LAUSR

Abstract In the field of optical metrology, the selection of the best model to fit experimental data is absolutely nontrivial problem. In practice, this is a very subjective and formidable task which highly depends on metrology expert opinion. In this paper, we propose a systematic approach to model selection in ellipsometric data analysis. We apply two well-established statistical methods for model selection, namely, the Akaike (AIC) and Bayesian (BIC) Information Criteria, to compare different dispersion models with various complexities and objectively determine the “best” one from a set of candidate models. The information criteria suggest the most optimal way to quantify the balance between goodness of fit and model complexity. In combination with screening-type parametric sensitivity analysis based on so-called “elementary effects” (the Morris method) this approach allows to compare and rate various models, identify key model parameters and significantly enhance process of ellipsometric measurements evaluation.