Probing Electron Beam Induced Transformations on a Single-Defect Level via Automated Scanning Transmission Electron Microscopy

Abstract

A robust approach for real-time analysis of the scanning transmission electron microscopy (STEM) data streams, based on ensemble learning and iterative training (ELIT) of deep convolutional neural networks, is implemented on an operational microscope, enabling the exploration of the dynamics of specific atomic configurations under electron beam irradiation via an automated experiment in STEM. Combined with beam control, this approach allows studying beam effects on selected atomic groups and chemical bonds in a fully automated mode. Here, we demonstrate atomically precise engineering of single vacancy lines in transition metal dichalcogenides and the creation and identification of topological defects in graphene. The ELIT-based approach facilitates direct on-the-fly analysis of the STEM data and engenders real-time feedback schemes for probing electron beam chemistry, atomic manipulation, and atom by atom assembly.

Creation of Single Vacancies in hBN with Electron Irradiation

“…Now that it is clear that hBN is remarkably radiation resistant at primary beam energies below 80 keV, the creation of color centers could potentially be greatly optimized by site‐selective irradiation using a focused STEM electron probe [ 48 ] coupled with low‐dose imaging and automation enabled by machine learning. [ 49 ]…”

Identifying and manipulating single atoms with scanning transmission electron microscopy

“…146 Automated atom detection using neural networks 147 running in real-time on the microscope 148 and combined with real-time feedback of the scattering signal 121,136 are enabling more than one group to build fully automated manipulation software. 149,150 A self-driving electron microscope 151 appears to be finally within reach.…”

Probe microscopy is all you need *

“…Many times, these can be extremely difficult for the human operator to notice (especially in the case of a single atomic defect) during experimental operation, therefore a rapid means for detecting and classifying atomic coordinates is of critical importance. In figure 5(b), a single layer of graphene is shown, where the atomic coordinates given by a trained deep ensemble neural network [152,155] are shown. Graph analysis is performed given the coordinates, from which vacancy defect structures like a Stone-Wales [156] (or, 5-7 ring) defect strongly stand out.…”

“…what geometries and positioning/type of defects are necessary to induce the required functional responses. This space is ripe for exploration with ML methodologies [152], and one can envision both generative as well as reinforcement learning approaches to tackle this inverse problem, ideally incorporating simulations where necessary to reduce data requirements.…”

“…Scale bar in (a) 2 nm, (b) and (c) both 1 nm. Reprinted with permission from[152]. Copyright (2022) American Chemical Society.…”