Si1-x-yGexCy epitaxial depositions (epi) on silicon wafers are of great interest as they allow dramatic performance improvements or even new devices fabrication. About 20 years ago, an actual breakthrough in SiGe epitaxy allowed an important extension of processes in terms of materials, chemistry and temperature range. Consequently, very different epitaxies are currently used in a variety of industrial technologies Historically, a number of epi techniques that have been developed to a more or less advanced degree: MBE, UHV-CVD, PE-CVD... Firstly, in this lecture, these techniques will be reviewed and benchmarked for industrial usage, thus their advantages/drawbacks, listed in table1, will be discussed. It appears that RTCVD presenting more advantages than competitors is indeed the most developed/adopted one, and that batch LPCVD is a serious challenger for depositions simple enough and for mass production (1). In a second part, important epi applications will be presented. The list is not exhaustive but based on the applications developed at STMicroelectronics for BiCMOS, Imagers, Photonics and CMOS technologies. Then, very different depositions will be reviewed, with thicknesses ranging from 3-6nm for CMOS to 3-6µm for Imagers, and from pure Si to pure Ge. A particular attention will be paid on the heterojunction bipolar transistors (HBTs) fabrication and on the problematic of epi on ultra-thin (3-5nm) SOI films, 2 items we have worked extensively on. Historically, HBTs first have used SiGe and advanced epi with a tremendous success. But the SiGeC:B base epi, especially when selective, is very demanding in terms of process control (complex stack and very thin elementary films) and in terms of structural quality (interface, defects and C incorporation). On the other hand, in advanced fully-depleted planar CMOS technologies, ultra-thin (3-6nm) SOI active zones are very sensitive to any non-oxidizing thermal budget, they require then special precautions before/during the epi process. The following part presents the challenges of advanced epi, namely: patterns, loading effect, limited thermal budget, selectivity, deposition control, uniformity, structural quality, morphology, defectivity, cost… And in industrial applications, contrarily to pure research activity, all of them have to be fulfilled at the same time; thus the process corresponds to a trade-off. Here, we will focus on the main difficulty of the RTCVD technique, i.e. the temperature control, and on the “small size effects” that becomes dominant in current (~14nm) and in future CMOS nodes.In current RTCVD single wafer systems, heating is radiative and there is no solution to measure accurately the wafer temperature with the desired accuracy, better than 1K. Thermocouples or pyrometers only measure the susceptor temperature. As the susceptor/wafer thermal coupling is not perfect, any vertical heat flux will gives rise to a temperature gradient and then to an offset between the set-point (susceptor) and the process temperature (wafer). The net heat balance varies with different parameters such as the bottom/top power ratio and the wafer optical properties, and the coupling depends on the wafer curvature and on the gas nature/pressure. Usually, the bottom heating is predominant and the vertical T gradient negative. Fig.1 reports a typical T offset under H2 and its variation as a function of the pressure. It is observed that the offset and its variation, are huge compared to the process requirements. They have then to be controlled and taken into account. On the other hand, in advanced nodes the lateral size of epi is small (10-30nm) and the thickness often very similar, epi have to be considered as 3D objects. In such a case, the crystal orientation plays a crucial role, and the epi morphology is very sensitive to small thermal budgets. In 3D objects, the epi lateral overgrowth (ELO) is significant and depends on the orientation via the faceting mechanism. As an illustration, Fig.2 displays active areas before epi and after epi in patterns orientated along <110> and <100>. It appears that <100> ELO is much more extended than the <110> one. This has some important consequences we have recently reported (2-3). Firstly, it changes the epi morphology. Secondly, it leads to new effects such as “anisotropic loading effect” and “time-non-linear deposition kinetics”. Furthermore, the small size of epi makes the surface energy dominant (larger than the giga-pascal stress of SiGe) and induces a rapid thermal rounding and even morphological instabilities more intense and rapid that Stranski-Krastanov ones. Both phenomena are illustrated in Fig.3. In the final part and as a conclusion, we will summary and discuss the recent and future trends in epi; and among the new processes, cyclic deposition illustrated in Fig.4 appears as the most promising. Figure 1
Read full abstract