chem., 2013, 5, 263–275.
40. N. Smith, S. Kevan and F. Disalvo, J. Phys. C: Solid State Phys., 1985, 18, 3175.
41. K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva and A. A. Firsov, Science, 2004, 306, 666–669.
42. K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. Grigorieva, S. Dubonos and A. A. Firsov, Nature, 2005, 438, 197–200.
43. K. S. Novoselov, D. Jiang, F. Schedin, T. Booth, V. Khotkevich, S. Morozov and A. K. Geim, Proc. Natl. Acad. Sci. U. S. A., 2005, 102, 10451–10453.
44. C. R. Dean, A. F. Young, I. Meric, C. Lee, L. Wang, S. Sorgenfrei, K. Watanabe, T. Taniguchi, P. Kim and K. L. Shepard, Nat. Nanotechnol., 2010, 5, 722–726.
45. Q. Lu and R. Huang, Int. J. Appl. Mech., 2009, 1, 443–467.
46. Q. Lu, M. Arroyo and R. Huang, J. Phys. D: Appl. Phys., 2009, 42, 102002.
47. T. J. Ko, H. Li, S. A. Mofid, C. Yoo, E. O. ue, S. S. Han, M. S. Shawkat, A. Krishnaprasad, M. M. Islam and D. Dev, Science, 2020, 11, 101676.
48. A. J. Mannix, B. Kiraly, M. C. Hersam and N. P. Guisinger, Nat. Rev. Chem., 2017, 1, 1–14.
49. T. Hu, X. Mei, Y. Wang, X. Weng and M. J. S. B. Wei, Sci. Bull., 2019, 64, 1707–1727.
50. M. Lanza, Q. Smets, C. Huyghebaert and L. J. Li, Nat. Commun., 2020, 11, 1–5.
51. M.-J. Lee, C. B. Lee, D. Lee, S. R. Lee, M. Chang, J. H. Hur, Y.-B. Kim, C.-J. Kim, D. H. Seo and S. Seo, Nat. Mater., 2011, 10, 625–630.
52. K. F. Mak and J. Shan, Nat. Photonics, 2016, 10, 216–226.
53. J. Hong, K. Li, C. Jin, X. Zhang, Z. Zhang and J. Yuan, Phys. Rev. B, 2016, 93, 075440.
54. C. Gong, Y. Zhang, W. Chen, J. Chu, T. Lei, J. Pu, L. Dai, C. Wu, Y. Cheng and T. Zhai, Adv. Sci., 2017, 4, 1700231.
55. X. Zhu, D. Li, X. Liang and W. D. Lu, Nat. Mater., 2019, 18, 141–148.
56. N. L. Rodenhouse, W. H. Lowe, R. L. Gebauer, K. P. McFarland and M. S. Bank, Sci. Total Environ., 2019, 665, 1125–1134.
57. V. K. Sangwan, H.-S. Lee, H. Bergeron, I. Balla, M. E. Beck, K.-S. Chen and M. C. Hersam, Nature, 2018, 554, 500–504.
58. X. Yan, K. Wang, J. Zhao, Z. Zhou, H. Wang, J. Wang, L. Zhang, X. Li, Z. Xiao and Q. Zhao, Small, 2019, 15, 1900107.
59. M. M. Rehman, J. Z. Gul, H. M. M. U. Rehman, W. Y. Kim and N. J. S. Ahmed, Sci. Technol. Adv. Mater., 2020, 21, 147–186.
60. J. H. Han, M. Kwak, Y. Kim and J. J. C. R. Cheon, Chem. Rev., 2018, 118, 6151–6188.
61. M. Bellardita, A. Di Paola, S. Yurdakal and L. Palmisano, Heterogen. Photocatal., 2019, 1, 25–56.
62. X. Yan, Q. Zhao, A. P. Chen, J. Zhao, Z. Zhou, J. Wang, H. Wang, L. Zhang, X. Li and Z. J. Xiao, Small, 2019, 15, 1901423.
63. W. Huh, S. Jang, J. Y. Lee, D. Lee, D. Lee, J. M. Lee, H. G. Park, J. C. Kim, H. Y. Jeong and G. J. A. M. Wang, Adv. Mater., 2018, 30, 1801447.
64. M. Kumar, D. K. Ban, S. M. Kim, J. Kim and C. P. Wong, Adv. Electron. Mater., 2019, 5, 1900467.
65. J. J. Liu, Z. Zeng, X. Cao, G. Lu, L. H. Wang, Q. L. Fan, W. Huang and H. Zhang, Small, 2012, 8, 3517–3522.
66. R. Ge, X. Wu, M. Kim, J. Shi, S. Sonde, L. Tao, Y. Zhang, J. C. Lee and D. Akinwande, ACS Nano Lett., 2018, 18, 434–441.
67. M. D. Pickett, J. Borghetti, J. J. Yang, G. Medeiros-Ribeiro and R. S. Williams, Adv. Mater., 2011, 23, 1730–1733.
68. X. Zhang, H. Xie, Z. Liu, C. Tan, Z. Luo, H. Li, J. Lin, L. Sun, W. Chen and Z. J. Xu, Angew. Chem., Int. Ed., 2015, 54, 3653–3657.
69. Q. A. Vu, Y. S. Shin, Y. R. Kim, V. L. Nguyen, W. T. Kang, H. Kim, D. H. Luong, I. M. Lee, K. Lee and D.-S. Ko, Nat. Commun., 2016, 7, 1–8.
70. R. Xu, H. Jang, M.-H. Lee, D. Amanov, Y. Cho, H. Kim, S. Park, H.-J. Shin and D. J. Ham, ACS Nano Lett., 2019, 19, 2411–2417.
71. X. Zhang, Z. Lai, Z. Liu, C. Tan, Y. Huang, B. Li, M. Zhao, L. Xie, W. Huang and H. J. Zhang, Angew. Chem., Int. Ed., 2015, 54, 5425–5428.
72. F. Zhang, H. Zhang, S. Krylyuk, C. A. Milligan, Y. Zhu, D. Y. Zemlyanov, L. A. Bendersky, B. P. Burton, A. V. Davydov and J. Appenzeller, Nat. Mater., 2019, 18, 55–61.
73. L. Valentini, M. Cardinali, E. Fortunati and J. Kenny, Appl. Phys. Lett., 2014, 105, 153111.
74. C. R. Dean, A. F. Young, I. Meric, C. Lee, L. Wang, S. Sorgenfrei, K. Watanabe, T. Taniguchi, P. Kim and K. L. Shepard, Nat. Nanotechnol., 2010, 5, 722–726.
75. M. Kim, R. Ge, X. Wu, X. Lan, J. Tice, J. C. Lee and D. J. Akinwande, Nat. Commun., 2018, 9, 1–7.
76. M. Yoshida, R. Suzuki, Y. Zhang, M. Nakano and Y. Iwasa, Sci. Adv., 2015, 1, 1500606.
77. Y. Liu, N. O. Weiss, X. Duan, H.-C. Cheng, Y. Huang and X. J. Duan, Nat. Rev., 2016, 1, 1–17.
78. A. J. Rockett, Mater. Sci. Semicond., 2008, 1, 505–572.
79. J.-H. Lee, E. K. Lee, W.-J. Joo, Y. Jang, B.-S. Kim, J. Y. Lim, S.-H. Choi, S. J. Ahn, J. R. Ahn and M.-H. J. S. Park, Science, 2014, 344, 286–289.
80. J. J. Yang, M. D. Pickett, X. Li, D. A. Ohlberg, D. R. Stewart and R. S. Williams, Nat. Nanotechnol., 2008, 3, 429–433.
81. S. Bhattacharjee, R. Wigchering, H. G. Manning, J. J. Boland and P. K. J. Hurley, Sci. Rep., 2020, 10, 1–8.
82. R. J. Van, M. M. Masis and E. Miranda, Quantum Point Contact Conduct., 2016, 22, 197–224.
83. M. Zidan, H. Omran, R. Naous, A. Sultan, H. Fahmy, W. Lu and K. Salama, Sci. Rep., 2016, 1, 1–9.
84. F. Alibart, L. Gao, B. D. Hoskins and D. B. Strukov, Nanaotechnology, 2012, 23, 075201.
85. S. Choi, P. Sheridan and W. D. Lu, Sci. Rep., 2015, 5, 1–10.
86. X. Cai, Y. Luo, B. Liu and H.-M. J. C. S. R. Cheng, Chem. Soc. Rev., 2018, 47, 6224–6266.
87. Y. Wang, Y. Zhao, X. Ding and L. Qiao, J. Energy Chem., 2021, 60, 451–479.
88. F. Wang, Z. Wang, T. A. Shifa, Y. Wen, F. Wang, X. Zhan, Q. Wang, K. Xu, Y. Huang and L. J. Yin, Adv. Funct. Mater., 2017, 19, 1603254.
89. F. Schedin, A. K. Geim, S. V. Morozov, E. Hill, P. Blake, M. Katsnelson and K. S. Novoselov, Nat. Mater., 2007, 6, 652–655.
90. M. Yi and Z. Shen, J. Mater. Chem. A, 2015, 3, 11700–11715.
91. H. Li, J. Wu, Z. Yin and H. J. Zhang, Acc. Chem. Res., 2014, 47, 1067–1075.
92. G. S. Shmavonyan, G. Sevoyan and V. J. Aroutiounian, Armenian J. Phys., 2013, 6, 1–6.
93. Y. Peng, Z. Wang and K. J. L. Zou, Langmuir, 2015, 31, 7782–7791.
94. Y. Zhang, J. P. Small, W. Pontius and P. Kim, Appl. Phys. Lett., 2005, 86, 073104.
95. Y. Hernandez, V. Nicolosi, M. Lotya, F. M. Blighe, Z. Sun, S. De, I. McGovern, B. Holland, M. Byrne and Y. K. Gun'Ko, Nat. Nanotechnol., 2008, 3, 563–568.
96. J. N. Coleman, M. Lotya, A. O’Neill, S. D. Bergin, P. J. King, U. Khan, K. Young, A. Gaucher, S. De and R. J. Smith, Science, 2011, 331, 568–571.
97. S. Jiang, M. Q. Arguilla, N. D. Cultrara and J. E. Goldberger, Acc. Chem. Res., 2015, 48, 144–151.
98. S. Jiang, S. Butler, E. Bianco, O. D. Restrepo, W. Windl and J. E. Goldberger, Nat. Commun., 2014, 5, 1–6.
99. X. Ling, H. Wang, S. Huang, F. Xia and M. S. Dresselhaus, Proc. Natl. Acad. Sci. U. S. A., 2015, 112, 4523–4530.
100. X. Li, W. Cai, J. An, S. Kim, J. Nah, D. Yang, R. Piner, A. Velamakanni, I. Jung and E. J. Tutuc, Science, 2009, 324, 1312–1314.
101. O. Penkov, H.-J. Kim, H.-J. Kim and D.-E. Kim, Int. J. Precis. Eng. Manuf., 2014, 15, 577–585.
102. V. Singh, D. Joung, L. Zhai, S. Das, S. I. Khondaker and S. J. Seal, Prog. Mater. Sci., 2011, 56, 1178–1271.
103. Y. Zhang, L. Zhang and C. J. Zhou, Acc. Chem. Res., 2013, 46, 2329–2339.
104. P. R. Somani, S. P. Somani and M. J. Umeno, Chem. Phys. Lett., 2006, 430, 56–59.
105. K. S. Novoselov, A. Mishchenko, A. Carvalho and A. Neto, Science, 2016, 353, 9439.
106. K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva and A. Firsov, Science, 2004, 306, 666–669.
107. A. K. Geim and K. S. Novoselov, Nature, 2010, 1, 11–19.
108. Y. Zhang, T.-T. Tang, C. Girit, Z. Hao, M. C. Martin, A. Zettl, M. F. Crommie, Y. R. Shen and F. Wang, Nature, 2009, 459, 820–823.
109. K. Novoselov, A. Mishchenko, A. Carvalho and A. C. Neto, Science, 2016, 353, 6298.
110. J.-W. Chen, S.-T. Lo, S.-C. Ho, S.-S. Wong, X.-Q. Zhang, Y.-D. Liu, Y.-Y. Chiou, Y.-X. Chen, J.-C. Yang and Y.-C. Chen, Nat. Commun., 2018, 9, 1–7.
111. S. Haigh, A. Gholinia, R. Jalil, S. Romani, L. Britnell, D. Elias, K. Novoselov, L. Ponomarenko, A. Geim and R. J. Gorbachev, Nat. Mater., 2012, 11, 764–767.
112. F. Xia, H. Wang and Y. Jia, Nat. Commun., 2014, 5, 1–6.
113. J. Ye, X. Hu and G. J. C. S. B. Sun, Chin. Sci. Bull., 2019, 64, 1620–1636.
114. K. Liu, P. Luo, W. Han, S. Yang, S. Zhou, H. Li and T. J. S. B. Zhai, Sci. Bull., 2019, 64, 1426–1435.
115. J. Pei, X. Gai, J. Yang, X. Wang, Z. Yu, D.-Y. Choi, B. Luther-Davies and Y. J. Lu, Nat. Commun., 2016, 7, 1–8.
116. J. Wan, S. D. Lacey, J. Dai, W. Bao, M. S. Fuhrer and L. J. Hu, Chem. Soc. Rev., 2016, 45, 6742–6765.
117. Z. Dai, L. Liu and Z. J. A. M. Zhang, Adv. Mater., 2019, 31, 1805417.
118. W. H. Pernice and H. J. Bhaskaran, Appl. Phys. Lett., 2012, 101, 171101.
119. J. R. Brent, A. K. Ganguli, V. Kumar, D. J. Lewis, P. D. McNaughter, P. O'Brien, P. Sabherwal and A. Tedstone, RSC Adv., 2016, 6, 86955–86958.
120. S. Yang, G. Lee, J. Kim, S. Yang, C.-H. Lee and J. C. Kim, J. Mater. Chem. C, 2020, 8, 8393–8398.
121. C. Lan, Z. Zhou, Z. Zhou, C. Li, L. Shu, L. Shen, D. Li, R. Dong, S. P. Yip and H. Johnny, Nano Res., 2018, 11, 331–3384.
122. C. Cong, J. Shang, X. Wu, B. Cao, N. Peimyoo, C. Qiu, L. Sun and T. Yu, Adv. Opt. Mater., 2014, 2, 131–136.
123. A. M. Van Der Zande, P. Y. Huang, D. A. Chenet, T. C. Berkelbach, Y. You, G.-H. Lee, T. F. Heinz, D. R. Reichman, D. A. Muller and J. C. Hone, Nat. Mater., 2013, 12, 554–561.
124. H. Han, Z. Xu, K. Guo, Y. Ni, M. Ma, H. Yu, H. Wei, J. Gong, S. Zhang and W. J. Xu, Adv. Intell. Syst., 2020, 2, 1900176.
125. V. K. Sangwan and M. C. Hersam, Nat. Nanotechnol., 2020, 15, 517–528.
126. H. Qiu, T. Xu, Z. Wang, W. Ren, H. Nan, Z. Ni, Q. Chen, S. Yuan, F. Miao and F. J. Song, Nat. Commun., 2013, 4, 1–6.
127. A. Sobhani, M. W. Knight, Y. Wang, B. Zheng, N. S. King, L. V. Brown, Z. Fang, P. Nordlander and N. J. Halas, Nat. Commun., 2013, 4, 1–6.
128. Y. Shi, X. Liang, B. Yuan, V. Chen, H. Li, F. Hui, Z. Yu, F. Yuan, E. Pop and H.-S. Wong, Nat. Electron., 2018, 1, 458–465.
129. X. Wang, M. Hossain, Z. Wei and L. J. Xie, Nanotechnology, 2018, 30, 034003.
130. M. Wang, S. Cai, C. Pan, C. Wang, X. Lian, Y. Zhuo, K. Xu, T. Cao, X. Pan and B. J. Wang, Nat. Electron., 2018, 1, 130–136.
131. S. Yin, C. Song, Y. Sun, L. Qiao, B. Wang, Y. Sun, K. Liu, F. Pan and X. J. Zhang, ACS Mater. Interfaces, 2019, 11, 43344–43350.
132. X. Yan, L. Zhang, H. Chen, X. Li, J. Wang, Q. Liu, C. Lu, J. Chen, H. Wu and P. J. Zhou, Adv. Funct. Mater., 2018, 28, 1803728.
133. K. Zhu, X. Liang, B. Yuan, M. A. Villena, C. Wen, T. Wang, S. Chen, F. Hui, Y. Shi and M. J. Lanza, ACS Mater. Interfaces, 2019, 11, 37999–38005.
134. M. S. Sokolikova and C. J. Mattevi, Chem. Soc. Rev., 2020, 49, 3952–3980.
135. I. Faraduan, I. Handayani, D. Diandra, H. Delima and I. J. Fathona, Mater. Res., 2021, 8, 016409.
136. P. Li, W. Wu, Y. Xu, J. Liu, S. Wu, Y. Ye, C. Liang and X. C. Zeng, J. Phys. Chem. Lett., 2021, 12, 1058–1065.
137. Y. Tian, Z. Guo, Z. Liu, H. Lin, X. Li, J. Chen, S. Deng and F. J. Liu, Mater. Today NANO, 2021, 14, 100112.
138. Y. Guo, K. Xu, C. Wu, J. Zhao and Y. J. Xie, Chem. Soc. Rev., 2015, 44, 637–646.
139. Q. Fang, Z. Zhang, Q. Ji, S. Zhu, Y. Gong, Y. Zhang, J. Shi, X. Zhou, L. Gu and Q. J. Wang, Nano Res., 2017, 10, 2761–2771.
140. Z. Shen, C. Zhao, Y. Qi, W. Xu, Y. Liu, I. Z. Mitrovic, L. Yang and C. J. Zhao, Nanomaterials, 2020, 10, 1437.
141. G. Cao, P. Meng, J. Chen, H. Liu, R. Bian, C. Zhu, F. Liu and Z. J. Liu, Adv. Funct. Mater., 2020, 4, 2005443.
142. N. Balakrishnan, E. D. Steer, E. F. Smith, Z. R. Kudrynskyi, Z. D. Kovalyuk, L. Eaves, A. Patanè and P. H. Beton, 2D Mater., 2018, 5, 035026.
143. B. Holst, G. Alexandrowicz, N. Avidor, G. Benedek, G. Bracco, W. E. Ernst, D. Farías, A. P. Jardine, K. Lefmann and J. R. Manson, Phys. Chem. Chem. Phys., 2021, 3, 7653–7672.
144. Y. Zhai, X. Yang, F. Wang, Z. Li, G. Ding, Z. Qiu, Y. Wang, Y. Zhou and S. T. Han, Adv. Mater., 2018, 30, 1803563.
145. M. Wang, S. Cai, C. Pan, C. Wang, X. Lian, Y. Zhuo, K. Xu, T. Cao, X. Pan and B. Wang, Nat. Electron., 2018, 1, 130–136.
146. W. Xu, J. Jiang, H. Ma, Z. Zhang, J. Li, B. Zhao, R. Wu, X. Yang, H. Zhang and B. J. N. R. Li, Nano Res., 2020, 13, 2091–2097.
147. Y. Li, M. D. Cole, Y. Gao, T. Emrick, Z. Xu, Y. Liu and T. P. J. A. A. E. M. Russell, ACS Appl. Energy, 2019, 2, 1634–1641.
148. L. Zhang, T. Gong, H. Wang, Z. Guo and H. Zhang, Nanoscale, 2019, 11, 12413–12435.
149. C. Liu, H. Chen, S. Wang, Q. Liu, Y.-G. Jiang, D. W. Zhang, M. Liu and P. J. N. N. Zhou, Nat. Nanotechnol., 2020, 15, 545–557.
150. V. K. Sangwan, H.-S. Lee, H. Bergeron, I. Balla, M. E. Beck, K.-S. Chen and M. C. Hersam, Nature, 2018, 554, 500–504.
151. V. K. Sangwan, D. Jariwala, I. S. Kim, K.-S. Chen, T. J. Marks, L. J. Lauhon and M. C. Hersam, Nat. Nanotechnol., 2015, 10, 403–406.
152. Y. Shi, X. Liang, B. Yuan, V. Chen, H. Li, F. Hui, Z. Yu, F. Yuan, E. Pop and H.-S. Wong, Nat. Electron., 2018, 1, 458–465.
153. R. Xu, H. Jang, M.-H. Lee, D. Amanov, Y. Cho, H. Kim, S. Park, H.-J. Shin and D. Ham, Nano Lett., 2019, 19, 2411–2417.
154. Y. Yang, P. Gao, S. Gaba, T. Chang, X. Pan and W. Lu, Nat. Commun., 2012, 3, 732.
155. T. Driscoll, H.-T. Kim, B.-G. Chae, M. Di Ventra and D. J. Basov, Appl. Phys. Lett., 2009, 95, 043503.
156. B. J. Choi, A. C. Torrezan, J. P. Strachan, P. Kotula, A. Lohn, M. J. Marinella, Z. Li, R. S. Williams and J. Yang, Adv. Funct. Mater., 2016, 26, 5290–5296.
157. J. Yin, F. Zeng, Q. Wan, F. Li, Y. Sun, Y. Hu, J. Liu, G. Li and F. J. Pan, Adv. Funct. Mater., 2018, 28, 1706927.
158. Y. Li, M. D. Cole, Y. Gao, T. Emrick, Z. Xu, Y. Liu and T. P. J. A. A. E. M. Russell, ACS Appl. Energy, 2019, 2, 1634–1641.
159. Y. Sun, C. Song, J. Yin, X. Chen, Q. Wan, F. Zeng and F. J. Pan, ACS Appl. Mater. Interfaces, 2017, 9, 34064–34070.
160. J. Zhao, Z. Zhou, Y. Zhang, J. Wang, L. Zhang, X. Li, M. Zhao, H. Wang, Y. Pei and Q. Zhao, J. Mater. Chem. C, 2019, 7, 1298–1306.
161. X. Yan, C. Qin, C. Lu, J. Zhao, R. Zhao, D. Ren, Z. Zhou, H. Wang, J. Wang and L. Zhang, ACS Appl. Mater. Interfaces, 2019, 11, 48029–48038.
162. A. A. Bessonov, M. N. Kirikova, D. I. Petukhov, M. Allen, T. Ryhänen and M. J. Bailey, Nat. Mater., 2015, 14, 199–204.
163. W. Huh, S. Jang, J. Y. Lee, D. Lee, D. Lee, J. M. Lee, H. G. Park, J. C. Kim, H. Y. Jeong and G. Wang, Adv. Mater., 2018, 30, 1801447.
164. I. Boybat, M. Le Gallo, S. Nandakumar, T. Moraitis, T. Parnell, T. Tuma, B. Rajendran, Y. Leblebici, A. Sebastian and E. Eleftheriou, Nat. Commun., 2018, 9, 1–12.
165. T. Z. Hua, R. Yang, K. Terabe, X. B. Yin, X. D. Zhang and X. Guo, Adv. Mater., 2016, 28, 377–384.
166. N. C. Frey, D. Akinwande, D. Jariwala and V. B. Shenoy, ACS Nano, 2020, 14, 13406–13417.
167. C.-H. Huang, J.-S. Huang, C.-C. Lai, H.-W. Huang, S.-J. Lin and Y.-L. Chueh, ACS Appl. Mater. Interfaces, 2013, 5, 6017–6023.
168. R. Lacerda, J. Ender, S. Fiorentini, W. Goes, S. Selberherr and V. Sverdlov, IEEE J. Electron Devices Soc., 2019, 9, 61–67.
169. C. He, J. Li, X. Wu, P. Chen, J. Zhao, K. Yin, M. Cheng, W. Yang, G. Xie and D. J. Wang, Adv. Mater., 2013, 25, 5593–5598.
170. J. Lee, C. Du, K. Sun, E. Kioupakis and W. D. Lu, ACS Nano, 2016, 10, 3571–3579.
171. P. Cheng, K. Sun and Y. H. Hu, ACS Nano Lett., 2016, 16, 572–576.
172. C. Pan, Y. Ji, N. Xiao, F. Hui, K. Tang, Y. Guo, X. Xie, F. M. Puglisi, L. Larcher and E. J. A. Miranda, Adv. Funct. Mater., 2017, 27, 1604811.
173. E. Yoo, M. Lyu, J.-H. Yun, C. Kang, Y. Choi and L. J. Wang, J. Mater. Chem. C, 2016, 4, 7824–7830.
174. E. J. Yoo, M. Lyu, J. H. Yun, C. J. Kang, Y. J. Choi and L. J. A. M. Wang, Adv. Mater., 2015, 27, 6170–6175.
175. C. Gu and J. S. Lee, ACS Nano, 2016, 10, 5413–5418.
176. B. Magyari-Kope, M. Tendulkar, S.-G. Park, H. D. Lee and Y. Nishi, Nanotechnology, 2011, 22, 254029.
177. V. K. Sangwan and M. C. Hersam, Nat. Nanotechnol., 2020, 15, 517–528.
178. S. Li, M. E. Pam, Y. Li, L. Chen, Y. C. Chien, X. Fong, D. Chi and K. W. Ang, Adv. Mater., 2021, 2103376 DOI:10.1002/adma.202103376.
179. S. G. Kim, J. S. Han, H. Kim, S. Y. Kim and H. W. Jang, Adv. Mater. Technol., 2018, 3, 1800457.
180. R. Devi Chandra and K. G. Gopchandran, ACS Appl. Electron. Mater., 2021, 3, 3846–3854.
181. T. Tuma, A. Pantazi, M. L. Gallo, A. Sebastian and E. J. Eleftheriou, Nat. Nanotechnol., 2016, 11, 693–699.
182. A. G. Andreou, R. C. Meitzler, K. Strohbehn and K. J. Boahen, Neural Netw., 1995, 8, 1323–1347.
183. A. Calimera, E. Macii and M. J. Poncino, Funct. Neurol., 2013, 28, 191.
184. I. A. Fleidervish, N. Lasser-Ross, M. J. Gutnick and W. N. Ross, Nature, 2010, 13, 852–860.
185. D. Kuzum, S. Yu and H. P. Wong, Nanotechnology, 2013, 24, 382001.
186. L. O. Chua, IEEE Trans. Circ. Theory, 1971, 18, 507–519.
187. A. Krishnaprasad, N. Choudhary, S. Das, D. Dev and P. L. Roy, Appl. Phys. Lett., 2019, 115, 103104.
188. R. A. J. A. i. Brooks, Artif. Intell., 1991, 47, 139–159.
189. S. Chen, Z. Lou, D. Chen and G. J. Shen, Adv. Mater., 2018, 30, 17054000.
190. H. L. Park, Y. Lee, N. Kim, D. G. Seo, G. T. Go and T. W. J. A. M. Lee, Adv. Mater., 2020, 32, 1903558.
191. K. Zhu, X. Liang, B. Yuan, M. A. Villena and M. J. Lanza, ACS Appl. Mater. Interfaces, 2019, 11, 37999–38005.
192. X. Lian, X. Shen, M. Zhang, J. Xu, F. Gao, X. Wan, E. Hu, Y. Guo, J. Zhao and Y. J. Tong, Appl. Phys. Lett., 2019, 115, 063501.
193. A. J. Arnold, A. Razavieh, J. R. Nasr, D. S. Schulman, C. M. Eichfeld and S. Das, ACS Nano, 2017, 11, 3110–3118.
194. Y. Chen, G. Gao, J. Zhao, H. Zhang, J. Yu, X. Yang, Q. Zhang, W. Zhang, S. Xu and J. J. A. F. M. Sun, Adv. Funct. Mater., 2019, 29, 1900959.
195. M. A. Villena, F. Hui, X. Liang, Y. Shi and M. Lanza, Microelectron. Reliab., 2019, 102, 113410.
196. M. M. Farronato, M. Melegari, S. Ricci, S. Hashemkhani, A. Bricalli and D. J. A. E. M. Ielmini, Adv. Electron. Mater., 2022, 2101161,  DOI:10.1002/aelm.202101161.
197. M. Lanza, Q. Smets, C. Huyghebaert and L.-J. Li, Nat. Commun., 2020, 11, 1–5.
198. Z. Lin, Y. Liu, U. Halim, M. Ding, Y. Liu, Y. Wang, C. Jia, P. Chen, X. Duan and C. J. Wang, Nature, 2018, 562, 254–258.
199. S. Chen, M. R. Mahmoodi, Y. Shi, C. Mahata, B. Yuan, X. Liang, C. Wen, F. Hui, D. Akinwande and D. B. Strukov, Nat. Electron., 2020, 3, 638–645.
200. J. Guo, L. Wang, Y. Liu, Z. Zhao, E. Zhu, Z. Lin, P. Wang, C. Jia, S. Yang and S. J. Lee, Matter, 2020, 2, 965–976.
201. Z. Zeng, Z. Yin, X. Huang, H. Li, Q. He, G. Lu, F. Boey and H. J. A. Zhang, Angew. Chem., Int. Ed., 2011, 123, 11289–11293.
202. M. Lanza, H. S. P. Wong, E. Pop, D. Ielmini, D. Strukov, B. C. Regan, L. Larcher, M. A. Villena, J. J. Yang and L. J. Goux, Adv. Electron. Mater., 2019, 5, 1800143.
203. M. A. Villena, F. Hui, X. Liang, Y. Shi, B. Yuan, X. Jing, K. Zhu, S. Chen and M. Lanza, Microelectron. Reliab., 2019, 102, 113410.
204. Y. Xuan, Y. Wu and P. Ye, IEEE Electron Device Lett., 2008, 29, 294–296.
205. S. Gaba, P. Sheridan, J. Zhou, S. Choi and W. J. Lu, Nanoscale, 2013, 5, 5872–5878.
206. M. Lanza, H. Wong, E. Pop, D. Ielmini, D. Strukov, B. Regan, L. Larcher, M. Villena, J. Yang and L. Goux, Adv. Electron. Mater., 2019, 1, 1800143.

---