[18a-PA1-1~23] 23.1 Joint Session N "Informatics"

[18a-PA1-17] Neural Network-Based Molecular Dynamics Simulations of Silica Etching Processes

〇Yuta Yoshimoto¹, Meguru Yamazaki¹, Naoki Matsumura¹, Yuto Iwasaki¹, Kazutaka Nishiguchi¹, Yasufumi Sakai¹, Hideyuki Jippo¹ (1.Fujitsu)

Keywords:

molecular dynamics,neural network potential,etching

We develop a high-fidelity neural network potential (NNP) for the Si-O-H-F system to elucidate the atomic-level reaction mechanisms governing the dry and wet etching of silica. By leveraging knowledge distillation from a large-scale pre-trained model, we efficiently construct a lightweight yet accurate NNP, significantly reducing the required volume of first-principles data. This NNP is subsequently employed in molecular dynamics simulations to analyze the dry etching process by hydrogen fluoride and the wet etching process by hydrofluoric acid.