Presentation Information
[7p-S202-9]Deep Learning Ellipsometry: Ultrafast and High-Accuracy Determination of Optical Constants, Thin-Film Structures, and Bandgaps
Ryosuke Oka1, Yuki Yamamoto1, James N. Hilfiker2, Hitoshi Tampo3, Takehiko Nagai3, Masahiro Hayashi1, 〇Hiroyuki Fujiwara1 (1.Gifu Univ., 2.J.A. Woollam Co., 3.AIST)
Keywords:
Deep learning,Ellipsometry,Bandgap
This study developed an innovative deep learning method to overcome challenges in optical analysis of thin-film materials using spectroscopic ellipsometry. By training on 100 million virtual datasets and incorporating noise-tolerant transfer learning, we successfully predict optical constants, film thickness, and bandgap within 2 seconds. The deep learning results show excellent agreement with conventional manual analysis, demonstrating both high precision and significantly faster processing.