[1B01-05] Sputtering and Plasma Processes 

3:00 PM - 3:30 PM JST(6:00 AM - 6:30 AM UTC)

[1B03【招待講演】] Decoupling sputter-deposition parameters using interpretable machine learning: a case study on transparent conducting film

*Naoomi Yamada¹, Oga Hayashi¹ (1. Department Applied Chemistry, Chubu University (Japan))

An IML framework was established to transition from empirical "black-box" optimization to a causal, mechanistic understanding of functional thin-film growth. Sputter deposition involves highly entangled thermodynamic and kinetic effects, making the process-property relationships complex to elucidate. In this study, an optimized tree-based XGBoost regression model was developed using a small experimental dataset of functional oxide thin films, treating substrate temperature, deposition pressure, and oxygen fraction as independent features. Without explicit domain instruction, the IML model autonomously rediscovered key physical laws, identifying that carrier density is governed by the thermodynamic contour of oxygen partial pressure, while electron mobility is driven by structural crystallinity. These data-driven hypotheses were experimentally validated, revealing a defect-compensation mechanism and a critical kinetic sweet spot. This synergy between data-driven inference and domain knowledge provides a transparent, scalable strategy broadly applicable to modern vacuum and materials science.

Password required to view