[20p-A21-1~11] 23.1 Joint Session N "Informatics"

2:00 PM - 2:15 PM JST(5:00 AM - 5:15 AM UTC)

[20p-A21-3] Reduction of cumulative error in self-regression model for grinder sim.

〇Keichi Osada¹, Teruyuki Katsuoka¹, Riku Tanaka¹, Fumiya Kawate¹, Shota Seki¹, Sepasy Saeed², Yoshifumi Watanabe² (1.Aixtal, 2.Mipox Corporation)

Keywords:

semiconductor,machine learning,process informatics

The manufacturing process includes a process in which the shape of a sample is sequentially changed through multiple processing steps. At this time, the final shape can be calculated by repeatedly making predictions using the same machine learning model that has been trained with the same input and output formats at each step, but the problem is that errors accumulate due to repeated predictions. In this report, we report the results of a study of a U-Net-based autoregressive model that reduces accumulated errors using the grinding process of semiconductor substrates as an example.