Symposium CH03—Advances in In Situ and Operando TEM Methods for the Study of Dynamic Processes in Materials III
Jonathan Hollenbach, The Johns Hopkins University
The New Operando—Incorporation Intelligent Decisions into In Situ TEM
Written by Henry Quansah Afful
Materials are typically studied using a bottom-down approach of breaking and analyzing them. Why not take advantage of advanced microscopy techniques to study individual atoms at the lattice level to build new structures just like legos? Combining machine learning (ML) with microscopy can do just that. For instance, electron energy loss spectroscopy (EELS) generates a ton of data (sometimes terabytes per second) but extracting meaningful information from these is very challenging for humans. ML can help analyze this data and obtain some key fingerprints necessary for tailoring materials properties. Jonathan Hollenbach demonstrates this in SrFeO3 in which a very noisy EELS spectra was obtained. A convolutional autoencoder was used to screen, denoise, and classify the experimental data according to computationally obtained spectra and generate usable information which would have been impossible otherwise. This was extended to study annealing of two-dimensional (2D) metal carbides (MXenes) and the information obtained can be used to spatially control the termination of the 2D structures to build them from ground up.