Abstract

Generally, spectroscopic ellipsometry data are compared with a model of the sample surface by using a nonlinear least-squares fitting procedure to determine characteristics of the surface, which include thin-film thickness and fractions of individual constituents in the layers. Usually, an unbiased estimator (i.e., without any spectroscopic weighting) is minimized to determine the parameters of the fit. In this paper, the use of the biased estimator in the fitting of spectroscopic ellipsometry data is examined and is applied to data from two-channel polarization modulation ellipsometry experiments. By including the error in the estimator, the more accurate regions of the spectrum are weighted more, and the ellipsometric data can be expressed in the most appropriate way (i.e., as complex P, W and A, complex e) without affecting the results of the fit. In addition, a goodness-of-fit can be calculated, which indicates whether a more complicated model is necessary or whether competing models are statistically different. Several thin films of SiC>2 on Si will be discussed as examples.

© 1990 Optical Society of America