Introductory lecture: origins and applications of efficient visible photoluminescence from silicon-based nanostructures

A variety of silicon-based nanostructures with dimensions in the 1–5 nm range now emit tunable photoluminescence (PL) spanning the visible range. Achievement of high photoluminescence quantum efficiency (PLQY) relies critically on their surface chemistry passivation and an impressive “tool box” of options have been developed. Two distinct PL bands are dominant. The “S-Band” (red–green emission with Slow microsecond decay) has PLQY that has steadily improved from 3% in 1990 to 65 5% by 2017. The “F-Band” (blue–yellow with Fast nanosecond decay) has reported PLQY values that have improved from 0.1% in 1994 to as high as 90% by 2016. The vast literature on both bands is surveyed and for the S-band, size-structure-PL correlations and selective photo-excitation studies are highlighted. Resonant photoexcitation and single quantum dot studies have revealed the key role of quantum confinement and the excitonic phonon-assisted nature of the radiative transitions. For the F-band, in contrast, specific phenomenological studies are highlighted that demonstrate similar emission without the presence of silicon nanostructures. Low PLQY F-band emission from pure silicon–silica core shell systems is probably associated with oxide-related defects, but ultrahigh PLQY from many lower temperature synthesis routes is likely to be from carbon-based nanostructures or chromophores, not silicon nanostructures. Potential applications for both PL bands include sensing, medical imaging, theranostics, photovoltaics, LED colour converters and nanothermometry. Emerging “green” synthesis routes are mentioned. If scalability and cost are significantly improved then a number of other proposed uses of ultra-efficient PL from “nano-Si” could become viable in cosmetics, catalysis, security and forensics.


Scope of review
Nanostructuring has introduced an exciting new capability to semiconducting silicon: the ability to emit visible light very efficiently under photoexcitation (photoluminescence (PL)). This property was quickly recognized in the early 1990s as a route to extend the functionality of an already powerful silicon technology. The topic has thus received intense study over the last 30 years, with now thousands of publications. This review focuses on the visible photoluminescence of silicon nanostructures; collating both experimental and theoretical work on most types of nanostructure; providing quantitative performance metrics, particularly on photoluminescence quantum yield (PLQY); photoluminescence wavelength tunability (PLWT) and photoluminescence decay times (PLDT).
There are two dominant PL emissions in the visible (see Fig. 1): the "Slow" (S) band in the red-yellow spectral range with long microsecond decay times (see Section 2) and the "Fast" (F) band in the blue-green with faster nanosecond decay times (see Section 3). In a given nanostructure, one or the other, or both can be present. Oen, as shown in Fig. 1, the intensity of the blue emission will oen increase with storage time. The origins of such behaviour have received intense study.
This review highlights many spectroscopy-based studies and selected phenomenological studies that provide insight into their likely PL mechanisms. The review also covers six application areas of efficient visible PL: sensing, photovoltaics, medical imaging, theranostics, white LEDs and nanothermometry. It also proposes a number of ways in which the eld might further progress, and other application areas that might emerge if cost and scalability issues are surmounted. There is no detailed discussion of nanostructure synthesis techniques and surface chemistry manipulation, as these have already been the focus of many prior reviews that are referenced.
Data on electrical transport, electroluminescence, optical gain, and thereby optoelectronics, are also omitted in order to entirely focus here on the potential of photoluminescent properties. Such a platform property is however oen combined with other crucial ones for specic uses, such as biodegradability and biocompatibility for medical applications. These additional properties are covered briey in the relevant application sections. In summary, this review collates and discusses some of the very large amount of data now available on many types of photoluminescent silicon nanostructures and many applications thereof. It is hoped to be of use both as a broad introduction to the vast literature of this maturing eld, but should also stimulate specic areas of discussion and perhaps further research in highlighted topics.

Electron-hole recombination processes in bulk silicon
When a semiconductor such as silicon is stimulated by light (photoexcitation) it may return to equilibrium by a variety of processes. 2 Some involve the re-emission of light (photoluminescence or uorescencethese terms are used interchangeably in the literature on luminescent silicon nanostructures) and are so called radiative transitions. Others redistribute the absorbed energy amongst vibrational modes of the lattice or electronic excitations and ultimately generate heatso-called non-radiative transitions.
Radiative processes in silicon include recombination from electrons and holes 3 free excitons, 4 biexcitons or excitonic molecules, 5-7 polyexcitons, 8 electronhole liquids, 9,10 bound excitons 11 and donor-acceptor pair luminescence. [12][13][14] Free excitons will diffuse through the silicon lattice 15 and can be trapped and emit light at isolated dopants, 16 clusters of dopants, 17 isoelectronic traps 18 and sometimes near extended defects [19][20][21][22] such as dislocations and precipitates. Such trapping can occur in the bulk and near surfaces. The majority of these radiative excitonic processes however are only observed at very low cryogenic temperatures, due to the small binding energy of the exciton. Not only that, but their efficiency is also normally low. The efficiency of a radiative process generally depends on the electronic bandstructure of the solid and for crystalline semiconductors the perfection of its lattice and surfaces, since most defects and surfaces promote non-radiative processes. Non-radiative processes 23 in silicon are dominated by Auger recombination 24 and recombination at "deep level" defects such as transition metal impurities in the bulk 25 or defects at surfaces and interfaces such as that created by ambient or thermal oxidation of silicon. 26

Photoluminescence of crystalline bulk silicon
It is instructive to rst survey the PL of bulk Si because this reveals certain "spectroscopic ngerprints" of radiative excitonic recombination processes in this crystalline material. The fundamental indirect energy gap in crystalline bulk silicon, between the top of the valence band and lowest conduction band is 1.170 eV at 0 K, decreasing monotonically to 1.125 eV at 300 K. 27 Interband near infrared photoluminescence from silicon was rst observed by Haynes and coworkers in the 1950s. 3 At room temperature its near bandgap emission consists of a single weak band at $1.09 eV (1134 nm) with a full width at half maximum (FWHM) that is about 100 meV. 3,28,29 At cryogenic temperatures recombination becomes excitonic. The exciton in bulk silicon has a Bohr radius of 4.2 nm and a binding energy of only 14 meV. 30 Fig. 2 shows how the spectra broaden considerably above room temperature and sharpen considerably at cryogenic temperatures, with a spectral position that follows the temperature dependence of the silicon bandgap.
At low temperatures (<40 K) and in pure silicon, free exciton (FE) recombination totally dominates over bound exciton (BE) recombination (Fig. 3). Transition linewidths are now narrow enough to resolve ne structures corresponding to the TA and TO phonons as shown in Fig. 2 and 3. Two phonon replicas are very weak. Note the extremely weak "zero-phonon" peak I 0 near 1.16 eV in Fig. 3 where electron and hole recombine without a momentum conserving phonon. The spacings of the features I TA and I TO from I 0 in the spectrum are determined by the corresponding phonon energies. Currently accepted values for bulk silicon are: TA phonon ¼ 18.4 AE 0.2 meV; LO phonon 56.2 AE 1 meV; TO phonon ¼ 58.0 AE 1 meV. The relative ratios of the phonon-assisted peaks are temperature dependent but the TO mode is dominant with approximate values in Fig. 3 31 Concerning defect-related emission let us rst consider the simplest point defects: substitutional donors, like P, As and Sb; interstitial donors like Li; and substitutional acceptors like B, Al, Ga and In. Free excitons migrating through the lattice can bind to the neutral donor or acceptor and binding energies lie in the range 3-50 meV. 31 When the exciton is localised on the impurity, there is a relaxation of the wave vector selection rule so the zero phonon line is thus much stronger for bound excitons (e.g. In 0 in Fig. 4) than for free excitons (e.g. I 0 in Fig. 3). Rapid progress in identifying many radiative defects began in the late 1970s when Ge photodetectors became available and were orders of magnitude more sensitive than PbS detectors in the near infrared. 31 The huge increase in resolution enabled isotope shis in sharp zero phonon line transitions to be studied, thereby unambiguously identifying which impurities were involved in specic vibronic band emission. When combined with uniaxial stress measurements to elucidate the defect symmetry, and correlations with other spectroscopic techniques, detailed models of radiative point defects emerged. 31 Fig. 3 Intrinsic free exciton PL from high purity silicon at 26 K. 16  Unfortunately, some luminescent bulk silicon crystals also display broad featureless bands at sub-bandgap wavelengths 19,21,32 providing a challenge for identication that researchers of today's silicon nanostructures are more than well aware. Fig. 5 shows an example of a broad and featureless band at 0.94 eV attributed to extended defects (hydrogen platelets) in bulk silicon. 21 1.4. Surface passivation chemistries for bulk silicon Surface or interface "passivation" refers to the removal of electrically or optically active defects and has become increasingly important in silicon solar cell technology due to the cost-driven reduction of cell thickness and corresponding increase in surface to volume ratio. There has been an associated signicant effort in understanding the origin of, and minimizing, such non-radiative recombination. That literature is utilized here to identify some of the most useful passivation chemistries that could be applied to luminescent silicon nanostructures, and indeed many, but not all, have already been used. Given that nanostructures have huge surface areas, the importance of minimizing non-radiative surface recombination becomes of paramount importance if high PLQY is to be realized.
Surfaces disrupt the band structure, creating energy states in the bandgap via strained or un-terminated ("dangling") bonds. These states can capture electrons or holes with capture velocities S n0 and S p0 that are governed by the product of carrier diffusion, capture cross-section and defect density, D it . A surface will have different steady-state carrier concentrations (n s , p s ) of electrons and holes to those in the bulk, and different to the intrinsic carrier concentration (n i ).
The total recombination at a surface can be characterized by an effective "surface recombination velocity (SRV)" that quantitatively assesses surface or interface passivation: where Dn d is the excess minority carrier concentration that dominates overall recombination rates. 33 What eqn (1) nicely illustrates are the two complementary techniques that have been investigated for reducing surface recombination in bulk silicon. One is "chemical" in the sense of new bonding arrangements of surface dielectric lms or organic monolayers and chemical species like hydrogen. These can lower capture rates by having fewer bandgap states (defects) or lowering the capture cross sections of existing defects. They therefore lower the carrier capture velocities S p0 and S n0. The other is "physical" in the sense of utilizing surface charge and thereby electric elds to lower the concentration of minority carriers at the surface or interface. This can be achieved by having a raised charge density at the surface or within the interfacial layer that acts to reduce the concentration of one type of carrier at the surface, where most defects reside. They therefore lower n s or p s or change their ratio.
The lowest values of SRV for bulk silicon have been achieved by utilizing both techniques. 33 Table 1 shows SRV values for a range of bulk Si surface and interface chemistries. Some observations from surveying this eld and the data in Table 1: Hydride passivation is effective but susceptible to native oxide growth which can seriously raise non-radiative rates. 36 Organic monolayers with Si-C bonding are also effective, but passivation also deteriorates with extended air exposure. 36 SRV values below 1 cm s À1 are achievable at bulk Si interfaces with optimized a-Si, thermal oxides and nitride coatings. Alumina, titania and silicon carbide lms can also passivate silicon surfaces but SRV values are generally a little higher.
Interface densities are much lower for atomically smooth interfaces. 46 Surface/interface charging has not been explored thoroughly versus PLQY of nanostructures.
Superacids and Naon can provide very low SRV, at least with bulk Si of moderate surface areas. [43][44][45] They are yet to be explored with silicon nanostructures.
SRV values are generally much higher for nanostructured silicon (see Section 5.1).

Luminescent Si nanostructure diversication
The easily reproduced demonstration of efficient 47 and wavelength-tunable 47,48 visible emission from porous silicon and silicon nanocrystals around 1990 led to a wide range of silicon nanostructures being explored, together with a variety of synthesis techniques. Fig. 6 illustrates many of these schematically. Each class of nanostructure is at least mentioned in this review, but with emphasis that very much reects their respective levels of study and development, and also their success in achieving high photoluminescence efficiency.
A variety of different terms for luminescent silicon nanostructures are used in the literature, so we will start by mentioning them here. The terms silicon "quantum" wells, wires, and dots are the smallest 1D, 2D and 3D conned crystalline silicon structures, respectively, with diameters below that of the Bohr exciton radius ($4 nm) in bulk silicon (see Section 1.3). Such structures are expected to have bandstructures that are size-dependent due to quantum connement effects as discussed in Section 1.6. Silicon "nanoparticles", nanowires, nanopillars or nanorods can oen, but not always, have larger diameters in the range 5-100 nm. These can exhibit carrier connement effects on PL but their bandstructure is similar to that of bulk Si. In the literature both terms are used interchangeably. "Nanoclusters", for example, oen denote the tiniest structures consisting of under 50 atoms. Likewise, "nanosheets" usually refer to freestanding layers of, at most, a few atoms thickness.
The experimental literature is currently dominated by luminescent silicon nanocrystals, porous silicon, silicon nanowires and silicon superlattices, in that order. The theoretical literature is dominated by calculations of silicon quantum wires and silicon quantum dots. Other silicon nanostructures have received some, but not substantial development. Examples of silicon nanostructures where efficient visible luminescence at 300 K is either not reported or not quantied to date include silicon nanoribbons, 49 nanoshells and nanotubes, 50 nanosheets, 51-54 molecular clusters, 55 allotropes 56 and metastable polymorphs. 57

Theory of quantum connement effects in Si nanostructures
Theoretical studies of how the silicon bandstructure evolves with size and shape in the critical 1-10 nm size range began in the early 1990s and there is now a substantial body of work. I am not aware of a prior review that tries to comprehensively analyze the considerable theoretical literature on silicon nanostructures and will not attempt to here. Numerous publications are provided here and a few concepts and trends are selected where there is a consensus with respect to issues that affect PLQY, PLWT and PLDT of silicon nanostructures.
Starting with the earliest calculations in 1992-1993 (ref. [58][59][60] and many others, theoretical work has continuously grown over the last 30 years, due to a myriad of potential applications for nano-Si, many of which do not utilize photoluminescence. Studies are grouped in Table 2, not by calculation technique, but by nanostructure morphology and surface/interface chemistry. The aim is to help readers access the available theory most relevant to their specic luminescent nanostructures. A large number of theoretical studies have focused on silicon quantum wires, but the literature on quantum dots is growing. Luminescent porous silicon has been modelled as idealized quantum wires and quantum dots or by introducing an articial periodicity of holes via supercell techniques. 109 In reality, morphology is variable but oen of mixed 2-3D connement due to undulating wire thickness (see Section 2.1) combined with other geometrical shape variations. 110 Surveying the theoretical data of Table 2 there would appear to be a consensus that: The silicon bandgap starts to widen signicantly for quantum wells (slabs/ nanosheets) with widths under 2 nm, quantum wires with diameters under 3 nm and quantum dots with diameters under 5 nm.
In principle the entire visible range is accessible for quantum dots and wires, with bandgap tunability extending from 1 eV to 3 eV. Exciton binding energy is signicantly enhanced due to quantum connement reaching values as high as a few hundred meV for 1.5-3 nm diameter quantum dots or 1-2 nm diameter quantum wires.
Surface chemistry signicantly affects the bandstructure of smaller nanostructures, e.g. oxidation narrows the bandgap of hydride passivated nanostructures. Table 3 provides some theoretical data for hydride passivated silicon quantum dots, quantifying how the bandgap and strength of exciton binding increase dramatically with decreasing size for 1-5 nm nanocrystals. Note how the predicted radiative recombination rates however, only become comparable with those of direct bandgap semiconductors for the smallest silicon nanocrystals around 1 nm diameter.
Aside from size, shape and surface chemistry, there are other factors that affect bandstructure and therefore PL. There have, for example, been a series of theoretical effects investigating the roles of strain. [116][117][118][119] For quantum wires, crystallographic orientation also has a signicant effect on bandstructure. 75,77,83,120

PL quantum yield (PLQY)
For any photoluminescent material or molecular species the efficiency of its radiative processes or "photoluminescence quantum yield" (PLQY) of its luminescence is an important quantitative property that will impact on its likely use. The PLQY depends on direct competition between the specic radiative process and all non-radiative processes: Table 3 Theoretical estimates of bandgaps, exciton binding energies and radiative recombination rates for 1-5 nm diameter silicon nanocrystals (quantum dots) with hydride passivation PLQY (%) ¼ photons emitted/photons absorbed ¼ k r /(k r + Sk nr ) (2) where k r is the radiative recombination rate and k nr the non-radiative rate.
Measurement of both PLQY h and luminescence decay time s PL enables one to separate the radiative and non-radiative lifetimes of excited states using the simple relation The intrinsic radiative recombination coefficient k r of bulk c-Si has been estimated to be 2 Â 10 À15 cm 3 s À1 which is about 100 000 times lower than direct bandgap semiconductors like GaAs. Absolute quantum efficiency estimates of interband recombination (see Fig. 2) in standard silicon crystals (wafers) with native oxide surfaces (see Table 1) typically range from 10 À6 to 10 À4 . However, eqn (3) emphasizes that if all non-radiative processes can be virtually removed, then PLQY can be quite high. This has been elegantly demonstrated by solar cell specialists (Green and co-workers). In 2003 they showed that in the highest purity oat zone silicon with optimized surface passivation, PLQY of near infrared emission could be as high as 6.1 AE 0.9% at 300 K. 121 Accurate measurement of PLQY is not as straightforward as measuring PL decay times and a number of different protocols exist, but they can be divided into comparative ones that utilize photoluminescence standards of known PLQY and absolute ones that directly measure the fraction of absorbed photons that are emitted as photoluminescence or the fraction of photons lost by non-radiative recombination. The latter use integrating spheres to collect all emitted light, or use photoacoustic/calorimetric methods to quantify heat production. Methodology can depend on the physical form of the photoluminescent material (e.g., lm versus dry powder versus colloidal solution) and the spectral positions of both excitation and emission. A number of important reviews are available that detail the theory, methodologies, standards, and sources of error involved. [122][123][124][125][126][127] Of particular relevance here is the review by Valenta on PLQY measurements of nanomaterials. 127 For comparative methods, the most established standards are quinine sulphate in 0.5 M sulphuric acid, uorescein in 0.1 M NaOH and rhodamine 6G in ethanol with peak emissions at 451, 515 and 552 nm, respectively. Their PLQY accuracy is better than 4% under well-dened conditions. 124 These standards are well suited to the spectral positions of the F-band (see Section 3).
The physical form of silicon nanostructures can have a pronounced effect on PLQY. For example, for thin solid lms light entrapment effects due to total internal reections and Fresnel transmission at the lm-air interface can dramatically lower PLQY, even if the internal quantum efficiency (IQE) is very high. A light extraction efficiency Ex PL for a thin lm on an absorbing substrate has been estimated. 128 The escape probability of emitted light is strongly dependent on refractive index: So, for example, with a solid silicon lm of refractive index 3.5 only 1.4% of light escapes directly from the top surface. For silicon nanocrystals embedded in silica lms with a refractive index of 1.64, only 8.7% escapes internal reections. For porous silicon nanostructures, PLQY and IQE values start to get closer as refractive index is lowered. High PLQY pSi structures have porosities in the range 70-90% with refractive indices as low as 1.1. Now more than 20% of light emitted internally can directly escape, lowering its self-absorption. Forming luminescent nanocomposites will oen normally lower the perceived brightness due to refractive indices then being higher again. Table 4 collates the highest PLQY values reported to date for all types of silicon nanostructure. It is clear that the highest PLQY values have been obtained in colloidal quantum dots, quantum dot-based superlattices and porous silicon and so these materials are given more consideration in this review. Amorphous nanoparticles have much lower PLQY than crystalline ones and are therefore not discussed. Molecular size clusters and size-dependent PL from ultrathin quantum wells have not had PLQY quantied to date. Nanotube studies have not reported visible PL.

Porous Si (2D to 3D connement)
Historically, relatively efficient ($3%) S-band emission at room temperature was rst observed in high porosity silicon 47 which is why we rst consider this nanostructure, despite its complex range of morphologies which make theoretical modelling a challenge. Freshly etched red-emitting high porosity silicon was found to contain "undulating" quantum wires 147 of below 3 nm average width (Fig. 7b).
Efficient S-band emissions are also evident in material where hydride passivation is replaced by native oxides, 137,149 anodic oxides, 150 high temperature oxides 30 and organic monolayers. 151 All these surface chemistries, with the notable exception of native oxides in some cases (see Table 1) produce low levels of nonradiative surface recombination in bulk Si. In oxide-free material tuning from the near infrared to the blue spectral region (1300 to 400 nm) has been achieved, as shown in Fig. 7a. 148,152 In oxidized material, from the near infrared to the yellow (1300 to 590 nm). 148 Typical PL lifetimes (see Table 5) are in the microsecond range but depend on wavelength and surface passivation quality (SRV) (see eqn (3)).
The data in Table 5 shows variability in PLDT at given wavelengths, due to differing silicon skeleton morphology and interconnectivity, levels of polycrystallinity and surface passivation. Nonetheless, the longest PLDT values of Gelloz 2005 probably reect primarily the lowest SRV values obtained to date by their wet oxide passivation technique.

Single quantum wells and nanosheets (1D connement)
Takahashi et al. 157 and Saeta et al. 158 were the rst to report visible emission from single 2D crystalline silicon structures. Both groups used SIMOX wafers, thinning the silicon layer with high precision via thermal oxidation. A red and near infrared emission band (1.65 eV) was observed with an intensity dependent on the well thickness but spectral position varying little. No luminescence was observed unless the quantum well thickness was less than 5-8 nm. PLQY peaked at $2 nm Table 4 Si nanostructure classes with the highest PL efficiencies (PLQY). For data on the F-band there is considerable uncertainty over the origin of the emission and specifically whether it arises from silicon nanostructures and question marks are included in the table to emphasize this (see Section 3) Class of nanostructure thickness but was very low at 10 À4 . 158, 159 Okamoto et al. 160 reported 2 K PL from SIMOX-derived quantum wells of only 0.6 and 1.6 nm. Emission was again at 1.66 eV. Pauc et al. 161 studied PL at 6 K from SIMOX derived wells of 3.9 down to 0.6 nm. A 1.6 eV band appeared for the narrowest wells with an estimated PLQY of only 0.1% at 6 K. The rst really size-tunable emission was achieved by Green and co-workers. 162 They used ELTRAN structures with superior defect densities (<50 cm À2 ), atomically at interfaces and lower SRVs. Fig. 8 shows TEM data and the 650-950 nm PL spectra as a function of well thickness over the range 1.1-2.7 nm. Their data on processed SIMOX wafers also reported an emission whose peak wavelength was insensitive to QW thickness but correlated instead with oxide thickness. Zhu et al. 163 studied (110) SIMOX wafers with $1-4.4 nm thick wells. Both QC and interface effects were discussed in relation to their 750-800 nm emission. Wagner

Superlattices and multilayers (1D to 3D connement)
Multilayer and superlattice structures have primarily been based on silica passivation, 164-171 but photoluminescent lattice matched Si/CaF 2 superlattices [172][173][174] have also been achieved. Lockwood and co-workers 164,165 rst reported a photoluminescent MBE grown Si/SiO 2 superlattice. The 6 period structure had 1.0-5.4 nm thick Si quantum wells with 1 nm thick oxide barriers. Visible emission was only observed for wells below 3 nm. PL peak energies were tunable over the 1.7-2.3 eV range but PLQY values were not given. A major advance was made by the Zacharias group in a series of  papers 171,175,176 using superlattices to exploit 3D rather than 1D connement. They also dramatically raised PLQY via improved interface passivation, size control and nanocrystal density. Fig. 9 shows a typical structure and the tunability of the PL. 171 Valenta et al. 176 provided detailed information on PLQY values, lying in the range 10-20% and doubling when the silica barrier layer thickness was increased from 1 to 3 nm. PLQY variation with temperature was also studied, peaking at 30% at 100 K. Very recently, 136 PLQY values reported for near infrared emission (1.35 eV/0.97 micron) corresponding to 4.5 nm nanocrystals were as high as 50%.

Nanowires and quantum wires (2D connement)
The rst free-standing isolated NWs of widths under 10 nm realized by lithographic etching techniques showed very weak photoluminescence [177][178][179] or no visible luminescence at all, ascribed to insufficient surface passivation. 180 Using laser ablation to create 13 nm wide ultralong nanowires, and then oxidation to thin the silicon core to 3.5 nm, Feng and co-workers reported red emission of unknown PLQY. 181 Complete oxidation of the wires removed the red emission but a distinct green emission band persisted (see Section 3.2). Brongersma and coworkers used catalysed CVD growth to create 20 nm wide nanowires and then oxidation to decrease core widths to below 5 nm. Tunable 800 to 720 nm emission with microsecond decay times was achieved. 182 Examination of Table 4 reveals that nanowire fabrication, excluding porous silicon, has been less successful to date with regards to achieving the highest PLQY values for S-band emission. Korgel's group reported a PLQY of 4.3% for 3-4 nm wide solid quantum wires. 141 For nanowires we need to distinguish those that are solid silicon from those that have larger diameters but are themselves porous. The latter "porous silicon nanowire arrays" will behave in a similar manner to porous silicon lms but with lowered nanocrystal densities and, to date, lower PLQY (see Table 4). Some much larger solid nanowires also have oxidized rough sidewalls containing nanocrystals that generate the visible photoluminescence. 183   184 These were extracted by-products of a plasma torch spherodization of micro-particulate powder. Once extracted via centrifugation, the various fractions showed PL that shied from 950 to 680 nm as the average Si core diameter evolved from $5 to $3 nm.

Nanoparticles and quantum dots (3D connement)
S-band PL from visibly luminescent silicon nanocrystals (nanoparticles) synthesized by a broad range of techniques have now received a vast amount of study. Fabrication techniques include plasma-assisted decomposition of silane or silicon tetrachloride; laser pyrolysis of silane; liquid phase synthesis; laser ablation; high energy milling and microemulsion growth. The reviews of Mangolini, 185 Huan et al. 186 and Cheng et al. 187 provide details on the majority of these synthesis techniques. Highlighted here is the recent study of the Ceroni and Korgel Groups using dodecene passivated nanocrystals in colloidal suspension. 188 Fig. 11 illustrates the excellent control over size. Fig. 12 shows how spectral position smoothly "blueshis" with decreasing nanocrystal size right from the 1.1 eV bandgap of bulk silicon (for 9.1 nm average size) to 2.1 eV (1.8 nm average size). Table 6 shows the decay times from a number of nanocrystal studies and it is informative to compare values therein with Table 5 (data from porous silicon). For example, at 800 nm, pSi decay times (Table 5) lie in the range 10-100 microseconds; isolated nanocrystals emitting at 800 nm (Table 6) have much longer decay times in the range 80-247 microseconds. It is likely that the internal surfaces of porous silicon nanostructures are more challenging to passivate than isolated or embedded nanocrystals. Table 7 summarizes the remarkable improvements made in raising the PLQY of Sband luminescence alone since 1990. Early values in the range 3-5% probably reected the lower levels of size control, combined with higher SRV levels. The highly interconnected silicon skeleton of porous silicon facilitates rapid exciton migration to lower bandgap parts of the network that have slower radiative rates. 137,193 Size control is thus key. This, in combination with differing light extraction efficiencies, explains why colloidal pSi nanoparticles typically have higher PLQY than microparticles and lms (see Table 7).    Fig. 13 presents a published and very useful compilation of PL data from 17 studies on nanocrystals made by a variety of techniques 208 and effective mass approximation (EMA) tting of nanocrystal bandgap versus size. There is clearly considerable scatter of quantitative data, partly because both S-band (this section) and F-band (see Section 3) data were used. For the S-band data, another complicating factor are the size-dependent optical absorption coefficients: smaller silicon nanocrystals have lower optical absorption which biases the PL spectra towards larger nanocrystals. 188 Fig. 14 illustrates how one parameter (nanocrystal size dispersion) can therefore strongly inuence the theoretical EMA modelling of PL tunability versus average size. Good agreement between such theory and experimental data is achieved once this is taken into account.

Wavelength tuning of S-band via surface chemistry
The ability to tune PL emissions by surface chemistry passivation, as well as by nanostructure size, would be desirable, particularly for fabrication techniques which offered high volume throughput, but lacked ne tunability of size. Theoretical calculations have shown that the bandstructure of ultrasmall nanostructures such as 1-1.5 nm diameter quantum dots and quantum wires are strongly affected by surface chemistry. 86,95,96 Spectral position of the S-band might then also be tunable by surface chemistry. However direct experimental evidence for extensive S-band tuning is sparse. One study reported that for larger nanocrystals (1.64 eV PL corresponding to $3.5 nm nanocrystals) changing from hydride to alkyl (dodecene) passivation had little effect on spectral position. 209 Not surprisingly however, PLQY and PLDT were affected, as the SRV will differ. In contrast, a study of amide and alkoxy functionalization relative to alkyl did nd a small but measurable effect on spectral position for 3.5-6.4 nm nanocrystals. 210 In addition, another study using alkynyl(aryl) groups also observed PL shis (in the range 685-800 nm). They attributed the effect to "surface-state induced bandgap changes". 211 In instances where chemical tuning "right across the visible range" is claimed for silicon nanocrystals, 212,213 it would be important to investigate whether the blue emission is in fact F-band emission by measuring PLDT as a function of tuning (see Section 3) and therefore of quite different origin to the yellow or red S-band emission. The extent to which the S-band can be tuned over a broad wavelength range via surface chemistry alone needs further investigation.

Selective photoexcitation of S-band emission
As we saw with many radiative transitions in bulk silicon (Section 1.3), sharp features in PL spectra can allow the determination of precise energies of electronic states, the symmetry of luminescent defects and the phonon energies involved. The large linewidth of the S-band PL obscures the spectroscopic information it contains. Photoexcitation at resonant wavelengths via tunable lasers can selectively excite subsets of the nanocrystal ensemble and lower the effects of inhomogeneous broadening, thereby revealing otherwise hidden spectral features. Starting with the studies of Calcott, Nash and co-workers it has now been investigated extensively with luminescent porous silicon 198,214-222 but also with silicon superlattice nanocrystals, 223,224 large silicon nanowires, 225 quantum wells 226 and nanocrystals made by various other techniques. 198,[227][228][229][230] Initial concerns raised about its applicability to high PLQY porous silicon emission at room temperature and the precise number of phonon features were resolved by 1997. 30 Highlighted here in Fig. 15 are S-band resonant photoexcitation data from a range of different silicon nanostructures. All show the TO and TA phonon-related features so characteristic of exciton recombination in crystalline silicon. For the S-band PL from a given nanostructure, at increasing levels of connement (shorter wavelength emission), the exciton binding energy increases (Table 3) and the no-phonon (NP) to phonon (TO, TA) intensity ratio increases, just as for localized bound excitons in bulk silicon ( Fig. 3 and 4).
Even more revealing than resonant photoexcitation can be the photoexcitation of isolated individual nanostructures. Single nanostructure spectroscopy can reveal homogeneous linewidths and most previously hidden spectral features, but is experimentally challenging. The rst such spectroscopy studies were performed on porous silicon particles in 1998 by Buratto and co-workers. 231 Progress has been recently thoroughly reviewed by the Linnros group 232 so only a few comments will be made here that relate to PLQY of nanostructure ensembles and the origin of S-band PL: Individual nanocrystals "blink" (periods of no emission under constant photoexcitation) and some are completely "dark"; others are "bright" with near unity IQE. The ensemble PLQY can thus be raised signicantly by lowering the density of dark silicon nanocrystals that have a very efficient non-radiative channel(s), by decreasing blinking rates, or by both routes.
PL linewidths for single nanocrystals are substantially narrower than for the ensemble (typically 100-160 meV at 300 K but values as low as 17 meV are now being reported 233 ). Both temperature and the medium surrounding the core have a strong inuence.
Substantial TO phonon involvement was again invoked to explain the emission spectral data.

Origin of S-band PL
The wavelength tuning via size (Section 2.7), right from the bulk silicon bandgap 201,205 to at least the yellow spectral region, convinced many that quantum connement was indeed responsible for S-band emission. Resonant photoexcitation experiments (Fig. 15), providing a spectroscopic signature of established silicon phonon-assisted transitions (Section 1.3), convinced many more. Single nanocrystal spectroscopy (Section 2.9) holds the most potential for unravelling the complexity of ensemble nanosystems where size and shape control is still improving, but not yet sufficient to avoid signicant inhomogeneous broadening and rapid carrier migration effects. 234 Nonetheless, aer many hundreds of studies, there is now substantial evidence, both experimental and theoretical, supporting a quantum connement model for S-band emission, at least for near infrared to green emission. Agreement between theory and experiment exists for PL position (excitonic bandgap) versus size and shape (Fig. 14), together with strongly wavelength-dependent measured decay times (Tables 5 and 6) and extracted radiative decay times. The reasons for rapid decline in PLQY for yellow to blue S-band emission; the ultimate PL linewidths achievable; the avoidance of "blinking"; improvements in SRV for smaller nanostructuresthese are all topics requiring further study.

Porous Si
Reports of blue-green emission ("fast"-nanosecond decay time and hence "Fband emission") from porous silicon started to appear 235,236 in 1992. Kovalev and co-workers 237 showed that its intensity could be signicantly raised by rapid thermal oxidation and since then it has received a series of studies. It is noteworthy that even moderate PLQY requires oxidation 238 and/or heat treatment of the porous silicon. Freshly oxidized porous silicon oen does not display F-band emission and it only emerges aer storage in ambient air. 239

Nanowires
Growth of nanowires via catalyzed laser ablation VLS growth or thermal evaporation oen results in short wavelength emission alone. [240][241][242][243][244] A silicon oxide phase is invariably present and oen silicon core width greatly exceeds that required for silicon bandgap enlargement (<3 nm). Sometimes there is no silicon phase at all. 240 PLQY values have not generally been provided.

Nanoparticles and quantum dots
For silicon quantum dots prepared by high temperature synthesis, such as silane decomposition, F-band emission also appears aer air storage (see Fig. 1).
A very large number of reports of F-band emission of high PLQY (only some are in Table 8) have appeared since the 2013 report of "one-pot aqueous synthesis" 245 by He and co-workers. Here, efficient blue-green emission is achieved immediately aer reduction synthesis using inexpensive precursors, low temperatures (e.g. <200 C) and short reaction times. Highlighted here though instead, is the recent reappraisal study of Oliinyk and co-workers. 246 They cast doubt on the likelihood of such conditions generating crystalline silicon nanostructures. Importantly, they also report that repeating the synthesis with a control (reducing agent and a silicon-free organic amine) produced remarkably similar photoluminescence. The de Cola group have also repeated such a synthesis under both short (15 minute) microwave irradiation and lengthy (300 minute) hydrothermal conditions. 247 Aer thorough characterization of repeat and control formulations (lacking reducing agent or silicon source) they attribute the blue-emitting product to the formation of silica and carbon dots rather than silicon nanocrystals. A further reappraisal study 248 has now also been conducted on a different wet chemical synthesis route that uses a magnesium silicide precursor developed by the Kauzlarich group. Once again the key control run, where the source of silicon was omitted, led to similar F-band emission to where the silicon precursor was included. These three studies [246][247][248] highlight the current uncertainty that surrounds the true origin(s) of F-band emission (see Section 3.5). Table 7 illustrates the dramatic rise in PLQY for the F-band over the last 5 years. In the rst reports of emission from oxidized porous silicon lms and microparticles, quantum yield was probably at best 0.1%. 238 Colloidal NP emission prior to 2010 was much better, but in the range 1-60%. Since 2015 however there have been a number of reports of PLQY above 75%, even as high as 90%. The table also emphasizes the dominance of bottom-up solution synthesis techniques in achieving the highest PLQY values. The question is, as discussed in the previous section, to what extent does such high PLQY have anything to do with silicon nanocrystals?

Selective photoexcitation and "ne structure" of F-band
As with the S-band, selective photoexcitation studies of F-band PL might provide some insight by revealing hidden spectral features. Currently, very few studies have been conducted 262,263 but these did reveal potentially useful ne structures that need further study. However, even under non-resonant excitation, some interesting structures of the emission band have been reported in some instances, 264 especially for NPs made by laser ablation in solvents. 265-267 By conducting control experiments where the solvent (e.g. toluene or octene) alone was UV irradiated, there appeared to be a good match of F-band spectra with those where silicon nanoparticles were created. Fluorescent solvent by-products need to be considered in such laser ablation studies.
The study of Kobayashi and co-workers 268 is also particularly insightful with regards to the potential inuence of nanostructure-adsorbed impurities at low concentrations. They utilized high energy milling to generate 5 nm NPs that were then suspended in hexane and generated a blue-green PL band with numerous sub-peaks. Control spectra of hexane alone, without NPs, yielded very much weaker spectra in the same location (Fig. 16). They attributed the PL to adsorbed dimethylanthracene (DMA), a highly uorescent impurity in hexane. DMA has been used as a standard to estimate comparative PLQY for blue-emitting silicon nanoparticles. 258 Importantly, intentional addition of only a 20 nM solution of DMA in hexane to another DMA-free solvent containing silicon NPs was sufficient for the vibronic structure to appear in spectra. Exciting light absorption via energy transfer from the silicon NP was estimated to raise PL output of DMA in hexane by a factor of about 3000. This study highlights the need for extensive characterization of the purity of potentially luminescent silicon nanostructures.

Origin(s) of F-band PL
Studies focused on elucidating the origin of the F-band emission have broadly drawn four proposals to date: that the emission is connected to the core of small silicon NPs; 269-273 radiative silicon-oxide interface states; 1,274,275 defect states in pure oxide; 1,276,277 or contamination of that oxide. 239,278,279 Valenta et al. 270 , for example, studied F-band emissions from small nanoparticles obtained by electrochemical etching, hydrogen peroxide treatment and ltration. They interpreted the very large spectral shis of the band under varying photoexcitation wavelengths (270-420 nm) to the behaviour of Si QD core emission, rather than defective oxide states. Data was tted to theoretical predictions of bandgap for small hydride-terminated nanoparticles. No FTIR data or chemical composition data was presented however, so the level of hydride remaining vs. oxide passivation (known to generate different bandgaps) is unknown.
F-band emission assignment to the oxide phase of oxidized silicon nanostructures is based on a number of observations, not least that it bears similarities to emissions from pure defective silica structures 280 and can still be observed when the nanocrystalline silicon phase has been removed. 240 It is interesting that high surface area silica, for example, shows similar "aging" and "regeneration" effects for its blue photoluminescence to oxidized porous silicon and nanocrystals from high temperature synthesis, attributed to interactions with water vapour and organic molecules in ambient air. [281][282][283] Are there specic contaminants that might be responsible? A very recent study 284 used pulverized pSi particles to investigate the effects of nitrogen contamination on F-band emission, following the work of Dasog et al. 285 reporting signicant inuence of nitrogen incorporation in nanocrystals made by other routes. The porous Si structures were heavily oxidized (FTIR spectra) and intentionally subjected to environments rich in nitrate and nitrates. They found increased nitrogen levels diminished, rather than induced, blue emission.
The study of Chen et al. 286 in 2013 instead provides strong support for a link to carbon contamination as one potential origin of the F-band in silicon nanostructures containing organic molecules that are subjected to oxidation, heat treatments, microwave treatments or UV laser treatments. They calcined mesoporous silica NPs containing organic molecules at 400 C and observed a 450-475 nm emission band. The uorescent particles had retained an ordered silica framework with 3 nm pores. When these $80 nm mesoporous silica spheres were subsequently soaked in 37% HF to dissolve the silica phase, the blue output was however completely preserved. Under subsequent TEM examination, the residue (see Fig. 17) was found to contain a high density of carbon nanoparticles ($3 nm diameter). The presence of carbon dots does not show that they are necessarily the origin of the luminescence. 287,288 It could be from molecular-sized organic chromophore by-products.
What this does demonstrate however, is that porous silica and oxidized porous silicon nanostructures can act as unintentional "nanoreactors" for carbon dot synthesis. Careful experimentation is required to check that F-band-like emission is not a result of carbon contamination of an initially hydrophobic and organophilic nanostructure. As pointed out many years ago, there is direct evidence of carbon contamination of hydride-terminated porous silicon from a variety of sources: extensive water rinsing used to remove HF residues; 289 atmospheric and vacuum storage; 149 plastic wafer and sample container outgassing. 279 Low levels of carbon contamination are likely to generate F-band emission that is detectable, but of low PLQY. Intentional carbon impregnation of mesoporous silica (e.g. by glycerol) is now used by the carbon dot scientic community to generate high PLQY 290 in the blue-green spectral range. It would be informative to search for carbon dots as well as silicon quantum dots in the now varied protocols for realizing ultrahigh F-band PLQY (see Table 7). It is also noteworthy that the carbon dot research community has similar issues with regards to unintentional uorescent impurities. 287,288 Without sufficient purication steps, molecular carbon-based chromophores can contribute the majority of the emission from bottom-up carbon dot formulations. Intentional introduction of both silicon and

Sensing
Very soon aer the demonstration of efficient visible photoluminescence from porous silicon 47 it became apparent that PL output could be sensitive to a range of adsorbates. 291 This has been the basis of many studies exploiting the photochemical properties of nanoscale silicon to attain sensors for gas and liquid analytes. Prior reviews include those of Sailor et al., 292 Gonzalez et al. 293 and Ji et al. 294 Chemical toxicants, pollutants and biomedical analytes have been the primary targets. Both PL quenching 295-299 and enhancements ("photobrightening") have been reported. 300,301 The vast majority of sensors have utilized PL quenching, which in many cases is reversible, but irreversible quenching has also been considered for single-use sensors. Fig. 18 illustrates S-band sensing of various explosives with silicon nanocrystals embedded within a convenient paper substrate.
Quenching can be static (binding), dynamic (collisional), oxidation-induced or deposition-induced. 302,303 Table 9 and 10 collate the majority of sensing studies performed together with Limit of Detection (LOD) values and separate them into those that use the S-band (red-yellow PL) from the many recent studies exploiting the F-band (blue-green PL).
General advantages of uorimetric (PL) sensing include low background signals and thereby high sensitivity, fast data acquisition and low cost. General challenges include sensor stability, specicity, repeatability, linearity, response and recovery times, and calibration. For Si-based sensors (Tables 9 and 10), approaches to improve specicity have included the immobilization of enzymes; 319 ratiometric PL spectroscopy; 302 combining PL quenching at a single wavelength with peak shis; 303 or decay time changes. 307 With porous Si, multilayer microcavities offer multiparametric sensing. 317 Recently, pSi-based biosensors have focussed on reectance-based interferometry rather than photoluminescence-based analysis 324 but as PLQY and more importantly, its stability, continues to improve, there could be a resurgence of interest in exploiting the S-band.

In vitro and in vivo imaging
Cellular and biomolecular imaging in vitro enables contactless studies of numerous physiological processes. Medical imaging facilitates noninvasive imaging of disease progression in vivo. These are critical tools in clinical diagnostics, physiology and cell biology. A signicant volume of literature now exists on using photoluminescent silicon nanostructures for both in vitro and in vivo imaging. It has been the focus of a few prior reviews 344-348 including a very recent one, 349 so only a short overview will be given here. The S-band emission has been used extensively in vitro as a "uorescent label" in cell cultures. [349][350][351][352][353][354][355] It is also increasingly being evaluated for in vivo use. [357][358][359][360][361][362][363][364][365][366] The facile synthesis for achieving very high PLQY of F-band emission has led to a recent surge in its use as a label in vitro and there is now also a signicant body of work on this. [367][368][369][370][371][372][373][374][375][376][377][378][379][380] In contrast, in vivo usage of F-band emission is very rare. 381 S-band emission can be used in vitro and in vivo, whereas F-band emission is generally restricted to in vitro use, because of high tissue absorption of both UV excitation wavelengths and blue-green emission wavelengths. Tissue has two "transmission windows" in the near infrared -700-900 nm ("NIR 1") and 1000-1700 nm ("NIR II"), where scattering and absorption of light is lower than adjacent wavelengths. 382 The S-band emission of Si nanostructures normally falls within the NIR1 window. Interestingly, with degenerate doping of nanocrystals to shrink their bandgap, room temperature PL output in NIR II has also been achieved, but the PLQY was relatively low at about 2%. 383 An ideal luminescent in vivo probe would have high PLQY with both excitation and emission occurring within these windows. Unfortunately, efficient PLQY of S-band emission necessitates short wavelength 300-450 nm excitation which limits the depth that can be imaged in vivo. For this reason there have been recent studies exploring two-photon excitation of the S-band 384,385 where both excitation and emission target the NIR I window. Sailor and co-workers have recently shown the feasibility of this with 850 nm excitation under two photon microscopy (TPM) and 560-740 nm emission. Although the efficiency of this process with 60 nm pSiNPs was signicantly lower than an established TPM molecular probe like rhodamine 6G, the porous silicon probe showed much better photostability, the ability to target tumours and carry drug payloads. 366 Gooding and co-workers have also utilized two-photon excitation, in combination with in vitro uorescence lifetime imaging, to excite F-band emission. 386 A wide range of luminescent nanostructured materials are being assessed for in vivo medical imaging 387 so I will summarize the perceived merits and challenges of using either the S-band or the F-band PL for biomedical imaging. There are 3 primary drivers for using S-band luminescence in Si nanostructures for in vivo imaging: (a) Wavelength tunable and now very efficient luminescence output (see Table  7) within tissue window(s); (b) Time-gated imaging that exploits the microsecond decay times to improve contrast; 357,388,389 (c) Biodegradability 356,390 combined with good in vivo biocompatibility in different body sites. [391][392][393] Challenges for in vivo use of the S-band include low optical absorption coef-cients, the relatively broad FWHM of PL, colloidal stability of the PL in body uids, and one-photon excitation wavelengths lying outside the tissue windows.
When using the S-band for in vitro cell imaging the primary advantages are tunable emission, the low cytotoxicity of silicon nanoparticles 394-400 and the exible surface chemistry "toolbox" that has been realized (see Section 5.3).
For in vitro imaging exploiting the F-band emission, the primary advantages are very high PLQY (see Table 8), PL stability (less photobleaching and stable over a large pH range 401 ) and narrower FWHM of PL than the S-band. Challenges for the F-band include the short excitation and emission wavelengths which overlap with tissue scattering and biological autouorescence, and current uncertainties with regards to the origin of PL.
It should also be noted that whereas uorescence labelling is a core technique used in in vitro cell biology, using luminescent agents in vivo is currently very much a developing research tool, rather than a mainstream clinical imaging modality. The latter point has relevance with regards to the likelihood of fast clinical translation of Si NP-based luminescence imaging and hence short-term impact. In this regard, the development of biomodal imaging probes, where photoluminescence imaging is combined with a more clinically established technique, has more potential for clinical translation. Unfortunately, silicon nanoparticles have low X-ray contrast compared to metals. Therefore more complex "nanoconstructs" are needed. There could be, for example, an opportunity for developing porous silicon with entrapped gas for a photoluminescent and biodegradable ultrasonic contrast agent. Table 11 provides examples from the literatureattention to date has been on accessing MRI rather than ultrasonic imaging.

Theranostics
Theranostic nanomedicine, where nanostructures simultaneously facilitate disease therapy and diagnosis, is currently dominated by drug delivery based therapies. 407 In particular, the pre-clinical evaluation of pharmacokinetics and the target site and healthy organ accumulation of drugs. 408 The potential of theranostics with nano-Si was initiated by the pioneering studies of Sailor and coworkers 357 who demonstrated in vivo imaging of S-band and organ targeting. The eld has been very recently reviewed 349 so only a short perspective is given here.
Whilst for many therapies the luminescence provides only a means to track the nanostructures, the photosensitization of oxygen scales with PLQY. 409 In contrast to photodynamic therapy, for photothermal therapy one wants to maximize nonradiative processes, so PL is unwanted. Thus luminescent Si nanostructures can themselves provide therapy (photosensitization 409,410 ) or carry payloads that perform therapy (drugs, radioisotopes, metals 349 ). Luminescent pSi is ideally suited to theranostics 411,412 because of the much higher drug payloads achievable via lling the pore volume, compared to solid nanoparticles where the drug is attached only to the outer surfaces. At 80% porosity for example, drug payloads in pSi can exceed 50 wt% by melt loading techniques and the drug is itself nanostructured by entrapment within 5-50 nm mesopores. Control over pharmacokinetics is achieved via a combination of restricted diffusion and carrier biodegradability. Fig. 19 illustrates the multifunctionality of porous silicon nanoparticles (pSiNP). 412

Photovoltaics
A major limiting factor in the ultimate efficiency of silicon solar cells is the spectral mismatch between the energy distribution of the incident solar spectrum and the bandgap. An efficient photoluminescent material can act as a spectral converter to raise the efficiency of a solar cell by down-shiing (DS), quantum cutting (QC) or "upconversion" (UC) as shown schematically in Fig. 20. 413 Visibly photoluminescent silicon can provide "down-shiers" 414-420 or quantum cutters [421][422][423][424][425] for traditional solar cells.
A second, recently emerged potential use is in "luminescent solar concentrators" (LSCs) for building integrated photovoltaics. [426][427][428] Here, the large Stokes shi between strong UV absorption and S-band emission is of great benet in minimizing light re-absorption over long distances. Solar concentrators are large plastic plates lled with photoluminescent material (commonly organic dyes) that redirect and concentrate solar energy to small photolvoltaic cells at their perimeter (see Fig. 21). A large fraction of the light can be effectively waveguided in the plastic sheet due to total internal reections. The tunable bandgaps of silicon nanostructures offer design freedom with regards to aesthetics and optical performance. Challenges for silicon solar concentrator use include low absorptivity, scattering due to nanocrystal agglomeration, scalability of production and above all, cost.

Colour converters for white LEDs
White light emitting LEDs of 5 lm W À1 brightness rst became commercially available in 1996, and by 2010, brightness had already improved to the 150-250 lm W À1 range. The performance and cost of white LEDs are now competitive for general lighting, automotive, communication and medical uses. 429 There are 3 methods for generating white light via an LED: (a) combining blue, green and red  LEDs; (b) combing a UV LED with blue, green and red phosphors; and (c) using a blue LED to excite green or yellow and red phosphors. Fig. 22 illustrates methods (b) and (c). The latter approach (c) has received increasing use following the development of ultra-bright blue GaN LEDs in the 1990s. For general lighting the total power emitted, power efficiency and "whiteness" (colour temperature and rendering) are critical technical parameters. A 60 W incandescent tungsten light bulb, for example, currently has a luminous ux of 1000 lumins (lm) and a luminous efficiency of 17 lm W À1 . The colour rendering index (CRI) is a quantitative measure of a light source's ability to reproduce the colours of objects against natural or reference colours. Perfect performance is 100. Fluorescent lights typically have values in the range 50-98. LEDs are typically around 80. Colour correlated temperature (CCT) quanties the temperature of an ideal blackbody radiating light of a colour comparable to the light source. Daylight ("cool") white has a CCT values above 5000 K, neutral white 3000-5000 K, whilst incandescent ("warm") white corresponds to a CCT value in the range 2500-3000 K.
In 2014 the start-up company Lumisands Inc, USA, reported the rst use of visibly photoluminescent Si to generate "warm" white LEDs by method (b), and combined the use with other green and blue phosphors. 430 The Si NPs were derived from porous silicon. Surprisingly high PLQY values for the S-band were estimated under blue (405 nm) excitation, especially since the authors reported their particles had "micron-size silicon cores with nanosize Si QD attached". One would expect signicant scattering and re-absorption of red emission with this type of structure. Colour rendering close to incandescent bulbs with a CCT $2800 K and CRI $95 were reported, without a value for luminous efficiency.
Broad band white PL has been reported from a number of silicon-containing nanostructures: heavily oxidized porous silicon, 431 oxidized and carbon containing porous silicon, 432,433 carbon doped silicon-rich silicon oxide 434 and silicon and carbon co-implanted silica lms. 435 Ghosh et al. 436 have also used method (b) but with white-light emitting Si nanocrystals containing material derived from rice husks (see Section 5.4) whose broad PL (230 nm FWHM) was possibly a combination of the S-band and F-band. In this regard, of particular relevance is the study where laser ablation of organosilica also produced white-emitting nanoparticles. 437 Both silicon and carbon nanoparticles were discovered by TEM and are jointly attributed with white light generation. The carbon dots were revealed aer HF removal of the silica phase, as used by Chen et al. 286 and discussed in Section 3.6. A study with commercial blue LEDs and porous silicon NPs embedded in an overlying polymer lm demonstrated the exibility of colour tuning by varying Si NP concentrations. 438 Luminous efficiency was again not reported. For maximizing brightness, thin down-converting lms should have low refractive indices to maximize light extraction efficiencies. 128 The F-band has also been investigated for this application. [439][440][441] Solution synthesis of nanoparticles was adapted to achieve broad yellow-green emissions centred at 540 nm under 410 nm excitation from a blue LED. 439 The 43.5% EQE of the blue LED was reported to be lowered to 8% for the resulting white LED. A luminance efficiency value of 13 lm W À1 was provided together with CCT of 7982 K and a CRI of 82. In a second study by He's group, luminescent nanorods achieved a similar luminance efficiency. 440 In a third study, green emitting NPs with PLQY of 60%, derived from rice husks, were combined with blue and red emitting materials. Under 3 W UV LED excitation the composite lm generated white light with a luminance efficiency of 75 lm W À1 . Photostability of the green-emitting material was monitored over 1 month (720 hours) and PLQY dropped by 12%.
However, high reliability and a very long operating lifetime (50 000-100 000 hours) is one of the key features of LEDs that have led to their remarkable uptake in recent years. Successful inorganic color-converting phosphors like YAG:Ce have ultrahigh chemical stability that can withstand continuous $100 mW mm À2 radiance from a blue LED. Long-term photostability testing under intense blue/ UV excitation is needed for Si nanostructures, as this might be a daunting challenge for this application area.

Nanothermometry
Measuring temperature with nanometer scale spatial resolution has potential widespread use in electronics, photonics and biomedicine but also a range of other niche areas. 442 There are diverse methods but using luminescent probes of temperature are one of the most attractive, as they are non-invasive, contactless and easy to use. 443 Most PL parameters are dependent to some extent on temperature: intensity (efficiency), spectral position, spectral shape, bandwidth, polarization and decay time. Which PL parameter(s) and which applications have the most potential for Si nanostructures? This topic is only starting to receive attention so it is difficult to gauge. Suggested specic applications to date range from tribology 446 to cryosurgery. 444 Researchers at INSA, Lyon, were the rst to realize the potential of S-band PL in this regard and they chose tribological studies of mechanical contacts as the target application area. 445-450 Functionalized 4-8 nm porous Si NPs with S-band emission were studied in alkanes and alkenes from 293-393 K. As expected, with increasing temperature, red shis of PL peak position were universally observed but data analysis revealed that shi magnitude also depended on solvent viscosity. They attributed this to Brownian motion causing energy transfer between nanoparticles of different size. 450 A consequence is that temperature sensitivity will have a complex dependence on liquid viscosity, NP concentration, NP size distribution and NP passivation. A signicant advance was therefore shown by Ferreira and Kortshagen's groups 451 who demonstrated that with better size control (2.0 AE 0.2 nm) of surface oxidized NPs, exciton migration effects were greatly reduced and the emission thermal dependence mainly follows the shrinkage of the bandgap. A single calibration then suffices for different media.
Use of the blue-green F-band in thermometry has also been reported. 448,452 The low temperature solution synthesis route 452 yielded very stable 480 nm emission with also very high 75% PLQY and monoexponential decay, but the PL origin is not established, and hence neither is the biodegradability of the luminescent entity (see Section 3.6). They chose to utilize decay times which linearly decreased from 23 ns at 0.5 C to 8 ns at 60 C, yielding a sensitivity of 0.24 ns C À1 with a thermal resolution under 1 C. Interestingly, when Ryabchikov and coworkers 448 compared the thermal sensitivity of Si NCs with carbon NPs (produced by anodization of Si and SiC wafers, respectively) they found spectral shis to be 3-4 times higher for the Si nanomaterial and of opposite sign to those of carbon dots (À0.18 vs. +0.05 meV K À1 , respectively). Future comparative studies of the temperature responses of high PLQY F-band emissions from Si-based formulations and those of established "carbon dots" may thus provide further insight into the origin of the former.
The quantitative thermal sensitivity values reported to date for both S and Fbands are shown in Table 12. It is clear that, at least for the S-band for biomedical uses, the standard gures of merit are currently signicantly lower than the highest values (1-5%/K) of other material systems. 442 Nonetheless, with further improvements of surface passivation, size control and PLQY, the thermal sensitivity should improve, and this combined with having a thermometer system of tunable biodegradability, makes this an interesting area for further development.

PL efficiency and passivation (stability)
Progress in improving PLQY of both the S-band (Table 7) and F-band (Table 8) over the last 30 years has been spectacular. Photostability of the S-band from derivatized silicon quantum dots is now competitive against many organic dyes with regards to biotagging. For applications like LED's and photovoltaics where very long term stability is essential, further improvements in passivation seem  Table 13.
As for bulk silicon, hydride and organic monolayer passivation is metastable with SRV values rising with air storage. SRV values for nanowires can be remarkably high, reecting perhaps challenges in avoiding defects along elongated nanostructures with highly curved interfaces.

Size control and wavelength tunability
If properties are size-dependent then size control and also metrology of nano-size distribution becomes critical. 457,458 Control over size can be strived for during synthesis, or aer synthesis, or a combination of the two. Some of the narrowest size distributions for as-synthesized silicon nanocrystals have been demonstrated with superlattice manipulation, 459 gas phase laser ablation, 460 plasma synthesis 461 and micro-emulsion techniques. 462,463 Some data are shown in Table 14. Various methods have also been applied to narrow size distributions of already fabricated luminescent silicon nanostructures (see Table 15). These include gas phase selection techniques, [460][461][462]465,466 size selective precipitation, 463 size exclusion chromatography 199 and density gradient ultracentrifugation. 196,468,469 The S-band has been tuned over a wide spectral range but currently very high PLQY (>50%) is restricted to the 700-1000 nm red-near infrared region (see Table  7). The F-band has been tuned over a complementary spectral region with very high PLQY (>50%) in the 440-600 nm blue to yellow region (see Table 8).

Surface chemistry toolbox
The well-developed surface chemistry of bulk silicon 470 was the platform from which many functionalization routes have been successfully applied to luminescent silicon nanostructures. 471 The availability of metastable hydride surfaces has been both a help and a hindrance in this regard: it has facilitated the graing of a wide range of organic capping ligands but 100% substitution seems virtually impossible due to steric hindrance issues. The resulting derivatized surfaces then  have a mixed chemical nature with the desired Si-C bonding plus residual Si-H bonding. Oxidation is suppressed but not eliminated. Complex surface chemistries are more exploitable in ex vivo uses like sensing than in vivo imaging where clinical translation is needed for success. For applications where long term PL stability is of paramount importance, attention should focus on inorganic coatings that have a proven track record with regards to surface passivation of silicon (cf. Table 1). The challenge is to nd scalable techniques that realize full surface coverage of nanostructures with low SRV values (cf. Table 13).

Scalability and cost of luminescent silicon
Nanostructures are typically more much expensive than bulk materials and more difficult to manufacture at scale. Table 16 summarizes some reported throughputs for silicon nanocrystals. It is evident that the majority of these for luminescent nanocrystals lie in the range 0.1-10 g h À1 (i.e. less than 0.01 kg h À1 ). Whilst pilot plant production rates of about 1 kg h À1 have been reported, the nanocrystals were not small enough to be photoluminescent. Exciting progress has been made with silicon nanostructure synthesis by bottom-up high temperature techniques with molecular precursors and conversion synthesis with biogenic silica structures. Continuous belt-furnace production using "silicon accumulator plant" nanostructures 478 is one very scalable technique for producing kilogrammes of biodegradable porous silicon. However, no efficient S-band emission has been reported to date from biogenic porous silicon. Where luminescent nanostructures have been achieved from plant material it has primarily been restricted to F-band emission 441,479,480 and thus might be from carbonaceous material rather than any silicon nanoparticles also present. A signicant goal is thus to achieve tunable S-band output of high PLQY from, for example, bamboo-derived silicon nanostructures. This would open a suite of new high volume and low cost uses (see next section). Dasog's group  477 have recently demonstrated that high S-band PLQY ($50%) can be achieved via magnesiothermic reduction techniques. 481 Synthetic sol-gel silica precursors were used. If this can be replicated with inexpensive biogenic silica feedstocks then the cost of visibly luminescent silicon should be signicantly lowered in the future.

Novel applications
If the cost and scalability of luminescent nanosilicon can be lowered substantially then a range of further applications could emerge as commercially viable. In most cases, ultrahigh PLQY and much lower cost would need to be supplemented with another key attribute for that application. I mention a few examples here from the diverse elds of dentistry, cosmetics, environmental remediation, forensics and security: Sunlight-driven photocatalyst (e.g. adsorbed toxicant degradation). Anti-ageing sunscreens (blue emitters and UV blockers).
Cosmetic skin hue converter (green-emitters used to mask red blemishes). Luminescent inks for banknotes (security features). Contrast agent in forensics (e.g. luminescent dusting powder).

Concluding comments
Remarkable progress has been made in raising the efficiency of visible emission from silicon-based quantum dots, superlattices and porous silicon over the last 30 years, with efficiencies of S-band red emission around 60-70% and green to blue F-band emission around 80-90% being the highest reported values. Surface chemistry plays a pivotal role in determining the efficiency and stability of emission. The widely tunable S-band arises from phonon-assisted exciton recombination within the silicon nanostructure. Emission in the F-band spectral range can have more than one origin, but carbon-based nanostructures and molecular chromophores are looking like the most likely origin of very efficient output. Some low temperature synthesis routes need to be revisited in this regard with more attention given to luminescent nanoparticle characterization and purication. Given the abundance of silica, low toxicity and biodegradability of luminescent silicon nanostructures compared to many other luminescent material nanosystems, their further development for a myriad of applications looks inevitable.

Conflicts of interest
There are no conicts of interest to declare.