[18a-S2_204-1~6] 23.1 Joint Session N "Informatics"

12:45 PM - 1:00 PM JST(3:45 AM - 4:00 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation Award

[18a-S2_204-6] Optimization of Process-Condition Sequences Using a Latent Space for Manufacturing Processes with Time-Evolving Systems

〇(M2)Takanao Sakamoto¹, Masaki Takaishi^2,3, Kentaro Kutsukake^1,2, Shunta Harada^1,2, Toru Ujihara^1,2 (1.Grad. Sch. Eng., Nagoya Univ., 2.IMaSS Nagoya Univ., 3.Aixtal)

Keywords:

Machine Learning,Optimization,Semiconductor

In manufacturing processes such as crystal growth, optimal process conditions change over time in response to the temporal evolution of interface geometry and physical fields. In this study, we propose a unified sequence optimization framework in which the time evolution of physical fields is predicted using a single model shared across all time steps, and the degree of variation in the physical fields is evaluated in a latent space constructed by a variational autoencoder (VAE). When applied to SiC crystal growth by the solution growth method, the proposed approach successfully identified an optimal condition sequence that promotes crystal growth while maintaining physical fields close to the initial state in the latent space.