Tailoring the surface density of silicon nanocrystals embedded in SiOx single layers

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés. Tailoring the surface density of silicon nanocrystals embedded in SiOx single layers S. Hernandez, Patrice Miska, M. Grün, S. Estradé, F. Peiro, B. Garrido, Michel Vergnat, P. Pellegrino


I. INTRODUCTION
In the last two decades, silicon nanocrystals (Si-NCs) have been investigated for their fundamental properties as well as for potential applications in fields of electronics and photonics. 1,2][5][6][7] Several different methods for fabricating Si-NCs, all compatible with the semiconductor industry standards, have been developed, such as ion implantation, 8 cosputtering, 9 or plasma enhanced or low pressure chemical vapor deposition (CVD). 10Even so, all these techniques face difficulties in accurately controlling the Si-NCs size and density, leading to a certain degree of uncertainty in interpreting how the actual Si-NCs morphology affects their electronic properties.
Many efforts have been put so far to control the Si-NCs size, either by controlling the silicon concentration, the postdeposition thermal process [8][9][10] or limiting the thickness of SiO x layers between SiO 2 stoichiometric barriers (as the one proposed by Zacharias et al. 11 ).Nevertheless, only few works can be found in literature dedicated to control and analyze the Si-NC areal density in similar systems. 12The approaches used are generally based on an estimation of the Si-NC volumetric density by transmission electron microscopy (TEM) in samples with either bulk silicon-rich oxide (SRO) systems or SRO/SiO 2 multilayered samples and, through it, the areal density with large uncertainty. 12In addition, changes in the density of nucleation centers can lead to Si-precipitates with structural properties strongly dependent on their environment. 13n the present work, we analyze the role of the silicon excess in the formation of Si-NCs in a single SiO x layer, focusing the study on the modification of the Si-NC areal density.We use a direct method based on energy filtered TEM analysis applied to a single Si-NC layer to quantitatively evaluate this property.Additional high resolution TEM images from the same areas have been acquired, in order to determine the crystalline size of the Si precipitates.Finally, the optical properties of the Si-NC were monitored by means of photoluminescence (PL) measurements at 77 K.A direct correlation between the structural and optical properties has been established, finding that PL intensity scales with the crystalline degree of the Si aggregates.These observations suggest that, for stoichiometries close to that of SiO 2 , the inhibited crystallization is related to the higher compressive local stress, which strongly affects the radiative transitions.

II. EXPERIMENTAL
Two different sets of samples with variable silicon concentration of SiO x were fabricated by evaporation on (100) silicon substrates at 100 C: samples with a single SiO 2 /SiO x /SiO 2 structure and samples containing 20 bilayers of SiO x /SiO 2 .The stoichiometric SiO 2 layers were deposited by evaporating targets of pure SiO 2 , whereas the sub-stoichiometric SiO x ones were obtained by co-evaporating SiO and SiO 2 from a thermal source and an electron beam gun, respectively.The final composition was obtained using different evaporation rates.Taking account the reaction SiO x !(1-x/2) Si þ x/2 SiO 2 , the resulting silicon excesses in the SiO x layers were of 25, 16, 10, and 5 atomic % (at.%) that correspond to relative oxygen-to-silicon concentrations of x ¼ 1, 1.25, 1.5, and 1.75, respectively.In all samples, the thickness of the SiO x layers was kept constant at 3 nm, while the SiO 2 barrier layers were of 9 nm and 5 nm, for the case of SiO 2 /SiO x /SiO 2 single layers and SiO x /SiO 2 multilayers, respectively.During the evaporation of either SiO x or SiO 2 , the rate was monitored by a quartz microbalance and maintained constant at 0.1 nm/s.After deposition, the samples were annealed at 1050 C for 5 min under nitrogen atmosphere in a rapid thermal annealing system.
The precipitation process was monitored by Fourier transform infrared spectroscopy in 200-nm thick SiO x samples annealed at different temperatures (not shown), by following the shift of the position of the Si-O-Si asymmetric stretching mode.We found an excellent phase separation for all the used compositions when the samples are annealed at temperatures above 900 C: in all samples, the Si-O-Si stretching mode presents a position (x % 1080 cm À1 ) and a lineshape typical for pure SiO 2 treated at this temperature or higher.
Slices for transmission electron microscopy using SiO 2 /SiO x /SiO 2 single layers were fabricated by the purely mechanical tripod method, in order to minimize any influence by the preparation method on the structural properties of the system.In-plane high resolution TEM (HRTEM) and energy filtered TEM (EFTEM) measurements were performed both in the same areas using a JEOL 2010F TEM operating at 200 keV coupled with a Gatan image filter, with a resolution in energy of 0.8 eV.In the case of EFTEM measurements, the Si contrast was enhanced by energetically filtering the TEM image, choosing only the electrons with an energy loss within a window around the Si plasmon energy (E Si ¼ 17 eV).Consequently, valuable information about the size and density of the precipitated Si nanoclusters of the single SiO x layers can be extracted directly from those images.
PL measurements were performed on both sets of samples at 77 K in a liquid-nitrogen-cooled optical cryostat.The samples were excited by using the 313-nm line of a mercury arc lamp with a power density of 5 mW/cm 2 , low enough to neglect any local temperature effects.The spectra were acquired in the range of 500-1000 nm and analyzed by using a monochromator equipped with a 150 grooves/mm grating and by a high sensitive CCD detector cooled at 140 K.

III. RESULTS AND DISCUSSION
TEM analysis was performed in SiO 2 /SiO x /SiO 2 single layers observed in planar view geometry.We present in Fig. 1, the energy filtered and high resolution TEM images of the precipitated Si-clusters in SiO 2 /SiO x /SiO 2 samples for different stoichiometries: temperature annealing for all the spanned range of stoichiometries, obtaining similar sizes and surface densities.For each sample, large areas have been analyzed to accurately determine both parameters.Consequently, size distribution of the whole cluster and their areal density can be extracted from EFTEM images, while the size distribution of the crystalline precipitates is obtained by means of HRTEM imaging (only the precipitates oriented along a high symmetry crystalline direction are observed using this configuration).Assuming that the Si-nanoaggregates have a spherical shape and are formed by a crystalline core surrounded by an amorphous shell (coreshell model 14,15 ), EFTEM and HRTEM configurations are complementary, providing information either from the whole clusters or only on their crystalline core, respectively.We found that the diameter of the Si-aggregates obtained by both configurations nicely follow a log-normal distribution, f(d clu,cry , r clu,cry ) where d clu,cry , d 0 clu,cry , and r clu,cry are the diameter, mean diameter, and broadening of the size distribution, respectively, for either whole clusters (clu) or crystalline cores (cry).Those parameters have been determined in all the samples by fitting the experimental distribution to the previous equation.In addition, the Si-NC areal density has been also determined using the data from EFTEM images.In Table I, we have summarized the results obtained from both techniques.
We have found that the diameter distribution of the whole cluster presents a similar distribution for all compositions, with a mean size d 0 clu around 2.7-2.8nm (see Table I).The small discrepancies from sample to sample in the size distribution are the consequence of the uncertainty in the size determination, indicating that an almost identical size distribution is obtained for all stoichiometries.Nevertheless, the amount of Si-nanoaggregates is modified for the different samples, finding a variation of their areal density from 3.0 Â 10 12 cm À2 to 3.8 Â 10 12 cm À2 , as the stoichiometry changes from x ¼ 1.75 to x ¼ 1.0.By using these values of areal density and mean sizes, the average distance between clusters was also evaluated.Even though the areal density variation is rather small, only around 30%, it is large enough to produce an important modification in the inter-dot distance.In fact, aggregates are separated by 2.2-2.4 nm for the highest silicon content (i.e., x ¼ 1.0 and 1.25), while the average separation is much larger, taking values of 2.7 and 3.0 nm, for the lower silicon content (x ¼ 1.5 and 1.75, respectively).This difference in the inter-dot distance can be directly observed on the EFTEM measurements: Si-NCs are very close to the adjacent ones in the samples with highest Si content [Figs.1(a Therefore, we demonstrated that, by employing SiO 2 /SiO x /SiO 2 structure-like systems and controlling the thickness and stoichiometry of the SiO x layer, it is possible to modify the surface density, while keeping almost constant the size of Si precipitates. Further analyses have been performed on the same set of samples (also on the same areas) by considering the images obtained by HRTEM [see Figs.1(e)-1(h)].We found that the mean crystalline diameter d 0 cry for each sample varies from 2.5 nm to 1.9 nm as the silicon content decreases (x from 1 to 1.75), values much smaller than the ones observed by EFTEM.As we mentioned above, this discrepancy is explained in terms of the amorphous Si-shell that surrounds the crystalline core. 14,15We found that the thickness of the amorphous shell gets larger as the stoichiometry approaches the one of pure SiO 2 , while the total cluster size is almost constant in all the explored range.The crystalline fraction (i.e., relative volume ratio between the crystalline and amorphous silicon regions) has been estimated by considering the size distribution of the whole clusters and the crystalline parts, obtained from the images of both TEM configurations.We observed that the crystalline fraction scales with the silicon content, with values ranging from 40% to 75% (see last column of Table I).Therefore, the stoichiometry of the SiO x layers is affecting both the crystallinity and the areal density, obtaining a reduction of both magnitudes as the Si content is reduced.
The optical properties of the precipitated Sinanoaggregates have been studied by means of PL.In Fig. 2(a), we present the PL emission of SiO 2 /SiO x /SiO 2 samples at 77 K, where a strong emission between 700 and 800 nm (typical from silicon nanoaggregates) 16,17 is observed only for samples with the highest silicon content (i.e., x ¼ 1.0 and 1.25).The PL peak position of samples with x ¼ 1.0 and 1.25 shifts to longer wavelengths, from 1.76 to 1.65 eV, respectively, as the silicon content decreases.Nevertheless, an intensity drop occurs as the silicon stoichiometry gets closer to SiO 2 , getting no signal for samples with the lowest silicon content; i.e., x ¼ 1.5 and 1.75.In fact, these samples with the lowest Si content present only a broad and weak defect-related emission in the blue-green region that is TABLE I. Size distribution from the whole Si-cluster and from the crystalline core determined by EFTEM and HRTEM, respectively.The surface density was determined directly from the EFTEM images.The crystalline fraction has been calculated considering the size distribution obtained from both techniques.

EFTEM HRTEM
x associated to oxygen vacancy defects characteristics to amorphous SiO x . 18n order to extract information of the optical emission from stoichiometries closer to SiO 2 , special multilayered samples containing 20 periods of SiO 2 /SiO x bilayers nominally identical to the single layers were fabricated to increase their active volume and, in turn, their optical emission.In Fig. 2(b), we present the PL spectra of the multilayered samples for different stoichiometries.An intensity increase with respect to the single layers was observed in all the multilayered samples, which scales with the number of multilayers, allowing to detect Si-NC related emission also from sample x ¼ 1.5.Despite the fabrication of this new set of samples, no signal was observed for the lowest silicon content, x ¼ 1.75, which is a strong indication that sufficient crystallization is not yet attained for this composition, in agreement with our TEM observations.
The PL peak position in the multilayered samples presents the same trend than in the single SiO x layers: there is a shift to shorter wavelengths as the silicon content decreases.However, there is a clear difference in the emission energy for single and multilayered samples with the same nominal stoichiometry: lower energy emission is observed for multilayered samples, compared to the single ones, of about 67 and 28 meV for x ¼ 1 and 1.5, respectively.In any case, the energy emission of both sets of samples lays well within the reported values found in the literature for samples with similar stoichiometry. 16,17ctually, the PL energy shift to higher energies observed in both sets of samples (single and multilayered ones) as the stoichiometry approaches to SiO 2 is related to an increase of electronic quantum confinement in small Si-NCs, favored by the low Si excess (the control of the Si content has been widely employed by many authors to tune their emission energy; see, for instance, Refs.19 and 20).Therefore, the PL energy should be correlated to the size of the Si-aggregates.Through TEM characterization, we found that there is a similar Si-cluster size for all samples, but presenting a crystalline size reduction for lower Si content, in agreement with the PL energy band displacement for different stoichiometries.Consequently, we represented, in Fig. 3(a), the PL peak energy of the two sets of samples (single layers and multilayered samples) as a function of the crystalline size.For both sets of samples, the PL peak energy decreases for larger crystalline sizes following an inverse power law: 21 being E Si the band-gap energy of bulk silicon, d the crystalline diameter, and d the decay factor.The obtained decay factor for multilayered samples was found to be 1.27, in very good agreement to report values in Si-NC embedded in SiO 2 . 14,21On the other hand, a larger d was found for single layer samples, obtaining a value around 2.1.However, in the latter case only two points have been used for its determination, giving rise to a large inaccuracy.Nevertheless, the higher emission energy for the two sizes is a clear indication that a much larger d value is associated to single layer samples.
Assuming that there is no change in the morphology of the precipitated Si-nanoaggregates in both sets of samples (the relative PL intensity between the two sets perfectly scales with the number of multilayers), the energy difference of some tenths of meV observed between single layer and multilayered samples could be related to a small loss of electronic confinement by interaction between adjacent SiO x layers.Actually, the distance between adjacent layers in the multilayered samples is short enough (the SiO 2 barrier thickness is 5 nm) that can induce the excitonic migration between clusters, either via tunneling of individual electrons or holes, or by resonance energy transfers by a dipole-dipole coupling. 22Both effects favorize the overlapping of the wavefunctions from Si-NCs in adjacent SiO x layers, reducing the electronic quantum confinement.Consequently, the PL emission from Si-NCs in multilayered samples would present a peak energy slightly lower than the ones in single layer samples, as the latter ones are close to an isolated system.This effect is well established for CdSe nanocrystals 23 and has been previously observed in Si-NCs/SiO 2 systems, presenting a slightly reduction in the electronic confinement energy as Si-NCs get closer. 24nother possible explanation arises by considering the different geometry of the two sets of samples (single or multilayers): thicker samples may be affected by a higher matrix induced compressive stress. 17,24,25The total SiO 2 thickness is of 105 nm for the multilayered samples and only of 18 nm in the case of the single layer samples, exhibiting a difference of more than a factor 5. So, Si-NCs in multilayered samples are more influenced by the SiO 2 matrix and are also submitted to a higher compressive stress than in single layers.Recently, Kůsov a et al. have reported the influence of the compressive stress in Si-NC/SiO 2 systems on the PL peak emission. 17They conclude that there is a large red shift in PL when the Si-NCs are submitted to a high compressive stress, which is typically induced by the surrounding SiO 2 matrix.The PL emission from our two sets of samples lays in the scattered data corresponding to compressed Si-NC in Ref. 17, which suggests that single layers are also subjected to a strong influence from the SiO 2 environment.Nevertheless, it is needed to also consider that, apart from the SiO 2 from the above and below layers, the SiO 2 between Si-NCs in the same suboxide layer may also contribute to the stress over the nanostructures.From the structural characterization, we observed that the inter-dot distance increases for lower Si content (due to the reduction of their areal density), increasing as well the amount of SiO 2 surrounding the nanoaggregates.Consequently, the Si-NCs from SiO x layers with lower Si content (i.e., smaller Si-NCs and larger inter-dot distance) are more likely to be submitted to a higher compressive stress than the ones from SiO x layers with higher Si content (i.e., bigger Si-NCs and lower inter-dot distance), affecting also their emission properties.
Once the possible origins of the different emission energies of the two sets of samples have been evaluated, we have analyzed the intensity emission as the stoichiometry of the SiO x layers is changed from x ¼ 1 to x ¼ 1.75.The multilayered samples were chosen for this comparison, as they present stronger emission than the single layers.As we commented above, there is a progressive intensity reduction of the PL integrated intensity as the stoichiometry approaches the one of SiO 2 , scaling with the crystalline fraction.In Fig. 3(b), we have depicted the relative PL intensity evolution as a function of the crystalline fraction, once normalized to the amount of Si atoms in the crystalline state (considering their areal density and mean crystalline size from Table I).With a good approximation, we observe that the intensity linearly depends on the crystalline fraction, presenting no detectable emission for crystalline fraction below 40%.Therefore, there is an onset of PL emission of about f c % 40%, indicating that the crystalline fraction is playing a major role in reducing nonradiative paths, and thus, enhancing the PL intensity.Similar results were reported for partially crystalline Si-nanoclusters, observing PL only from nanoaggregates with a sizeable crystalline component and no emission from those with large amorphous component. 26he oxygen out-diffusion has been proved to be the mechanism involved in the local phase separation that allows Si crystallization. 13Both structural and optical data point to an enhanced retardation of this diffusion for reduced Si excess, where the crystalline core of the precipitates is submitted to larger compressive stress.Higher annealing temperatures during longer times may produce Si-NCs with bigger crystalline domains, red-shifting and enhancing the PL emission and, at the same time, reducing the Siamorphous phase.However, the use of large thermal budgets may induce the destruction of the multilayered/single layer structure, leading to a loss of control in the Si-NC size, density, and crystalline fraction.

IV. CONCLUSIONS
We analyzed the formation of silicon nanocrystals in single SiO x layers with different stoichiometries, in order to explore the possibility of controlling the silicon nanostructure areal density.Using EFTEM and HRTEM, we have determined the size distribution of clusters and crystalline nanoaggregates.The size of the Si-clusters was found to be almost constant as the silicon content increases in the layers, whereas the areal density is slightly larger.The crystalline fraction of the nanoaggregates for each stoichiometry has been determined by combining the data from both TEM modes.The PL measurements show emission energies that increase as the stoichiometry gets closer to SiO 2 , together with an intensity reduction.These observations are in good agreement with the TEM data, which indicate a reduction of the crystalline sizes and a loss of the crystalline fraction for lower Si excesses.Moreover, the PL emission linearly scales with the crystalline fraction, finding an onset of the PL emission for 40% crystalline fraction.In the layers with a low Si excess the weak PL emission, together with the reduced crystalline fraction, indicate that the phase separation is not sufficiently attainted for the outer suboxide shell of the small nanoaggregates that could have its origin in the high compressive local stress.
FIG. 1. Transmission electron microscopy images of SiO 2 /SiO x /SiO 2 single layers for different silicon stoichiometries: x ¼ 1, 1.25, 1.5, and 1.75.Images from (a) to (d) were acquired by energy filtering the energy loss spectra around the Si-plasmon energy, while images from (e) to (h) were obtained using high resolution TEM (the crystalline clusters seen in the images are encircled in red).EFTEM and HRTEM images of each sample were taken in the same region.
) and 1(b)], and are well separated in the samples with the lowest Si content [Figs.1(c)and 1(d)].
d clu;cry ; r clu;cry !; in SiO x [This article is copyrighted as indicated in the article.Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP: