Pycroscopy and AECroscopy—Reproducible and Open-Source Workflows for Automated Microscopy Experiments

Microscopy workflows are becoming incresingly complex, at times necessitating coordination between simulations, experiments, and computational resources. When attempting to automate microscopy workflows, most application programming interfaces (APIs) available from vendors are sui generis, creating yet more barriers for users who need to couple multiple instruments together for true autonomous workflows. Additionally, with increasing automation, maintaining a record of every step conducted becomes a challenge, and ensuring metadata is stored for full reproducibility is a must.<br/><br/>To tackle these challenges, we developed a python-based package termed 'AEcroscopy', short for AutomatedExperiments for microscopy, that consists of a hardware and software component that can be utilized to automate many different microscopes, including scanning tunneling microscopes, scanning transmission electron microscopes, and atomic force microscopes. Users can call specific functions to quickly code up their experiments in python, without need to change code for different microscopes. Moreover, the software auotmatically logs every function call made and every parameter set, to ensure reproducibility. All datasets are stored in an open source dataset object termed a sidpy dataset, which are objects built on top of dask arrays. These objects contain information relevant to the dimensionality of each variable, automatically include all the metadata, and offer features such as easy visualization, parallelization, and can be written to HDF5 files. We show examples of reproducible workflows for different systems, including sparse scanning measurements, Bayesian optimization and reinforcement learning workflows for optimized materials manipulation and physics discovery.