Seminar Series - Computational Imaging, data analysis, and machine learning in scanning / transmission electron microscopy (S/TEM)

Tuesday, November 10, at 2 p.m. via Zoom

Speaker: Dr. Colin Ophus, National Center for Electron Microscopy, Lawrence Berkeley National Laboratory

Title: Computational Imaging, data analysis, and machine learning in scanning / transmission electron microscopy (S/TEM)

Abstract:  Transmission electron microscopy (TEM) is a very powerful tool for both biological and materials science, especially since the introduction of high-speed direct electron detectors. The unmatched spatial resolution of TEM combined with the high quantum efficiency, fast frame rate, and large fields of view possible with direct detectors has enabled many new experiments to characterize complex nano-scale samples. However, these experiments often produce extremely large datasets with very low signal-to-noise ratios and nonlinear information transfer, necessitating the use of advanced computational methods to extract the desired sample properties. In this talk, I will show several examples of computational methods in TEM experiments. These include simple methods such as aligning time series to correct for sample drift, measurements of basic sample properties, and 3D imaging using tomographic reconstruction. I will also show examples of complex algorithms such as those used for phase contrast inversion under multiple scattering conditions, and the scattering-matrix formalism for simulating the quantum mechanical scattering of electron waves.  Finally, I will also show a machine learning example for TEM datasets, which outperforms conventional image processing methods by a wide margin.