Estimating cross-section semiconductor structure by comparing top-down SEM images

Abstract

Scanning electron microscope (SEM) images for semiconductor line-width measurements are generally acquired in a top-down configuration. As semiconductor dimensions continue to shrink, it has become increasingly important to characterize the cross-section, or sidewall, profiles. Cross-section imaging, however, requires the physical cleaving of the device, which is destructive and time-consuming. The goal of this work is to examine historical top-down and cross-section image pairs to determine if the cross-section profiles might be estimated by analyzing the corresponding top-down images. We present an empirical pattern recognition approach aimed at solving this problem. We compute feature vectors from sub-images of the top-down SEM images. Principal component analysis (PCA) and linear discriminant analysis (LDA) are used to reduce the dimensionality of the feature vectors, where class labels are assigned by clustering the cross-sections according to shape. Features are extracted from query top-downs and compared to the database. The estimated cross-section of the query is computed as a weighted combination of cross-sections corresponding to the nearest top-down neighbors. We report results obtained using 100nm, 180nm, and 250nm dense and isolated line data obtained by three different SEM tools.