IJAM, Volume 28, No. 1 (2015)

Abstract. In the current work, we consider the inverse ellipsometric problem for thin film characterization which consists in determining the shape of a diffracting feature from an experimental ellipsometric data. The reformulation of the given nonlinear identification problem was considered as a parametric optimization problem using the Least Square objective function. The evaluation of the latter is often expensive as it implies the solution of the direct problem for each iteration. In this work, we propose a design procedure for global robust optimization using a probabilistic approach based upon Kriging method. Robustness is determined by the Kriging model to reduce the number of real functional calculations of Least Square objective function. The technical of the global optimization methods is adopted to determine the global robust optimum of a objective function considered.