Low-temperature direct bonding of InP and diamond substrates under atmospheric conditions

An InP substrate was directly bonded on a diamond heat spreader for efficient heat dissipation. The InP surface activated by oxygen plasma and the diamond surface cleaned with an NH3/H2O2 mixture were contacted under atmospheric conditions. Subsequently, the InP/diamond specimen was annealed at 250 °C to form direct bonding. The InP and diamond substrates formed atomic bonds with a shear strength of 9.3 MPa through an amorphous intermediate layer with a thickness of 3 nm. As advanced thermal management can be provided by typical surface cleaning processes followed by low-temperature annealing, the proposed bonding method would facilitate next-generation InP devices, such as transistors for high-frequency and high-power operations.

For efficient heat dissipation, semiconductor researchers have developed an integration technique for devices on a diamond heat spreader, which has the highest thermal conductivity amongst solid materials (2200 W/m/K). For example, Ksenia Nosaeva et al. transferred a diamond heat-spreading layer on the InP HBT embedded in benzocyclobutene (BCB) resin 21 . Andreas Beling et al. integrated InP photodiodes on a diamond sub-mount by flip-chip bonding using metal bonding layers 22,23 . Direct bonding of the device and diamond is ideal to mitigate thermal resistance because the thermal conductivities of these bonding materials are an order of magnitude smaller than that of diamond. In particular, there have been intensive studies regarding the direct and indirect bonding of Ga-based materials (i.e., GaN 24-26 , GaAs [27][28][29][30] , and InGaP 31 ) onto diamond substrates. However, studies on the direct bonding of InP and diamond substrates are scarce.
Our research group developed and reported a direct bonding method for semiconductor substrates (i.e., Si, Ga 2 O 3 ) on a diamond heat spreader [32][33][34][35][36] . We found that OH groups were formed on a diamond surface treated with oxidizing solutions, such as H 2 SO 4 /H 2 O 2 32 and NH 3 /H 2 O 2 mixtures. Moreover, the OH-terminated diamond surface forms direct bonding with the OH-terminated semiconductor substrate by thermal dehydration at approximately 200 °C. The semiconductor substrates are typically OH-terminated using plasma activation 37 . While studies on the bonding of InP and diamond are scarce, optoelectronics scientists have developed multifunctional devices using direct bonding techniues [38][39][40] and achieved the direct bonding of oxygen-plasmaactivated InP lasers and Si waveguides [41][42][43][44] . Consequently, the InP surface activated by the oxygen plasma can be directly bonded with the OH-terminated diamond surface. To test this hypothesis, we proposed direct bonding of InP and diamond substrates and investigated nanostructures of the InP/diamond bonding interface, as illustrated in Fig. 1 www.nature.com/scientificreports/ 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 34 ). If the environmental cleanliness and substrate flatness are improved, direct bonding will be formed at most of the contacted area. When a shear   Surfaces are required to be sufficiently smooth for direct bonding; the root mean square (RMS) roughness is preferably less than ~ 5 Å 45 . 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 36 . 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 roughness after the oxygen plasma irradiation was similar as the RMS roughness was 3.03 ± 0.3 Å; it was sufficiently smooth for bonding formation.

Results
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 NH 3 /H 2 O 2 mixture was terminated with the C-OH groups 36 . 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 a 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. 5, 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. Figure 6 depicted the compositional analysis obtained using energy-dispersive X-ray spectroscopy (EDX). The amorphous layer at the bonding interface is composed of In, P, O, and C. It is known that the InP/Si interface bonded using oxygen plasma is composed of In, P, and O 44 . The C atoms supposedly diffused into the oxide layer on the InP substrate formed by the oxygen plasma; the formation of the intermediate oxide layer is unavoidable in the case of the bonding under atmospheric conditions. It was assumed that the thermal conductivity of the intermediate layer was low but significantly thin compared with conventional approaches (e.g. 2-4-µm-thick metal layers 21,22 ). Thus, it was supposed that the InP/diamond bonding technique would contribute to efficient heat dissipation from the InP electronic devices.

Conclusions
In this study, we demonstrated the direct bonding of InP and diamond substrates to improve the heat dissipation of InP-based electronic devices. The InP substrate activated by oxygen plasma was contacted with the diamond substrate that was cleaned with a mixture of NH 3 , H 2 O 2 , and H 2 O under atmospheric conditions. Direct bonding was formed by annealing the contacted specimen at 250 °C. As both surfaces were atomically smooth after the pre-bonding treatments, the InP and diamond substrates successfully generated direct bonding with a shear

Method
In this study, commercially available InP and diamond substrates were directly bonded as received conditions, as illustrated in Fig. 1. Three-square-millimeter diamond (111) substrates with a thickness of 300 μm (from EDP Corp.) were bonded on a three-inch-diameter InP (100) wafer with a thickness of 500 μm (from Sumitomo Electric Industries, Ltd). The diamond substrates were cleaned with a mixture of 10 mL of NH 3 solution (28%), 10 mL of H 2 O 2 solution (35%), and 50 mL of deionized water at 75 °C for 10 min. The diamond substrates were rinsed in deionized water and blown by nitrogen gas for drying. The InP substrates were activated with reactive ion etching equipment (QAP-1000, Bondtech). The plasma at a power of 200 W irradiated the InP surface for 30 s under an O 2 pressure of 60 Pa and an O 2 mass flow rate of 20 mL/min. In the contacting step, the activated InP substrate was placed on www.nature.com/scientificreports/ a Peltier cooler at 14 °C for approximately 30 s in our clean room (temperature: 23 °C, relative humidity: 40%), and then the diamond substrate was placed on the InP substrate. The cooling process developed condensed water molecules that are believed to promote hydrogen bond networks between the InP and diamond substrates. The contacted specimen was annealed at 250 °C for 24 h under a load of approximately 1 MPa. The bonding quality was evaluated using a shear tester (4000Plus, Nordson DAGE). The surface roughness of the InP substrate was investigated using the AFM (L-trace, Hitachi). The surface chemical composition was studied using XPS (VG Theta Probe, Thermo Fisher Scientific). The nanostructure of the InP/diamond bonding interface was investigated using the TEM and EDX (JEM-ARM200F, JEOL).  www.nature.com/scientificreports/