Towards Automating Structural Discovery in Scanning Transmission Electron Microscopy

Scanning transmission electron microscopy (STEM) is now the primary tool for exploring functional materials on atomic level. Often, features of interest are highly localized in specific regions material, such as ferroelectric domain walls, extended defects, or second phase inclusions. Selecting to image structural and chemical discovery via atomically resolved imaging has traditionally proceeded human operators making semi-informed judgements sampling locations parameters. Recent efforts at automation physical have pointed towards use "active learning" methods that utilize Bayesian optimization with surrogate models quickly find relevant interest. Yet despite potential importance this direction, there a general lack certainty selecting control algorithms how balance priori knowledge material system derived during experimentation. Here we address gap by developing automated experiment workflows several combinations both illustrate effects these choices demonstrate tradeoffs associated each terms accuracy, robustness, susceptibility hyperparameters discovery. We discuss possible build descriptors using raw data deep learning based semantic segmentation, well implementation variational autoencoder representation. Furthermore, workflow applied range feature sizes including NiO pillars within La:SrMnO$_3$ matrix, domains BiFeO$_3$, topological defects graphene. The code developed manuscript open sourced will be released github.com/creangnc/AE_Workflows.