Gasb-to-Si Direct Wafer Bonding and Thermal Budget Considerations for Photonic Applications

Abstract

Extensive progress has been made in recent years in heterogeneous integration (HI) of III-V materials on Si photonics wafers [1], including direct chip-to-chip bonding of GaSb-based infrared (IR) lasers [2], greatly expanding the application space and performance of either material system on its own. This includes active research at Sandia National Labs developing lasers based on GaAs, GaSb, GaN, and InP substrates directly integrated with silicon photonics at the chip and wafer scale. A “device last” process flow is desirable in this case both for scalability and because of the tight alignment tolerance required between the Si waveguides and active III-V laser elements to optimally couple light between the III-V and the Si. In this process flow, the Si photonics wafer is fabricated separately before bonding with an unpatterned III-V wafer or chip. The III-V laser is then defined after bonding, allowing lithographic registration to Si photonics structures. Thus, preparation of both semiconductor surfaces and preservation of the bond between them become critical for device performance and constrain subsequent process steps. Here, we focus on GaSb-based devices.The first consideration is preparing the GaSb and Si surfaces for bonding. Argon sputtering to prepare the GaSb surface for bonding to Si has been investigated, though the emphasis was on GaSb surface roughness, bonding temperature, and electronic transport through the interface [3]. Here, we use x-ray photoelectron spectroscopy (XPS) to investigate the impact of the GaSb oxide composition on the quality of the bond to oxidized Si wafers. In Fig. 1(a)-(c) we use confocal scanning acoustic microscopy (CSAM) to compare as-received 3” GaSb substrates with and without chemical oxide removal to epitaxial GaSb. In the CSAM images, dark areas indicate bond adhesion. We find that bonding improves with decreasing Ga-oxide content realized on the epitaxial GaSb.The second consideration is constraining the laser process steps in order to preserve the bond between the Si photonics and GaSb wafers. Our HI process steps involving GaSb thinning and substrate removal have been described elsewhere [4], though investigation of the thermal budget for survival of the direct GaSb/Si bond through a laser fabrication process flow is lacking in the literature. As the thermal expansion coefficient of GaSb is approximately three times that of Si, temperatures must be limited to prevent the GaSb from expanding excessively and delaminating from the Si wafer when running the III-V laser fabrication after integration with the Si photonics wafer. We investigate the robustness of the bond between epitaxial GaSb and the Si photonics wafer through cross-sectional transmission electron microscopy (TEM) of the bond interface after various process steps typical to laser fabrication and the implications of the thermal budget on these process steps. For example, we find that extended periods of time (more than an hour) at 250°C, which is typical for chemical vapor deposition (CVD) of SiO2 or Si3N4, cause the thinned III-V wafer to delaminate from the Si photonics wafer. While we don’t observe destructive delamination at lower processing temperatures, high-angle annular dark-field imaging (HAADF) analysis in Fig. 1(d) shows significant Sb diffusion into the Si photonics native SiO2 layer.In summary, for optimal performance of GaSb-based hybrid lasers directly bonded to Si photonics, understanding and preserving the bond between these materials is critical. These considerations include minimizing the Ga-oxide at the GaSb surface prior to bonding to promote bond success and keeping extended thermal processes below 250°C to maintain the bond between the GaSb and the Si photonics wafers. These results will inform processes required for full laser fabrication flow, particularly in steps that traditionally require higher or longer temperature parameters.SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.[1] D. Caimi, et al., IEEE Trans. Elect. Dev. 68, p. 3149 (2021).[2] A. Spott, et al., Optica 5, p. 996 (2018).[3] F. Predan, et al., Appl. Surf. Scie., 353, p. 1203 (2015).[4] M. Wood, et al., in CLEO 2023 Technical Digest Series, STh3H.5 (2023). Figure 1