Chapter 5 Physical Characterization of Si Nanocrystals
5.3.2 Effects of the Annealing Temperature on the Silicon Agglomeration
In order to investigate the Si agglomeration behaviors, different RTA temperatures have been tested for the in-situ annealing. 3.5-nm-thick a-Si layers were annealed at different temperatures (700, 750, 800, and 850 °C) for 5 min. A constant annealing duration of 5 min was used in order to limit the thermal budget in the process. The capability of the UHV system was also considered while setting the experimental variables. For example, the annealing temperature at 850 °C was close to the highest temperature that could be handled by the currently-used in-situ RTA system.
The dot densities and mean sizes were taken by SEM as a function of the annealing temperature, as shown in Fig. 5.6. Figures 5.6(a), 5.6(b), and 5.6(c) show SEM images of samples annealed at 750, 800, and 850 °C, respectively. The sample annealed at 700 °C did not reveal resolvable features in the SEM analysis. When the annealing temperature was at 750 °C or higher, however, evidence of Si-dot
Si-dot density and a larger averaged dot size, although the changes in the dot densities and sizes were not significant.
However, AFM analysis with the sample annealed at 750 °C has revealed an incomplete agglomeration of silicon, as shown in Fig. 5.8(b). Therefore the dot size and density observed by SEM for this specific sample can not represent the real results of a complete dot agglomeration process.
Figures 5.8(a)–5.8(d) show AFM scan images (1 μm by 1 μm) for samples annealed at different temperatures. When the annealing temperature was ≥800 °C, dense dots were observed after 5-min RTA, as shown in Figures 5.8(c) and (d). When the annealing temperature was ≤700 °C, no silicon dot could be found under SEM, and AFM analysis showed an rms surface roughness less than 0.2 nm. Figure 5.8(a) shows the AFM image of a sample annealed at 700 °C for 5 min, and the rms roughness is measured 0.18 nm, which is only 50% higher than the as-deposited a-Si roughness (0.12 nm). This observation suggests that an annealing temperature lower than 700 °C either is not enough to trigger the thermal agglomeration or needs an annealing duration much longer than 5 min to complete the process.
In Fig. 5.8(b) the Si-dot agglomeration is found evident after annealing at 750
°C. However, darker regions surrounding the Si dots are spotted in the AFM image, indicating recessed areas which are possibly exposed tunnel-oxide surfaces. The intermediate areas between the bright dots and the dark spots are therefore representing the flat Si surface, and it becomes clear that the agglomeration process is incomplete with this sample. The in-situ annealing at 750 °C did start the Si-dot formation in a 3.5-nm a-Si film, but the annealing time of 5 min was not enough for the entire film to completely agglomerate into nanoparticles. Figure 5.9(a) shows a magnified portion of the AFM image, with the cross-sectional surface profile along the Y–Y’ line plotted in Fig. 5.9(b). The recessed area is labeled and correspondingly
marked in the cross-sectional plot by two dashed lines. The height differences between the recessed areas and the flat silicon surfaces are approximately 2–4 nm, showing consumption of Si top layer. Considering the original Si thickness of 3.5 nm, the surface coalescence of the Si film can contribute to a rougher surface and a less uniform thickness. It is noteworthy that a slightly thicker edge can be observed along the remaining Si film, which coincides with the void-edge thickening observed and discussed in the thermally-induced de-wetting of ultrathin SOI films [48, 123, 124].
Some small dots on the flat-Si regions are observed in Fig. 5.8(b), which suggest the nucleation of Si nanoparticles on a clean surface. Figure 5.10(a) shows a magnified AFM image, with a cross-sectional profile along the X–X’ line plotted in Fig. 5.10(b). The X–X’ line is placed through the middle of a small dot on the Si surface. It is obvious in the AFM profile that a Si dot has agglomerated on a relatively flat silicon surface, possibly due to the surface migration and nucleation of Si atoms.
The surface diffusion and nucleation of Si have been studied in explaining the HSG formation on a thick a-Si film [23, 39, 40]. In present study, the dot formation on the Si surface can be also resulted from similar mechanisms.
Figure 5.11 shows cross-sectional high-resolution TEM (HR-TEM) images of the sample appeared in Fig. 5.8(b). A separated Si nanocrystal on the tunnel-oxide layer is observed. The nanocrystal is formed after annealing a 3.5-nm a-Si film in situ at 750 °C for 5 min, and the self-assembly of crystallized Si nanoparticles is evident under HR-TEM. Because there seems no remaining a-Si layer in Fig. 11(a), the image is supposedly taken from within the dark (recessed) areas in Fig. 5.9, where the Si top
dash line, with two arrows indicating two mirrored arrays of Si atoms. When thinner a-Si layers were annealed, smaller Si nanocrystals with single-crystal structures would be obtained. The effects of initial a-Si thickness on the agglomerated nanocrystal properties will be discussed later in section 5.3.4.
Figure 5.12 shows another HR-TEM image of the same sample appeared in Fig. 5.8(b). Different from what is observed in Fig. 5.11, there is a remaining a-Si layer beneath the Si nanocrystal in this micrograph. Therefore one can assume that Fig. 5.12 is more representative of a partially agglomerated Si dot on the silicon surface, as being the case in Fig. 5.10. There are two observations of Fig. 5.12 which may provide insights of the Si agglomeration behavior. First, there are amorphous regions at the sides of the hemispherical crystal. The diffusing Si atoms may migrate along the Si-dot surface and accumulate near the edge before eventually aligned and crystallized with the nucleated core. Second, the crystallized structure in the middle of the dot has propagated into the underlying a-Si layer, forming a curved interface between the crystal and the a-Si layer. This suggests that the nanocrystal growth could have started at nucleation sites near the a-Si surface. After the surface nucleation the nanocrystal can grow along two fronts. One is the re-crystallization process propagating into the underlying a-Si thin film. The other is the alignment of impinging Si atoms, which can migrate on a clean surface while thermally activated, to the crystalline structure.
Briefly, a 5-min RTA at 750 °C did start the thermal agglomeration process in a 3.5-nm a-Si layer, but it was not enough to transform the entire film into separate nanocrystals. A longer annealing time should be needed if the thermal agglomeration process had to be completed at 750 °C. In the following experiments, an in-situ RTA at 850 °C for 5 min was used for the Si-nanocrystal formation, unless specified otherwise.
5.3.3 Effects of the Native Oxide Growth on the Silicon Agglomeration and Nanocrystal Formation
If the thermal agglomeration technique is to be integrated into the Si-based semiconductor manufacturing process, it will be important to consider its integration capabilities with other processing steps. Unlike the bottom-up CVD techniques, which randomly deposit Si dots all over the wafer surface, the thermal agglomeration technique has the potential of selectively patterning the initial a-Si layer prior to the vacuum annealing. Therefore it becomes possible to confine the dot formation in selected areas with a top-down, subtractive process. However, adding process steps in between the a-Si deposition and vacuum annealing steps would expose the wafer to air and therefore grow a thin native oxide on the a-Si surface. It has been emphasized that a clean surface is critical for the HSG formation, and surface oxide passivation can severely inhibit the surface diffusion and nucleation of Si atoms [23, 126]. In this section, different oxidation/annealing conditions are tested, and effects of the surface oxide growth on the nanocrystal agglomeration are discussed. It is found that a dilute-HF dip before the vacuum RTA is efficient to remove the surface oxide and ensure Si-dot agglomeration.
Ex-situ RTA in N2 ambient
A 3.5-nm a-Si layer was deposited on a tunnel-oxide/Si substrate and then annealed ex situ at 850 °C for 5 min in N . AFM measurements were made to
layer grown in air or during the N2 RTA has prevented the a-Si film from coalescence and agglomeration.
Figure 5.13 shows O 1s and Si 2 p spectra acquired by ex-situ XPS analysis before and after the N2 RTA. Normally after a silicon sample is exposed to air, a native oxide layer growth occurs on the surface. Therefore it is reasonable to assume that a native oxide on a-Si has contributed to the oxide signals in the XPS spectrum for the as-deposit sample. Intensities of the oxide peaks (O 1s and Si4+ 2p) slightly increased after the N2 RTA. This implies an oxide re-growth during the RTA procedure. A small amount of residual oxygen may have existed in the RTA processor chamber and reacted with the a-Si surface during the N2 RTA. Atmosphere moisture, which is usually absorbed on the wafer surface or on the chamber walls, can also contribute to the oxide re-growth. In Fig. 5.13(b) the silicon peak (Si0 2p) shows no distinguishable changes, which indicates that only a very small amount of a-Si is oxidized during the RTA. The XPS analysis is more sensitive to the native oxide re-growth than it is to the silicon consumption because the native oxide is the very top layer which has the most efficient photoelectron response.
In-situ oxidation in low-pressure O2
In order to verify that the silicon agglomeration is hindered by the surface oxide bonding, an in-situ oxidation test was performed. A sample was deposited with a 3.5-nm a-Si layer and then moved to another chamber without breaking vacuum.
The sample was then oxidized under a low-pressure condition. The oxidation was performed at 650 °C for 20 min, while the chamber was purged with pure O2 flow (250 sccm) and maintained at 12 milli-Torr. After the low-pressure oxidation, the sample was moved back to the RTA chamber (without breaking vacuum) for an in-situ annealing at 850 °C for 5 min.
Figure 5.14 shows O 1s and Si 2 p spectra acquired by in-situ XPS analysis after each step. Measurements on the as-deposited a-Si have shown small oxide peaks (O 1s and Si4+ 2p) which can be attributed to the tunnel-oxide interlayer. The energy separation (ΔEB) between the O 1s and Si4+ 2p peaks reads 429.3 eV. The same ΔEB value is also observed with the a-Si/oxide/Si samples discussed in section 5.3.1. After the low-pressure oxidation at 650 °C, the O 1s intensity increases significantly due to oxide growth. The decrease of the Si0 2p peak intensity is also a result of the silicon consumption and the top-oxide growth. After the oxidation the O 1s peak position shifts toward a smaller binding energy. The Si 2p oxide peak also appears to have a tail on the lower-EB side after the oxidation. These observations can be attributed to the existence of sub-oxides, or sub- stoichiometric SiO2 (SiOx or Sin+ where n=1, 2, 3) [118]. A multi-peak fit of the Si 2p spectrum recorded after oxidation reveals a combination of Si0, Si1+, Si2+, Si3+, and Si4+ peaks, as shown in Fig. 5.15. It appears that the oxidation at a relatively low temperature has produced a transition layer of sub-oxides.
After in-situ annealing of the partially oxidized silicon film, the XPS spectrum only shows a small increase in the peak intensities for all peaks in Fig. 5.14. The peak shapes and positions stay the same. AFM analysis on the sample shows no Si-dot formation. The rms roughness is measured 0.16 nm (sample 1329 in Table 5.3), which is only a little rougher than the as-deposited a-Si surface (0.12 nm). The surface may have been roughened by the a-Si crystallization during RTA. A slightly rougher morphology can hence provide more surface responses to the XPS detector and
A 5-nm a-Si film was immersed into an SPM solution to grow chemical oxides on it and then dipped in dilute-HF for the oxide removal. The sample was then immediately loaded back into the UHV system for the vacuum RTA at 850 °C for 5 min (see sample 1328 listed in Table 5.3).
Figure 5.16 shows O 1s and Si 2 p spectra taken by in-situ XPS analysis before and after the chemical-oxide growth/removal treatments. A decrease of the Si0 2p peak (EB = 99.3 eV) and increases of the oxide peaks (O 1s and Si4+ 2p) are observed after the SPM+HF treatments. These observations are attributed to the a-Si thickness reduction, which also allows more photoelectrons from the oxide interlayer to be detected. Obvious the growth of chemical oxides has consumed the a-Si top layer and reduced its thickness. It is worth noting that a tail of the O 1s peak (on the lower-EB side) is found after the SPM+HF treatments. Figure 5.17 shows curve-fitting results for the O 1s peak. Two components are extracted (peaks at 532.8 and 531.4 eV) while the entire spectrum is referenced to the Si0 2p line at EB = 99.3 eV. The O 1s peak component at 532.8 eV is attributed to the tunnel-oxide interlayer. The other O 1s component at the peak position of 531.4 eV can be indicating hydroxyl groups or hydroxide contaminations on the a-Si surface [52].
Figures 5.18 and 5.19 show the SEM and AFM results taken on the sample after the 850 °C vacuum RTA. The rms roughness is approximately 4.9 nm measured by AFM, as listed in Table 5.3 (sample 1328). The thermal agglomeration of silicon dots is evident, even though the sample has been through ex-situ treatments prior to the vacuum RTA. Therefore one can conclude that this agglomeration technique is applicable even after the sample has been exposed to air for additional process steps, provided that the native oxide is removed prior to the vacuum annealing. One can also envision the possibility of removing the a-Si in selected areas by a subtractive process before the vacuum annealing; thus Si agglomeration and nanocrystal formation can be
confined in selected areas. Though the additional photolithography and a-Si etching steps may increase the manufacturing costs, which is a trade-off.
5.3.4 Dependence of Nanocrystal Sizes and Densities on the Initial a-Si Thickness
Figure 5.20(a) shows an SEM image of Si nanoparticles agglomerated after in-situ annealing of a 3.5-nm-thick a-Si layer. The size distribution of Si dots observed under SEM is summarized in Fig. 5.20(b). There is one group of larger dots, with their base radii around 10–14 nm. Also there is another group of smaller dots with base radii around 4–8 nm. This bimodal size distribution has been reported in studies of the thermally-induced island formation with a-Si [42, 43] or SOI [47] thin films, and it can be explained by a multi-stage agglomeration process.
In the first stage nucleation centers (seeds) are created on the silicon surface by strain relaxation in the form of defects or micro-roughness [40]. Then the three-dimensional aggregation of Si atoms forms small hemispherical crystallites on the Si surface, as shown in Figures 5.10 and 5.12. The aggregation requires surface atoms to migrate; therefore it is critical to maintain a clean Si surface. The hemispherical crystallites grow laterally when the impinging atoms merge into the crystal structure [23]. The crystallites also act as seeds and re-crystallize the underneath a-Si. In other word, the growth of crystallites also propagates from the a-Si surface toward the a-Si/oxide interface. The crystal growth eventually reaches down to the a-Si/oxide interface and depletes the surrounding a-Si film, which
obtained when the free energy changes no longer favor the crystal growth. A group of larger Si dots are formed when Si atoms are sufficiently supplied until the critical size is approached. Some remaining a-Si will then agglomerate into a second group of nanocrystals whose sizes are smaller because of inadequate Si atoms.
Figures 5.21(a) and 5.22(a) present SEM images of Si dots agglomerated from initially 2.5-nm and 1.8-nm a-Si layers, respectively. The dot size distributions are plotted in Figures 5.21(b) and 5.22(b). It appears that thinner a-Si layers have resulted into smaller and denser nanoparticles, in agreement with the experimental and theoretical studies on the thermal agglomeration of ultrathin silicon films [42, 48].
The bimodal size distribution can still be seen, for example, in Fig. 5.22(b).
Cross-sectional TEM analysis has again verified the hemispherical shape of Si nanocrystals. Figure 5.23(a) shows an annular dark-field scanning TEM (ADF-STEM) image of the Si dots self-assembled from a 1.8-nm-thick a-Si film. It is clearly shown that hemispherical Si dots are distributed on the tunnel-oxide layer and separated from each other. Some seemingly merged dots are actually separated but with their images overlapped in the plane of focus. HR-TEM images of nanocrystals are also taken, as shown in Figures 5.23(b)–(d). Clear lattice fringes are observed. For example, in Fig.
5.23(c) the lattice spacing of 0.313 nm coincides with that of the Si (111) planes. The single-crystal structures in these photos suggest that more single-grain nanocrystals are formed with the thinner (1.8 nm) a-Si layer.
Figure 5.24 shows an AFM image of Si dots self-assembled after annealing a 0.9-nm-thick a-Si film. The rms roughness is 0.55 nm. The Si dots on this sample are so small that SEM encounters difficulties resolving the features. Therefore AFM analysis is used to evaluate the dot size and density. Si nanoparticles with a density as high as 3.9×1011 cm-2 and a mean radius of 5.1 nm were observed by using AFM.
However, it should be noted that AFM technique can potentially introduce probe
artifacts. When an pyramidal AFM tip goes over a nanoparticle attached to a flat surface, the side of the tip may interact with the nanoparticle and hence cause broadening features in the image [127]. Besides, the proximity of neighboring small nanoparticles can appear as a larger dot in scanning probe microscopic images [62].
Thus the mean size obtained by AFM is possibly overestimated, with the density possibly underestimated.
Figure 5.25 summarizes the nanocrystal dot densities and base radii as a function of the initial a-Si layer thickness. After an in-situ RTA at 850 °C for 5 min, the thinner a-Si resulted into denser and smaller Si nanocrystals. This result provides a simple route of the nanocrystal size and density control. Studies on the surface-energy instabilities have also suggested that the size and density of agglomerated dots are mainly determined by the initial Si thickness [42, 48].
Wakayama et al. [42] have discussed the mechanism of Si-island formation from a thin silicon film in relation to the free energy of the Si/SiO2 system. Agglomeration of the Si crystallite occurred forming Si islands, thus preventing a further increase in free energy. In our work, however, a higher dot density has been obtained, which can be attributed to the use of RTA instead of the equilibrium annealing conditions used by Wakayama and co-workers. In addition, Nuryadi et al. [45] have studied the thermal agglomeration of single-crystalline silicon-on-insulator (SOI) layers. They also found that the size of agglomerated islands increased with increasing SOI film thickness, and that the critical agglomeration temperature was lower for the thinner SOI film. In our study, a much higher dot density was obtained by annealing a thin a-Si layer. The
In this section a simple model is established to estimate the Si-dot size by using in-situ XPS analysis results. After the nanocrystal size is extracted from XPS data, the dot density can be calculated accordingly.
Figure 5.26 is a schematic showing one unit area of a silicon film. The a-Si film is deposited over a tunnel-oxide layer, and therefore any photoelectron signals from the oxide interlayer must go through the Si top-layer. After in-situ annealing, presumably all the a-Si atoms in the unit area agglomerate into one hemispherical dot, and the oxide surface becomes partially exposed. Figure 5.27 shows schematics of an
Figure 5.26 is a schematic showing one unit area of a silicon film. The a-Si film is deposited over a tunnel-oxide layer, and therefore any photoelectron signals from the oxide interlayer must go through the Si top-layer. After in-situ annealing, presumably all the a-Si atoms in the unit area agglomerate into one hemispherical dot, and the oxide surface becomes partially exposed. Figure 5.27 shows schematics of an