[15p-P06-1~15] 23 Joint Session N "Informatics" (Poster)

[15p-P06-5] Data-Driven Method for Discovering Low Adsorption Energy Molecules on Si Surfaces

〇Hiromori Murashima¹, Koki Uonami¹, Shogo Kunieda¹, Yosuke Hanawa¹, Yuta Sasaki¹, Hitoshi Kamijima², Toshiaki Shintani², Ryo Yoshida³ (1.SCREEN Holdings Co., Ltd., 2.Research Institute of Systems Planning, Inc., 3.The Institute of Statistical Mathematics)

Keywords:

machine learning,materials informatics,semiconductor

In recent years, the miniaturization and three-dimensionalization of semiconductor devices have advanced, exacerbating the issue of micro-pattern collapse during the drying process in semiconductor cleaning. Pattern collapse can be suppressed by the sublimation drying method, which solidifies and sublimates the liquid between the patterns. One challenge of this method is organic residues, necessitating efficient methods for exploring low-residue materials from numerous candidates. A data-driven exploration method, combining sequential experimental design by Bayesian optimization (BO) and property calculations, has been proposed. This study aims to confirm the effectiveness of data-driven exploration and select candidates for low-residue sublimation agents. We conducted an automatic exploration of molecules with low adsorption energy on Si substrate surfaces using a surrogate model predicting adsorption energy values and property calculations using Matlantis.