Empirical models in semiconductor processing: optimization and assessment as simulators

摘要

Abstract: If any empirical model of an experimental system is to be used to make predictions its success as a simulator needs to be determined. This is especially so in semiconductor manufacturing where process runs are expensive making the need for a reliable process simulation even more important. With many current model assessment techniques, for example `coefficients of determination', too much information about the model's fit is hidden by the attempt to describe the model's success in terms of a single variable value. In this paper the description is given of a computational approach together with a standard visual display technique which allows the simulation capabilities of a model to be more fully understood. The method described is applicable to all modelling algorithms and as such allows the utility of competing modelling philosophies to be assessed.!10