Estimation of effective reaction radius for catalytic chain reaction of chemically amplified resist by Bayesian optimization

Abstract

The modeling of electron beam (EB) lithography processes from exposure to development is important for resist pattern prediction and material design. The effective reaction radius for deprotection indicates the capability of chemically amplified resists. However, this parameter cannot be directly measured by experiments. On the other hand, the concentration of protected units determines the resist pattern after development. To simulate the lithography processes, these two parameters must be estimated. In this study, we developed a strategy to determine these two parameters at the same time by Bayesian optimization to reduce the computational time. The simulation results were compared with the scanning electron microscopy images of resist patterns obtained by EB lithography. As a typical Bayesian inference method, Gaussian process regression with the Matérn kernel was applied to the analysis, which reduced the iterative calculation from 140 to 35. The probable effective reaction radius was found.