[17a-S2_204-1~13] 23.1 Joint Session N "Informatics"

11:00 AM - 11:15 AM JST(2:00 AM - 2:15 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation Award

[17a-S2_204-8] Virtual Screening of High-κ Oxides via Physics-Based Machine Learning

〇(M2)Atsushi Takigawa^1,2, Shin Kiyohara², Yu Kumagai^2,3 (1.Tohoku Univ., 2.IMR, Tohoku Univ., 3.OAS, Tohoku Univ.)

Keywords:

dielectrics,machine learning

This study proposes a physics-based machine-learning framework for efficient screening of high-k oxides. Born effective charges and phonon properties are predicted separately and directly substituted into first-principles formulas via a joint model to accurately estimate ionic dielectric constants. Screening of 8,717 oxides identified 38 dynamically stable high-performance candidates, demonstrating superior accuracy, physical consistency, and extrapolation robustness over conventional approaches.