Figure 2 shows the diamond substrate bonded on the surface of the InP substrate. The bonding interface can be observed through the transparent diamond substrate. Diffused reflection due to the gaps between the substrates was observed where the surfaces were not bonded. While there were some bright spots, Fig. 2 indicates that three-quarters of the contacted area was successfully bonded. Voids with diameters of approximately 0.1 mm were formed due to particles on the substrate surface. The large unbonded regions at the corners of diamond substrates resulted from the convex diamond surface (see the supplement of 30). If the environmental cleanliness and substrate flatness are improved, direct bonding will be formed at most of the contacted area. When a shear force of 9.3 MPa (84 N for 3 × 3 mm) was applied to the bonded diamond substrate, fracture at the bonding interface and cleavage along the InP (110) face were observed.
Surfaces are required to be sufficiently smooth for direct bonding; the root mean square (RMS) roughness is preferably less than ~5 Å 39. The diamond substrate used in this study had an atomically smooth surface with an RMS roughness of less than 3 Å, which was reported in our previous study 33. The InP substrate surface was investigated using an atomic force microscope (AFM), as shown in Fig. 3. The RMS roughness of the InP substrate surface was initially 2.76 ± 0.3 Å. Thereafter, the surface became rough by the oxygen plasma irradiation as the RMS roughness increased to 3.03 ± 0.3 Å. However, it was still sufficiently smooth for bonding formation.
The surface chemical composition of the InP substrate was investigated through angle-resolved X-ray photoelectron spectroscopy (XPS) , as depicted in Fig. 4. The measurement depth depended on the take-off angle of the photoelectrons; the inelastic mean free path (IMFP) was calculated at approximately 1 and 4 nm for angles of 10.75° and 63.25°, respectively. Before plasma irradiation, the amounts of In-O and P-O bonds were relatively small, and organic contaminants were present on the surface. This indicated that the OH groups detected at the surface probably resulted from C-OH bonds, owing to contaminants. However, organic contaminants rarely existed, and In-O and P-O bonds were present on the plasma-activated InP surface. Thus, the OH bonds detected on the surface were possibly attributed to the In-OH, P-OH, or both groups generated on the InP substrate. Our previous study suggested that the diamond substrate cleaned with the NH3/H2O2 mixture was terminated with the C-OH groups33. Consequently, the OH groups on the InP and diamond substrates probably reacted with each other during the bonding process.
The nanostructure of the InP/diamond bonding interface was observed using an transmission electron microscope (TEM), as shown in Fig. 5. For the observation, the thickness of the InP substrate, bonded with diamond, was reduced to 10 µm by grinding. Subsequently, the ultra-thin TEM specimen was prepared using a focused ion beam (FIB). The incident angle of the electron beam was set parallel to the InP <110> direction. As shown in Fig. 1, the InP and diamond substrates formed atomic bonds without cracks or nanovoids. Moreover, an amorphous layer with a thickness of approximately 3 nm was observed at the bonding interface. The presence of the intermediate layer corresponds to the previous studies on InP/Si direct bonding; the layer is composed of In, P, and O due to oxygen plasma irradiation 38. It was assumed that the thermal conductivity of the intermediate layer was low. However, it was supposed that the negative effect of heat dissipation was limited because the layer was atomically thin.