R. Unocic
North Carolina State University,
United States
Summary:
Recent advances in autonomous experimentation are rapidly transforming the capabilities of direct-write nanofabrication, enabling precise manipulation of matter from the atomic to nanoscale with an unprecedented degree of control. At the forefront of this transformation is the integration of automation and machine learning (ML) with scanning transmission electron microscopy (STEM)-based direct-write approaches, which together are redefining the way materials can be engineered and modified at the smallest length scales. In this work, we demonstrate a comprehensive methodology that couples real-time feedback-controlled STEM imaging with ML-based image analysis to guide atomic-scale transformations with high spatial and temporal resolution. Building upon prior developments in atomic defect engineering in two-dimensional (2D) materials, and leveraging high-throughput, deep learning-enhanced STEM workflows, we deploy a closed-loop experimental platform. This system employs deep convolutional neural networks (DCNNs) to classify evolving atomic configurations during dynamic electron-beam-induced transformations, enabling on-the-fly adjustment of beam irradiation parameters. Such adaptive control allows for precision operations, including layer-by-layer material removal, the formation of sub-nanometer pores, and the stabilization of transient metastable phases within Ti₃C₂Tₓ MXenes. Our in-situ observations under varying thermal and electron beam irradiation conditions reveal key mechanisms governing atomic mobility and beam-matter interactions, aligning with previously reported behaviors in thermally activated edge reconstructions in 2D transition metal dichalcogenides (TMDs). By combining real-time analysis, automated feedback control, and multimodal environmental capabilities, this work establishes a robust platform for autonomous, atomically precise fabrication. The resulting system serves as a powerful toolset for materials discovery, defect engineering, and dynamic structural modulation offering new opportunities for scientific exploration in both fundamental and applied research.