Identifying Single-Atom Defects of 2D Materials on the Million-Atom Scale via Deep Learning

Aberration-corrected scanning transmission electron microscopy (STEM) is an important tool to study how atomic defects such as vacancies and substitutions impact the structure and properties of 2D materials^1-2. Yet, high-precision characterization of defects in 2D materials remains challenging because they are irradiation sensitive, making it difficult to achieve high resolution and signal-to-noise ratio (SNR) measurements without modifying the intrinsic structure. Here, we combine automated acquisition, class averaging, and machine learning to acquire and analyze large datasets of STEM images of 2D materials. We analyzed an atomic resolution, ~million atom dataset to determine the precise atomic structures and spatial distributions of 7 different types of point defects in a 2D transition metal dichalcogenide (TMDC). By summing thousands of images from nominally identical defects, we improve the measurement precision to ~0.3 pm, sufficient to detect pm-scale strain field oscillations from a single atom defects up to ~1 nm away. This technique also allows us to observe how defect clusters interact with each other through coupled strain fields.

1. O. Krivanek et al., Nat. 464 (2010), p. 571-574

2. P. Y. Huang et al., Sci. 342 (2013), p. 224-227