[17p-M_103-1~12] 8.2 Plasma deposition of thin film, plasma etching and surface treatment

3:15 PM - 3:30 PM JST(6:15 AM - 6:30 AM UTC)

[17p-M_103-6] Investigation of Feature Engineering Methods for Improving Machine Learning Accuracy Based on Plasma and Material Information Science (PaMIS)

〇KUNIHIRO KAMATAKI¹, W. Sukma Fitriani², Yushi Sato¹, Yosei Kurosaki¹, Tsukasa Masamoto¹, Daisuke Yamashita¹, Takamasa Okumura¹, Naho Itagaki¹, Kazunori Koga¹, Masaharu Shiratani¹ (1.Kyushu Univ. ISEE, 2.Kyushu Univ. IMI.)

Keywords:

Plasma and Material Information Science (PaMIS),Feature Engineering for Machine Learning,Plasma Enhanced Chemical Vapor Deposition (PECVD)

Plasma deposition processes are inherently non-equilibrium and nonlinear, and machine-learning (ML) models using raw experimental variables often suffer from limited accuracy and generalizability. In this study, we propose a feature engineering approach based on Plasma and Material Information Science (PaMIS). Physically interpretable plasma descriptors are derived from optical emission spectra in a TEOS-PECVD process and incorporated into an ML model, enabling accurate prediction of the deposition rate.