Pattern recognition in high-resolution electron microscopy of complex materials

Abstract

Structural features like defects or heterointerfaces in crystals or amorphous phases give rise to different local patterns in high-resolution electron micrographs or object wave functions. Pattern recognition techniques can be used to identify these typical patterns that constitute the image itself, as was already demonstrated for compositional changes in isostructural heterostructures, where the patterns within unit cells of the lattice were analyzed. To extend such analyses to more complex materials, we examined patterns in small circular areas centered on intensity maxima of the image. Nonsupervised clustering, namely, Ward's clustering method, was applied to these patterns. In two examples, a highly defective ZnMnTe layer on GaAs and a tunnel magneto resistance device, we demonstrate how typical patterns are identified by this method and how these results can be used for a further investigation of the microstructural properties of the sample.