Presentation Information

[7p-S202-3]Reliable optimization of MOCVD condition of GaN by objective function considering conventional conditions

〇Shota Seki1,2, Fumiaki Mizuno1, Tsuyoshi Matsuoka1, Tsutomu Sonoda3, Tokio Takahashi3, Hisashi Yamada3, Reiko Azumi3, Kentaro Kutsukake2, Toru Ujihara2 (1.Aixtal, 2.Nagoya Univ., 3.AIST)

Keywords:

semiconductor,machine learning,optimization

When optimizing manufacturing conditions using machine learning models, it is effective to add an indicator of proximity to conventional conditions to objective functions to avoid significant changes from conventional conditions. Furthermore, it is thought to be effective in improving the reliability of predictions in optimal solutions. In this study, we applied this to the optimization of the GaN vapor phase growth process and obtained reliable optimal solutions that closely matched the experimental results.