NTT's AI film platform creates reusable rules; world-first β-Ga₂O₃ sputtering

NTT's demonstration addresses a known bottleneck in materials development: process conditions like temperature and gas flow affect film quality in complex ways, so finding the right settings is slow. Traditional AI accelerated the search but hid its reasoning. By pairing Bayesian optimization with automated sputtering and optical evaluation, NTT closed the loop without human intervention. Then, using random forest analysis on the gathered data, it extracted which parameters matter and how they interact.

The reported threefold speedup in the experiment cycle and the world-first single-crystal β-Ga₂O₃ film by sputtering are concrete results from this pipeline. β-Ga₂O₃ is positioned as a next-generation power semiconductor material, so the process advance may support its practical use. The company frames this as groundwork for AI-driven laboratories that turn autonomous experiments into scientific knowledge.

Looking ahead, the value of this approach will likely depend on whether the extracted rules transfer from one machine or material to another, which is the stated goal. NTT says it will pursue optimization with multiple evaluation metrics and AI models that incorporate material and process knowledge. Whether that leads to faster commercialization of high-performance power semiconductors remains to be seen, but the underlying logic is that interpretable rules accelerate development beyond what black-box optimization alone can achieve.