Yongtao Liu1
1Oak Ridge National Laboratory
Microscopy has significantly advanced our understanding of structure-function relationships at the nanoscale, becoming a staple in characterization laboratories. However, traditional microscopy methods have largely been constrained by manual operations centered around human intervention. Therefore, we present the integration of application program interface (API) with machine learning to address these limitations. We developed AEcroscoPy, a cross-platform Python API designed to automate microscopy experiments, and showcase the combined power of human expertise, ML efficiency, and API-driven automation for accelerating scientific discovery. Our development of automated and autonomous experiment (AE) in scanning probe microscopy (SPM) facilitates the exploration of material functionalities and mechanisms. Using AE-SPM, we discovered coexistence and interplay of two ferroelectric subsystems in wurtzite ferroelectric thin films. Employing ML-driven approaches, we have probed ferroelectric materials to study phenomena such as domain wall dynamics and switching mechanisms, as well as the interactions between domain structures and local properties. By incorporating physical hypotheses in active learning model, our approach has enabled the microscope to autonomously discover the physical laws influencing domain switching. Although these methodologies were applied to specific materials, they possess broad potential to revolutionize various characterization techniques, including the assessment of stiffness and adhesion through force-distance curves in SPM.