Robust solution to the inverse problem in optical scatterometry.

In optical scatterometry, the least squares (LSQ) function is usually used as the objective function to quantify the difference between the calculated and measured signatures, which is based on the belief that the actual measurement errors are normally distributed with zero mean. However, in practice the normal distribution assumption of measurement errors is oversimplified since these errors come from the superimposed effect of different error sources. Biased or inaccurate results may be induced when the traditional LSQ function based Gauss-Newton (GN) method is used in optical scatterometry. In this paper, we propose a robust method based on the principle of robust estimation to deal with the abnormal distributed errors. An additional robust regression procedure is used at the end of each iteration of the GN method to obtain the more accurate parameter departure vector. Simulations and experiments have demonstrated the feasibility of our proposed method.