Defect complexes in CrSBr revealed through electron microscopy and deep learning

TL;DR

This study combines electron microscopy, deep learning, and ab-initio calculations to identify and classify various defect complexes in bilayer CrSBr, revealing their structures, electronic states, and potential optical activity, advancing defect understanding in layered materials.

Contribution

The paper introduces a novel machine learning workflow for defect detection and classification in CrSBr, uncovering new defect types and linking them to electronic and magnetic properties.

Findings

Identification of multiple defect complexes in CrSBr

Localization of electronic states due to defects

Prediction of optical activity of defect states

Abstract

Atomic defects underpin the properties of van der Waals materials, and their understanding is essential for advancing quantum and energy technologies. Scanning transmission electron microscopy is a powerful tool for defect identification in atomically thin materials, and extending it to multilayer and beam-sensitive materials would accelerate their exploration. Here we establish a comprehensive defect library in a bilayer of the magnetic quasi-1D semiconductor CrSBr by combining atomic-resolution imaging, deep learning, and ab-initio calculations. We apply a custom-developed machine learning work flow to detect, classify and average point vacancy defects. This classification enables us to uncover several distinct Cr interstitial defect complexes, combined Cr and Br vacancy defect complexes and lines of vacancy defects that extend over many unit cells. We show that their occurrence is in agreement with our computed structures and binding energy densities, reflecting the intriguing layer interlocked crystal structure of CrSBr. Our ab-initio calculations show that the interstitial defect complexes give rise to highly localized electronic states. These states are of particular interest due to the reduced electronic dimensionality and magnetic properties of CrSBr and are furthermore predicted to be optically active. Our results broaden the scope of defect studies in challenging materials and reveal new defect types in bilayer CrSBr that can be extrapolated to the bulk and to over 20 materials belonging to the same FeOCl structural family.