Fast 4D-STEM-based phase mapping for amorphous and mixed materials

Abstract

All materials are made from atoms arranged either in repeating (crystalline) or in random (amorphous) structures. Diffraction measurements probe average distances between atoms and/or planes of atoms. A transmission electron microscope in scanning mode (STEM) can collect spatially resolved 2-dimensional diffraction data, effectively creating a 4-dimensional (4D) hyperspectral dataset (4D-STEM). Interpretation strategies for such 4D data are well-developed for crystalline materials, because their diffraction spectra show intense peaks, allowing for effective phase and crystal orientation mapping at the nanoscale. Yet, because of the continuous nature of the diffraction data for amorphous and mixed materials, it is challenging to separate different amorphous contributions. Nonnegative matrix factorization (NMF) allows separation of 4D-STEM data into components with interpretable diffraction signatures and intensity maps, independent of the structure. However, NMF is a non-convex optimization problem and scales ~ O(nmk) with n the number of positions probed, m the number of diffraction features and k the number of components, making analysis of large 4D datasets inaccessible. Here, we apply QB decomposition as a preprocessing step for NMF (Randomized NMF or RNMF) to achieve scaling independent of the largest data dimension (~O(nk)), opening the door for NMF analysis of 4D-STEM data. We demonstrate our approach by mapping a thin TiO$_2$ layer on top of SiO$_2$, and a LiNi$_{0.6}$Co$_{0.2}$Mn$_{0.2}$O$_{2}$ (NMC) - Li$_{10}$GeP$_2$S$_{12}$ (LGPS) mixed crystalline-amorphous battery interface, illustrating strengths and limitations of using RNMF for structure-independent phase mapping in 4D-STEM experiments.