Dec 4, 2024
8:45am - 9:00am
Hynes, Level 2, Room 209
Richard Liu1,Utkarsh Pratius1,Roger Proksch1,2,Jason Bemis2,Sergei Kalinin1,3
The University of Tennessee, Knoxville1,Oxford Instruments Asylum Research2,Pacific Northwest National Laboratory3
Richard Liu1,Utkarsh Pratius1,Roger Proksch1,2,Jason Bemis2,Sergei Kalinin1,3
The University of Tennessee, Knoxville1,Oxford Instruments Asylum Research2,Pacific Northwest National Laboratory3
The rapid development of computation power and machine learning algorithms has paved the way for accelerating scientific discovery with automated scanning probe microscopy (SPM). Here we introduce a fully automated SPM powered by reward-driven machine learning algorithms. We start with the implementation of an interface library enabling Python-based control of the instrument from both local computers and remote clusters. We will also show how to abstract routine SPM operations into automated workflows. Next, we present the reward-driven automation of advanced SPM operations such as tapping mode topography mode, hysteresis loop spectroscopy and Kelvin Probe force microscopy mode. These automations are the foundations of fully automated SPM and rule out the operator biases in the SPM research. At a higher level, we show how to automate material discovery workflows with user-defined rewards and policy. Finally, we present example workflows including automated material discovery based on structural features, automated exploration based on spectral features (Deep Kernel Learning), and high-throughput automated discovery of combinatorial libraries. The fully automated SPM can promote collaboration in the SPM community by offering standardized results and allows providing fast feedbacks to material synthesis and theory by high-throughput material characterization.