Imaging ellipsometry : a multivariate analysis strategy based on k-means and hierarchical clustering for fast data inversion

In this paper, we consider two-steps statistical algorithms to identify the pixels of the image characterizing
regions of the sample having a similar optical response. Typically, 128 or 256 regions are selected
and the ellipsometric equations are numerically solved for each of them. The solution is back-propagated
on the initial images to get a thickness or a refractive index map. This procedure considerably decreases
the processing time of the data. Examples concerning SiO2 on Si samples, with (Fig. 1A and 1B) and
without patterns, will be given as well as simulations results.