Invited Talk: Exploring the Atomic World: Advancing Electronic Microscopy with Deep Learning and Generative Models
Chia-Hao Lee,
Cornell University, NY, USA
Abstract.
Understanding atomic structure is essential for designing and engineering next-generation materials. While electron microscopy has long been a powerful tool for visualizing the atomic world, it remains challenging to acquire and analyze large-scale atomic-resolution datasets, and to infer meaningful 3D information from inherently incomplete or noisy 2D projections. In this talk, I will present how advances in deep learning and generative models can address key bottlenecks in atomic-scale materials characterization with unprecedented clarity.
This talk highlights three complementary efforts: (1) leveraging fully convolutional networks (FCN) to perform large-scale defect identification and sub-picometer precision strain mapping at atomic resolution from scanning transmission electron microscopy (STEM) images; (2) using CycleGAN to generate realistic synthetic microscopy images, which serve as high-quality training data for downstream machine learning tasks—especially valuable in data-scarce regimes; and (3) integrating diffusion-based generative priors with iterative solvers in electron ptychography to recover 3D structural information with enhanced depth resolution, by leveraging rich structural information from large scale materials databases.
Together, these approaches demonstrate how deep learning techniques and generative AI can be tightly coupled with physical imaging systems to push the frontier of high-throughput, high-precision materials characterization at atomic scale.
References:
[1] C.-H. Lee, A. Khan†, D. Luo†, et al., “Deep Learning Enabled Strain Mapping of Single-Atom Defects in Two-Dimensional Transition Metal Dichalcogenides with Sub-Picometer Precision”, Nano Letters (2020) DOI: 10.1021/acs.nanolett.0c00269
[2] A. Khan†, C.-H. Lee†, P. Y. Huang, et al., “Leveraging generative adversarial networks to create realistic scanning transmission electron microscopy images” npj Computational Materials (2023) DOI: 10.1038/s41524-023-01042-3
[3] C.-H. Lee and D.A. Muller, “Enhancing Depth Resolution of Multislice Ptychography with Data-Driven Prior and Regularization” Microscopy and Microanalysis (2024) DOI: 10.1093/mam/ozae044.942