Skip to content
KOEN
Technical note

Ultrathin InP and Ga₂O₃ layers on an RBSC platform for photonic and power semiconductors

CoolsPublished

https://cools.co.kr/en/insights/in-2026-019/

About this document

InP and Ga₂O₃ share two constraints: low thermal conductivity and costly single-crystal substrates. This Technical Note describes a common manufacturing platform that transfers only the functional layer onto a reaction-bonded silicon carbide (RBSC) support and joins it by interface-selective optical heating. The full report is available in Korean. An English summary is provided on this page.

Download

An English edition is not available yet. The Korean document is below.

Summary

InP and Ga₂O₃ have low thermal conductivity and expensive single-crystal substrates, yet only the top few micrometers do any work in the device. The functional layer is split off by layer transfer and bonded to RBSC, while the parent wafer is repolished and reused. Light in the 1500 nm band passes through both materials and generates heat only in an interfacial absorber, and residual silicon in the RBSC melts locally to form the bond. The calculations behind the thermal-resistance and cost effects are included.

Conclusion

The bottlenecks of the two materials reduce to the same structure. InP (about 68 W/m·K) and Ga₂O₃ (about 11-27 W/m·K) conduct heat poorly, so the substrate directly beneath the device becomes a thermal barrier rather than a heat path, and the costly single-crystal substrate caps both cost and output. Yet the layers that determine device performance are only a few micrometers thick, less than 1% of the wafer. A single change, keeping the functional layer and replacing the rest with a highly conductive, low-cost support, addresses heat, cost and handling together. The platform has three elements: net-shape RBSC substrates, layer transfer, and bonding by interface-selective optical heating. RBSC can be tuned to a thermal conductivity of 120-200 W/m·K and a thermal expansion coefficient of 4-5 ppm/K through its residual silicon content and porosity, and since it barely shrinks during sintering it can be made as large plates. Layer transfer, by ion-implantation-induced or stress-induced splitting, removes only the thickness the device needs and leaves the rest of the wafer intact. The absorption edge of InP lies in the 900 nm range and β-Ga₂O₃ absorbs only in the ultraviolet, so light in the 1500 nm band passes through both. This is why one light source and one optical system can process two different semiconductors. The optical absorber is a heat source, not an adhesive. The actual bonding medium is the residual silicon phase at the RBSC surface, which melts, wets and solidifies locally to join the semiconductor layer to the ceramic. Thermal diffusion lengths calculated from literature properties for a 1 μs exposure are about 4.5 μm in Si, 6.8 μm in InP and 2.1 μm in Ga₂O₃. With exposure kept at or below a microsecond, the heat-affected region stays within a few micrometers of the interface. Two integration routes are available, epi-first, which uses existing epitaxial wafers, and bonding-first, which forms the active region after bonding, and both share the same layer-transfer line and bonding optics. In a one-dimensional conduction model for a 1 mm² heat source, excluding lateral spreading, thermal resistance beneath the device improves 1.9 times for the InP structure and 10.6 times for the Ga₂O₃ structure (calculated values). The gain is larger for Ga₂O₃ because nearly all of its original thermal resistance comes from the substrate itself. The number of times a parent wafer can be reused, N, follows from thickness arithmetic: a 350 μm wafer losing 3 μm per transfer plus 12 μm per repolish, with 300 μm usable, gives N of about 20 (arithmetic value). Cost competitiveness therefore comes down to a single variable, the precision of the reclaim process.

Share this insight

Topics

  • Power semiconductors
  • Photonics