In Figure 1, we show the photoluminescence spectra of the large and small Si NPs in powder form before incorporation into the aerogels (the spectra are normalised to the PL peak height to highlight changes only in the shape of the bands). The PL spectrum of the large porous grains is dominated by silicon nanoparticles within the porous shell of each grain, so that the PL spectra of both PSi NPs and Si NSs are similar, though the PL band of the Si NSs is centred at a slightly higher energy and has a slightly lower width, suggesting a narrower size distribution for the Si NSs. When incorporated into aerogel, we see that the PL spectra of the composites are essentially those of the Si NPs still until the lowest concentrations are reached (going from top to bottom of Figure 1, which shows data for aerogels containing LH-type particles). For the lowest concentrations, the Si NP PL becomes weak in comparison to the broader, higher-energy emission from the silica aerogel itself, which is shown in the bottom PL spectrum and is typical of luminescent silica aerogels 21 . Essentially the same sequence of PL spectra is obtained also for the Si NPs of type LO in aerogel. This demonstrates that, in a first approximation, we can consider the PL of the composites as a superposition of the spectra of the Si NPs and the silica matrix; there is no evidence from this data for any interaction between the two.

At the top of Figure 1, we show also the PL spectrum of a sample of the LH particles in powder form with and without oxygen present (the arrow shows how the spectrum drops on the introduction of oxygen); this demonstrates the typical degree of quenching of the luminescence due to the energy transfer process. Detailed discussions of the evidence for this quenching mechanism have been given elsewhere 12 . By contrast, when oxygen is admitted into the aerogel samples containing Si NPs, there is no detectable change in the PL. Thus, incorporation in aerogel brings about a change in the surface state of these particles that reduces the efficiency of the energy transfer process. This is likely to be oxidation of the hydrogen-terminated surface, leading to an increased spatial separation between the confined exciton of the nanoparticle and the oxygen orbitals and, therefore, a reduction in their coupling; further work is, however, needed to quantify this. It is not yet known at which stage of the aerogel preparation this surface oxidation takes place.

Although the colour change of the composites already indicates the state of the Si NPs after incorporation, Raman scattering is useful as a quantitative tool and because, as is usually assumed, quantum confinement and relaxation of momentum selection rules produce a shift of the frequency of the silicon Γ-point phonon mode from its bulk value of 521 cm⁻¹. This shift allows one to estimate where the peak of the size distribution of the NPs lies 17 22 23 24 25 . We note that the influence of the nanocrystal surface provides an alternative physical explanation of the Raman peak shift 25 , though in that model, the quantitative relationship between NP diameter and Raman shift is not altered.

In Figure 2, we show the Raman spectra of a representative selection of composite samples. At the top, we show the Raman spectrum of a pure silica aerogel; the bands at 485 and 620 cm⁻¹ are, respectively, the D₁ and D₂ defect bands of the silica matrix (their microscopic origin has been widely debated: see 1 26 and references therein). The first-order Raman scattering of the Γ-point mode of bulk silicon gives a line at 521 cm⁻¹, and as Figure 2 shows, the LH aerogel composites show a line close to this position (with a Lorentzian lineshape, centre at 522 cm⁻¹, FWHM 12 to 14 cm⁻¹). This implies that the Raman scattering is now dominated by the remaining solid silicon core of the PSi particles and, therefore, that there is very little of the porous shell remaining (however, the PL spectra of Figure 1 do demonstrate that there are some NPs still present in the LH aerogel).

On the other hand, the aerogels containing LO particles show a Raman band that is clearly asymmetric and is shifted to lower frequency (peak at 515.5 cm⁻¹, FWHM 31 cm⁻¹). On the basis of these two facts and following earlier work, e.g. 22 , we could attribute this to Raman scattering from Si NPs of diameter around 4 nm. However, bearing in mind the core-shell structure of these particles, we modelled the lineshape instead using a contribution at the bulk Si phonon frequency (intensity and FWHM varied but position fixed at 521 cm⁻¹) and a second peak due to the nanoparticles; the fit then gives a position of 510 cm⁻¹ for the second band (and FWHM 23 cm⁻¹), which then implies a slightly smaller mean diameter of 2.3 nm for the nanoparticles in the remaining shell. A representative fit and its components (including the aerogel peak) are shown for one LO aerogel in Figure 2: the band arising from the porous shell is clearly dominant. Finally, the SO particles are solid and approximately spherical and so are rather analogous to Si NPs grown by ion-implantation in bulk silicon; in aerogel, we obtain a mean diameter of 2.5 to 3 nm if we interpret their Raman peaks (at 505.5 cm⁻¹, FWHM 23 cm⁻¹) according to the same model 22 .

We now consider the concentration-dependence of the Raman scattering bands. Clearly, as the nanoparticle concentration increases and the resulting composites change from highly transmitting to opaque at the excitation energy, the penetration depth of the excitation light and the total scattering volume reduce dramatically. However, the well-defined aerogel Raman D₂ band at 620 cm⁻¹ provides a convenient means of comparing peak intensities, because it must arise from scattering within the same macroscopic volume of sample as the silicon Raman band. By taking the ratio of the integrated intensity of the silicon band (treated here, for simplicity, as a single band for any given nanoparticle type) to the aerogel D₂ band, we should obtain a quantity proportional to the nanoparticle concentration. The validity of this normalisation depends on the structure of the silica aerogel itself not being modified by the presence of the embedded material. This may not be true; it is known, for example, that the ratio of the D₁ to D₂ bands gives an indication of water content in silica aerogels 26 and is modified after the UV-induced formation of CdS crystallites 1 ; however, we do not expect the introduction of silicon into silica to be as dramatic a modifier of the Si-O network as water.

Figure 3 shows a test of this idea: the ratio of the silicon Raman band to the aerogel D₂ band is plotted as a function of the mass density of Si NPs introduced into the gel preparation, for each type of nanoparticle. Linear fits to each set of data are shown; these were not constrained to pass through the origin and do not do so exactly. It can be seen, however, that the fits converge near the origin and that the assumption of a linear relationship is reasonable. It is also apparent that the gradient of the linear fits depends on the type of nanoparticle, and this can easily be understood. Firstly, we assume that, in the case of the SO nanoparticles, a significant proportion of the particles is likely to be lost during the gel preparation process, because their mean diameter is much less than the aerogel pore size (typically 10 to 50 nm), and so they are less easily immobilised in the gel network. This accounts both for the weak degree of colouration of these composites and the low rate of increase of the Raman band with concentration. Secondly, the reduction in Raman strength of about a factor of 2 between the LO particles and the LH particles (for the same initial concentration) is qualitatively consistent with our earlier conclusion that the porous shell of the LH particles becomes oxidised, leading to a loss of scattering cross-section of the silicon phonon mode. Of course, the fundamental Raman scattering cross-section for these different types of particle is not necessarily identical, and any variation of this will also lead to a difference in slope of the fits in Figure 3. To estimate the importance of this effect, we plan to investigate the dependence of the Raman spectra on excitation energy.

Finally, we have tested the above conclusions using FTIR, since this provides a sensitive identification of Si-O-Si and Si-H vibrational modes via their IR absorption bands at around 1,030 and 1,180 cm⁻¹ (Si-O-Si) and 2,000 to 2,200 cm⁻¹ (Si-H, Si-H₂ and Si-H₃ stretches) 9 27 . On Figure 4, the vertical dashed lines show the positions of the bands associated with the three Si-H stretching modes. These modes are readily seen in, for example, conventional electrochemically etched PSi (top) even after it is several months old and substantially oxidised (which is indicated, for instance, by the presence of the band at 790 cm⁻¹ marked by the dotted line 9 ). The Si-H modes are also weakly present in the IR spectra of the nanoscale silicon particles before and even after aerogel preparation (second from top) despite the fact that they are mostly oxidised. The FTIR spectra of the composite aerogels made using LH particles, however, do not show any sign of these bands, and we obtain spectra that are dominated by the IR absorption of silica aerogel (shown at the bottom of Figure 4 for comparison) even for samples that contain sufficient concentrations of nanoparticles that they are opaque for visible light. The silica aerogel IR spectra show the expected Si-O modes but also two weak modes arising from Si-CH₃ groups indicated by the two arrows 28 . The fact that these are only weak is consistent with the aerogel surface being hydrophilic (a large Si-CH₃ coverage leads to a hydrophobic surface). Finally, as one would expect, we see no sign of Si-H modes in the aerogels containing large, oxidised particles, and their FTIR spectra (not shown in Figure 4) are very similar to those of our silica aerogels and those reported in the past 28 ; we note that the Si-Si modes of the bulk silicon cores will not be detected via IR absorption.