Automated atomic site determination by four-dimensional scanning transmission electron microscopy data analytics

Abstract

Automated atomic column detection and identification constitutes an active open front in advanced scanning transmission electron microscopy techniques. In this work we use clustering algorithms in combination with dimensionality reduction techniques to identify specific columns in a series of very different cutting-edge materials, ranging from ultrathin 2D materials to bulk semiconductors or complex oxides, which include different types of columns (heavy and light), and thus pose a challenge towards automated detection. By implementing a three-stage cascaded clustering pipeline, we are able to automatically identify all atomic column sites of our test materials and resolve them from the background interatomic space. This approach could enable new data-driven in-depth analysis of materials, allowing the automatic detection of chemical and structural characteristics of materials.