Crystal Phase Mapping by Scanning Precession Electron Diffraction and Machine Learning Decomposition

Abstract

Analyzing the distribution of crystal phases is required to understand nanostructures. Conventional analysis in the transmission electron microscope (TEM) include acquiring dark-field (DF) images for each expected crystal phases. However, this becomes laborious if a series of phases are present, a nm-scale resolution is required and larger areas (> 0.5 µm2 ) are to be analyzed. The latter is even worse for highresolution (HR) TEM. An alternative approach is scanning precession electron diffraction (SPED), allowing for automatic determination of the crystal phase distribution. Here we combine SPED with machine learning to a model system of a ~400 nm-thick GaAs nanowire (NW) with multiple axial GaAsSb-based superlattices [1].