Wet and Siconi® Surface Preparation Sequences for SiGe Epitaxial Regrowth

Abstract

The efficiency of the “Siconi®” process, which is based on the use of a NH3 / NF3 remote plasma and anneals at temperatures less than 180°C, was recently assessed on SiGe surfaces. Siconi® removed GeO2 and SiO2, but was less efficient on GeOx. The presence of residual GeOx after a Siconi® process was minimized with the use, beforehand, of a wet oxidation. It consisted on dips in Standard Cleaning 1 (SC1) solutions made of NH4OH, H2O2 and H2O or in water with a few ppm of Ozone. A SiO2 - rich oxide was generated, then (P.E. Raynal et al, Microelec. Eng. 187-188, 84 (2018)). In this study, we elaborate on those findings by comparing the performances of (i) wet, (ii) Siconi® and (iii) “wet - Siconi®” sequences on re-epitaxy on SiGe surfaces. Those sequences were evaluated on 15 nm thick Si0.6Ge0.4 layers with different Queue-times (15 min and 8 hours) between wet and Siconi® treatments. Without any air break, the Siconi® process was followed by a low temperature H2 bake (at 20 Torr and a temperature less than 700°C) and a re-epitaxy at 600°C of 15 to 17 nm of Si0.6Ge0.4. Secondary Ions Mass Spectrometry (Figure 1) showed that a “chemical oxide - Siconi” sequence yielded a definite reduction of the interfacial oxygen concentration. Even with 8 hours of Q-time, concentrations were very low with these sequences. They were (i) more than one order of magnitude lower than with a regular “HF-HCl” (i.e. a “HF-Last”) wet cleaning and (ii) several times lower than with Siconi® only. Atomic Force Microscopy (AFM) images of the surfaces (Figure 2) of ~ 30 nm thick Si0.6Ge0.4 layers (after epitaxial regrowth) showed the presence of numerous islands after the use of “chemical oxide - Siconi®” surface preparations. Those defects were associated with poor crystalline quality layers. This was surprising. Indeed, such sequences yielded a superior oxide removal efficiency (Fig. 1) and smooth and defects-free SiGe surfaces just after their use (i.e. without any re-growth). Other epitaxial regrowth schemes using solely “HF/HCl” or Siconi® treatments did not result in such defects. The use of “chemical oxide - Siconi®” sequences must have resulted, after oxide removal, in SiGe surfaces richer in germanium than the “bulk” concentration (40%). They were then more sensitive to 2D-3D transitions during the H2 bake than “regular” SiGe 40% surfaces. Ge-rich islands were thus formed during the H2 bake that followed the “chemical oxide-Siconi®” sequence, destabilizing the epitaxial re-growth. In order to prevent surface islanding after "chemical oxide-Siconi®" sequences, we have then evaluated the interest of adding dichlorosilane (DCS) to H2 during the temperature stabilization (just after loading inside the epitaxy chamber), the temperature ramping-up then the H2 bake itself. Our aim was to encapsulate the extreme surface with a few atomic planes of Si in order to prevent 3D transitions. The reason why DCS was used was the following: with this chlorinated precursor, the homo-epitaxial Si growth rate decreased exponentially with the temperature (Ea > 2eV, typically), with a lower “boundary” of 650°C, at which it was only 5 Å/min. When sent on a SiGe surface at T < 650°C, growth with DCS essentially stopped as soon as a switch over to a pure Si surface occurred, resulting in really thin Si caps. The H2 annealing prior to SiGe epitaxy was divided into three steps:1) low temperature stabilization of the wafer (after loading) ;2) ramping-up to the bake temperature ;3) low thermal budget H2 bake (2 min. at T < 700°C).Three epitaxial regrowth schemes, with "HF/HCl-SC1-Siconi®" surface preparations then injections of dichlorosilane during step 1) only, steps 1) and 2) or steps 1), 2) and 3), were evaluated. As shown by AFM (Figure 3), injection of DCS during the various annealing phases drastically reduced the density of 3D defects or suppressed them. The Si spacer thickness between the 15 nm thick Si0.6Ge0.4 layer and the 17 nm of re-epitaxy after the addition of DCS during steps 1) and 2) was 2 nm only (Figure 4). This is acceptable from an integration point of view, especially in SiGe:B sources and drains grown on top of high hole mobility SiGe channels, as such stacks will be germano-salicided afterwards, anyway. Figure 1