Scanning transmission electron microscopy (STEM) with aberration correction can map atomic order with sub-ångström resolution. Yet, many nanomaterials are sensitive to high-energy electrons, which limits the applicable electron dose. This results in images that present weak contrast, hindering an unambiguous determination of the atomic structure. Jarmo Fatermans et al. now propose an advanced fitting procedure that enables a more reliable and unbiased detection of single atoms in high-noise, low-contrast electron microscopy images.
The researchers combine a known model fitting with an advanced statistical selection method and apply this to STEM images of atomic-scale systems. The fitting approximates the contrast of atomic columns in an STEM image with Gaussian functions on a constant background. Then, they introduce an approximate analytical implementation of a probability rule, which, based on the pixel intensities, determines the probability for a specific number of atomic columns to give rise to this image. The result is a set of configurations with different numbers of atomic rows and their position together with the probability for each configuration. This method is more reliable than conventional peak-fitting methods, avoids the bias in visual inspection and provides the likeliness of one atomic configuration compared to alternative structures.