LAUSR

The advent of high-speed, high dynamic range, and low-noise pixelated detectors ushers in a new regime for scanning transmission electron microscopy (STEM) [1]. The efficient acquisition of a full diffraction pattern at each point in a scan enables analyses across a vast range of length scales. The information-rich 4D datasets are well suited to factor analysis, and machine learning techniques have proven invaluable for feature extraction in multidimensional datasets [2]. Efficient and physically transparent analyses remain challenging. Here we discuss a physically-motivated basis for efficiently representing scanning diffraction data and present interpretable results of component analysis.