Abstract
In thin film photovoltaic devices, the control of grain structure and local crystallography are fundamental for high power conversion efficiency and reliable long-term operation. Structural defects, grain boundaries, and unwanted phases can stem from compositional inhomogeneities or from specific synthesis parameters, and they need to be thoroughly understood and carefully engineered. However, comprehensive studies of the crystallographic properties of complex systems, including different phases and/or a large number of grains, are often prohibitively challenging. Here, the use of 4D Scanning Transmission Electron Microscopy (4D-STEM) is demonstrated on cross-sections to unravel the nanoscale properties of three different materials for photovoltaics: Cu(In,Ga)S(2), halide perovskite, and Sb(2)Se(3). These materials are chosen because of the variety of challenges they present: the presence of multiple phases and complex stoichiometry, electron beam sensitivity, and very high density of grains. 4D-STEM provides comprehensive insights into crystallinity and microstructure, but navigating its large datasets and extracting actionable, statistically sound information requires advanced algorithms. How unsupervised machine learning, including dimensionality reduction and hierarchical clustering, can extract key information from 4D-STEM datasets is demonstrated. The analytical framework follows FAIR principles, employing open-source software and enabling data sharing.