Seminar Date: Tuesday, February 11, 2025
Time: 11:00 AM PT
Location: 67-3111 & Zoom
Talk Title: Efficient Electron Microscopy: Harnessing Automation for High-Throughput Materials Characterization
Zoom link
Abstract:
Scanning transmission electron microscopy (STEM) is a powerful tool for characterizing material structure and composition at atomic scales, capable of simultaneously acquiring real-space, diffraction-space and spectroscopic information. However, STEM is typically a low-throughput technique and insufficient to meet the growing demand for characterization generated by recent advances in data-driven materials discovery and automated synthesis. One reason for this is that, despite its capabilities and widespread use, several important aspects of STEM operation, including lens alignment and data acquisition, remain mostly manual processes, which greatly reduces the efficiency of microscope sessions. With the processing power of modern desktop computers and availability of open-source computer vision, image analysis and machine learning software, however, many of these key processes can now be automated. Complex workflows can then be built around these processes and tailored to individual samples and experiments for high-throughput data acquisition, thereby accelerating the discovery and characterization of new materials. Incorporating high-performance computing (HPC) into these workflows offers new opportunities for real-time analysis and HPC in-the-loop decision-making.
My presentation will cover three topics. First, the use of Bayesian optimization for automated on-sample election lens aberration correction, the development of a custom-built automation program for high-throughput acquisition of atomic-resolution HAADF- and 4D-STEM data, and a new system for streaming the large datasets generated by 4D-STEM (~100 GBs per dataset) directly to HPC for near real-time analysis. I will discuss the main methods underpinning these topics, including Bayesian optimization, computer vision tools and the streaming architecture, that I believe would prove useful in a wider variety of scientific applications.
