OPTIMIZING PRECISION OF HIGH-ANGLE ANNULAR DARK FIELD IMAGES IN SCANNING TRANSMISSION ELECTRON MICROSCOPY

Abstract

Scanning Transmission Electron Microscopy (STEM) is a key technique for analyzing crystal structures on an atomic scale. When fully optimized, high angle annular dark field STEM (HAADF-STEM) yields images in which atoms appear as bright spots on a dark background, making it easy to extract atomic positions quantitatively. It is crucial for us to optimize the precision of these measurements in order to be able to determine any possible atomic shifts that can arise due to the material's local properties such as ferroelectricity or charge order. In this thesis, I investigate the limits of precision in determining atomic column positions in HAADF STEM data and how image acquisition parameters and data analysis can be optimized. In each experiment, I registered fast-acquisition stacks of SrTiO3 (STO) images, determined the atomic centers in the registered images using a mixture of Gaussians for each atomic column, and finally measured the distances between each Sr column and its nearest Ti/O neighbors. Compared to previous work that required non-rigid image registration to obtain high precision from slow-scan stacks, I demonstrate that for ultra-fast scans, simple rigid registration is sufficient and precisions reached (about 2-3pm) are comparable to those from non-rigid registration. Furthermore, no improvements are observed as a result of binning, nor when correcting for the scan direction with REVSTEM. I additionally compared my experimental data to simulated data with varying levels of noise added, and concluded that noise cannot possibly be the main factor in limiting my precision. In order to reach a precision of 2-3 picometers in atomic column distance measurements, the image acquisition parameters have to be optimized. Increasing stack size improved precision, though this still seems to level off around 2 or 2.5 pm. Additionally, in comparing various combinations of possible pixel densities and dwell times, I report that 1024 x 1024 x 0.25 us/pixel scan parameters gives the best precision for typical beam currents at 300 keV and fields of view on the order of 6.6 nm. Although some of these experimental parameter changes did improve the precision, none of my data sets were able to get a precision in the Sr-Ti/O column distances in SrTiO3 below 2 pm, indicating that there is some underlying systematic process that limits my precision. Future experiments performed on different samples or at cryogenic temperatures could determine if this issue is due to the microscope setup or due to a physical shift in the lattice itself.