[1Ep01-05] Development of Data Processing and Data Utilization in Surface Analysis: Surface Analysis Division's Session

[1Ep04] Introduction to sample structure prediction from XPS data using Bayesian estimation and SESSA Simulator

*Hiroshi Shinotsuka¹, Kenji Nagata¹, Malinda Siriwardana¹, Hideki Yoshikawa¹, Hayaru Shouno², Masato Okada³ (1. National Institute for Materials Science, 2. The University of Electro-Communications, 3. The University of Tokyo)

We have developed a framework for solving the inverse problem of XPS by incorporating the SESSA, into Bayesian estimation to obtain an overall picture of the distribution of plausible sample structures from the measured XPS data. In this framework, the procedure for running the simulator, which originally required human trial and error, was fully automated. As a concrete example, an artificial sample with a four-layered structure of C₂O (10 Å)/HfO₂ (25 Å)/SiON (16 Å)/Si was designed, and the angle resolved XPS intensity data were generated. We performed an inverse-problem solution in our framework to obtain Bayesian posterior distributions of the composition and thickness of each layer of the pre-designed sample, and we succeeded in estimating the sample structure, including the true model.