REDUCED AND COMPRESSED DATA REPRESENTATIONS IN 4D SCANNING TRANSMISSION ELECTRON MICROSCOPY

Abstract

4D Scanning Transmission Electron Microscopy (STEM) contains rich momentum-resolved scattering information, which contains annular bright field (ABF), low annular annul dark field (LAADF), and high annular annul dark field (HAADF) images. However, 4D-STEM data’s large file size makes it hard to store and analyze. We show that for quasi-periodic molybdenum disulfide (MoS2) samples imaged with a focused probe, their 4D-STEM data is highly redundant. Therefore, distinguishing and extracting domain features with an acceptable range of loss from 4D-STEM data is possible and necessary. Based on principal component analysis (PCA) and Zernike component methods, we show that the 4D-STEM data is highly redundant; hence it can be reduced and compressed with minimal impact on imaging. We also examine the physical interpretation of specific Zernike modes of the 4D STEM data: information about the probe’s beam while scanning the sample and the phase contrast of different elements in the model. In our MoS2 4D-STEM data, our Zernike projection suggested a possible “stacking” of sulphuric atoms at rare interstitial sites, which I plan to examine using simulations. Overall, this work justifies our method to reduce and compress 4D-STEM data in a physically meaningful and interpretable manner.