Fabrication of Atomic-scale Defect Structures within 2D Materials through Automated Electron Beam Control

Atomically precise manipulation of matter, conducted through the use of the sub-Å sized electron (e-) beam of an aberration-corrected scanning transmission electron microscope (STEM), has become a reality in recent years. This has allowed for the fabrication of unique and desired nanostructures both within and along the edge of monolayers of 2D materials such as in transition metal dichalcogenides (e.g. MoS2)[1-4]. Prior work has demonstrated that these types of 1D-2D edge heterostructures and internal defect structures can locally alter the functional properties (e.g. electronic, magnetic, optical, and catalytic properties) [5-7]. However, while previous works have shown that control over the e- beam scanning pathways can have a great impact on the formation of different defect structures [7-8], a greater degree of automated feedback and control is required to make atomic fabrication techniques feasible and precise. In this work, atomic transformations in MoS2 were investigated using a deep convolutional neural network (DCNN) beam control system that works in-line with the STEM workflow through a programmable Python-enabled interface, known as the Atomic Fabrication Controller (AFC) [9-11]. The experimental workflow was used to perform automated atomic fabrication experiments can be seen in Figure 1. For this work, structures such as the Mo6S6 nanowire-type (MoS-NW) nanopore edge structure and single vacancy line (SVL) defects were fabricated using this AFC workflow. Utilizing this automated system, the images acquired during experimentation are quickly decoded into atomic coordinates and defect types in real time via the ensemble learning iterative training (ELIT) approach for DCNNs [11]. Using this information, the e- beam can then be precisely positioned over the desired atom species to break the local bonds and to induce atomic transformations. Through feedback control, the e- beam then moves to the next decoded location while periodically re-decoding the image to inform the system of the fabrication progress. Different workflows were developed and employed to facilitate the formation of the different defect structures described previously. Through this method the full fabrication and characterization of nanostructures could be achieved reliably without altering the surrounding material. Additionally, through repeated decodings, the automated e- beam system can reveal trends in the formation of different defect structures that help inform future atomic fabrication endeavors. In this regard, in-depth studies of this fabrication process and further usage and implementation of this technique are expected to lead to the fabrication of full nanoscale devices using automated STEM techniques [12]. Atomic fabrication through AI-enabled decoding (a) Experimental schematic of the Atomic Fabrication Controller within the STEM workflow. (b-c) Initial STEM image of MoS2 and its accompanying decoded image with Mo atoms in blue, S2 sites in yellow, the single vacancy line Mo in white, and the selected S targets atoms marked in red. (d-e) Final STEM image with the Mo6S6 nanowire edge type formed along the bottom marked in orange.