Automatic and Interactive Correlation Partitioning Compared: Application to TiN/Ti/SiO2

A new method has been developed for the partitioning of sets of images with the objective of automatically identifying the number and locations of different regions in a material, The method is called automatic correlation partitioning and it involves the identification of clusters in the n-dimensional intensity histogram of a set of n images that are spatially registered, The method uses the peaks located in the simple intensity histograms of each image in the set to produce a list of all possible clusters in the entire data set, This list is then searched in order to find the actual clusters, The method is tested using data from a multi-imaging Auger electron microscope, which yields sets of Auger images characteristic of the spatial distributions of selected kinds of atoms in the surface of a solid, The first tests involve the use of a model sample consisting of a W overlay pattern on a Si substrate, The second tests are done on a TiN/Ti/SiO2 planar layer structure that has been ion beam bevelled to reveal a cross-section of the composition depth profile, The first set contains two images and the second set contains five images, The results of the new automatic method are compared with those obtained by the analyst working interactively with the data set to identify the clusters subjectively, Cluster analysis of the second sample reveals details of the interfacial layer chemistry not revealed by the interactive method and is consistent with published XPS depth profiling experiments reporting a titanium silicide layer at the Ti/SiO2 interface.