Deep Learning Enabled Strain Mapping of Single-Atom Defects in Two-Dimensional Transition Metal Dichalcogenides with Sub-Picometer Precision

Abstract

Two-dimensional (2D) materials offer an ideal platform to study the strain fields induced by individual atomic defects, yet challenges associated with radiation damage have so far limited electron microscopy methods to probe these atomic-scale strain fields. Here, we demonstrate an approach to probe single-atom defects with sub-picometer precision in a monolayer 2D transition metal dichalcogenide, WSe2–2x Te2x . We utilize deep learning to mine large data sets of aberration-corrected scanning transmission electron microscopy images to locate and classify point defects. By combining hundreds of images of nominally identical defects, we generate high signal-to-noise class averages which allow us to measure 2D atomic spacings with up to 0.2 pm precision. Our methods reveal that Se vacancies introduce complex, oscillating strain fields in the WSe2–2x Te2x lattice that correspond to alternating rings of lattice expansion and contraction. These results indicate the potential impact of computer vision for the development of high-precision electron microscopy methods for beam-sensitive materials.

Picometer-Precision Atomic Position Tracking through Electron Microscopy

“…Most recently, the deep learning algorithm was employed to identify various types of point defects in 2-D transition metal dichalcogenides monolayers from a huge amount of STEM image data. Later, the measurement was conducted on the averaged image of different types of defects and this method also demonstrated sub-picometer level precision on the distortion around those defects 18 . Consequently, as a future plan for increasing the analysis capacity, we are in the progress of developing and implementing more advanced algorithms such as deep learning.…”

“…Another example is the visualization of the polar vortex structures achieved in the SrTiO 3 -PbTiO 3 superlattices, achieved through calculation of the titanium atomic column displacements with respect to the strontium and lead column positions 14 . Finally, the advances in computer vision algorithms, such as image denoising with non-local principle component analysis 15 , Richardson and Lucy deconvolution 16 , drift-correction with non-linear registration 17 , and pattern recognition with deep learning, have significantly strengthened the accuracy of the measurement to sub-picometer precision 18 . One such example is the alignment and image registration of multiple fast-scan cryogenic-STEM images to enhance the signal-to-noise ratio.…”

TEMImageNet training library and AtomSegNet deep-learning models for high-precision atom segmentation, localization, denoising, and deblurring of atomic-resolution images

“…LeBeau et al used AlexNet, a version of deep CNNs primarily used for classification, to preprocess SrTiO 3 convergent beam electron diffraction patterns and determine crystal thicknesses 23 . Huang et al used CNNs to locate defects and extract strain fields in 2d materials 24 . Xin et al used generative adversarial models to inpaint and restore the missing-wedge information in electron tomography datasets 25,26 .…”

Ensemble learning-iterative training machine learning for uncertainty quantification and automated experiment in atom-resolved microscopy

“…In STM, the application of DL for image analysis was pioneered by Ziatdinov et al for molecular resolved imaging 41 and Wolkow for atomically resolved imaging 42 , and in STEM by Ziatdinov et al 43 . This initial effort has grown exponentially in recent years [44][45][46] .…”