CaDS Seminar 2024 - Sep. 10

Dr. Paul Baumeister (SDL Electrons and Neutrons)

Data Compression for Live Transmission Electron Microscopy

Abstract:

Scanning Transmission Electron Microscopy (STEM) has become a powerful imaging technique with resolutions enabling to spot single atoms. STEM devices produce vast data volumes while scanning the probe, so there a need for both, fast processing pipelines and data compression. We present an innovative technique to compress and post-process STEM images on-the-fly providing visual feedback to the scientist operating the microscope. An essential ingredient to this are harmonic function sets allowing to avoid several Fourier transforms in the post-processing pipeline completely. Furthermore, transformation of images into the representation in harmonic functions can be as efficient as matrix-matrix-multiplications on the GPU.


Short bio of the speaker: 

Paul Baumeister leads the SimDataLab Electrons & Neutrons at Forschungszentrum Jülich. He received a JSPS scholarship (Japan) in 2009 and his PhD in 2012 from RWTH Aachen University while working at the Peter-Grünberg-Institute 1 at Jülich. His background is algorithm development and high performance computing (HPC) for electronic structure calculations in the framework of density functional theory (DFT). By moving from solid state physics deeper into HPC, he got into touch with several other simulation-assisted sciences and various programming paradigms. He is a coauthor of the 2016 ISC Hans-Meuer-award paper of JSC and IBM. Currently, he is an active developer of the GPU-accelerated linear-scaling DFT application AngstromCube.