Using Gaussian process regression for efficient parameter reconstruction

Abstract

Optical scatterometry is a method to measure the size and shape of periodic\nmicro- or nanostructures on surfaces. For this purpose the geometry parameters\nof the structures are obtained by reproducing experimental measurement results\nthrough numerical simulations. We compare the performance of Bayesian\noptimization to different local minimization algorithms for this numerical\noptimization problem. Bayesian optimization uses Gaussian-process regression to\nfind promising parameter values. We examine how pre-computed simulation results\ncan be used to train the Gaussian process and to accelerate the optimization.\n