Final Project Report Project Title: Machine Learning Enabled Advanced Electron Tomography for Resolving Chemical Inhomogeneity and Materials Dynamics in Lithium-Ion Battery Electrodes Award Number: DE-SC0021204 PI: Huolin Xin Department of Physics and Astronomy, University of California, Irvine Project Period: 09/01/2020 – 08/31/2025 1. Project Objectives The objective of this project is to develop machine learning-assisted electron microscopy, together with three-dimensional, cryogenic, and in-situ imaging techniques, for resolving chemical inhomogeneity and materials dynamics in lithium battery electrodes and interfaces. The project aims to expand the spatial, temporal, and dimensional resolution of transmission electron microscopy and to enable quantitative analysis of beam-sensitive battery materials. 2. Major Accomplishments Over the course of the project, we established a suite of machine learning–enabled imaging and spectroscopy methods and applied them to key scientific problems in lithium-ion and lithium-metal batteries. First, we developed foundational deep-learning tools for atomic-resolution electron microscopy. This includes the TEMImageNet training library and AtomSegNet deep-learning models for atom segmentation, localization, denoising, and super-resolution imaging(1). These models enable robust analysis of experimental atomic-resolution STEM images across a wide range of materials and imaging conditions. The methods outperform conventional fitting-based approaches and have been released as open-source tools. Using these capabilities, we resolved atomic-scale degradation mechanisms in high-Ni layered cathodes, including irreversible O3 to O1 phase transformations, oxygen loss, and interfacial strain accumulation(2, 3). We subsequently applied these machine learning–enabled imaging techniques to technologically relevant cathode systems. We resolved the role of residual grain boundaries in quasi-single- crystalline layered oxide cathodes and demonstrated their impact on chemomechanical degradation and capacity fade(4). We also uncovered atomic-scale stabilization mechanisms in ultrahigh-Ni single-crystalline cathodes, showing how reduced lithiation inhomogeneity and suppressed oxygen-loss-induced phase transitions lead to improved cycling stability(5). In parallel, we developed cryogenic STEM and cryogenic electron tomography methods for lithium metal and solid-state battery interfaces. Using cryoSTEM and cryoET, we directly resolved the structure and chemistry of the Li metal–solid polymer electrolyte interface(6). These measurements showed that sustained chemical reactions prevent the formation of a stable passivating interphase in certain polymer electrolytes and lead to stress corrosion cracking of lithium deposits. Based on this insight, we demonstrated electrolyte modification strategies that stabilize the Li surface and improve cycling performance in solid-state batteries. We further developed in-situ TEM methods to study lithium plating and stripping in solid-state and constrained environments. These experiments revealed tension-induced cavitation during lithium stripping and identified a quantitative criterion linking mechanical stability of supporting structures to stripping behavior and dead lithium formation(7). Additional in-situ studies resolved the incipient-stage nucleation and growth of lithium metal, revealing a reversible liquid-to-solid-like transformation driven by size-dependent lithium diffusivity(8). We also demonstrated a reactive wetting strategy that enables conformal and reversible lithium deposition through substrate- mediated interfacial chemistry(9). We also identified the nanoscale origin of the soft-to-hard short-circuit transition in inorganic solid- state electrolytes using in situ transmission electron microscopy(10). Electronic-leakage-induced Li⁰ precipitation at voids and grain boundaries causes transient soft short circuits that evolve into irreversible hard short circuits through Li⁰ penetration and chemomechanical fracture. Introducing a mechanically resilient and electronically insulating polymer network suppresses Li⁰ precipitation and electronic leakage, leading to improved electrolyte stability. In an ultrahigh-Ni cathode material, using in situ atomic-resolution transmission electron microscopy, we directly resolved the electrochemically driven nucleation, motion, and annihilation of dislocations in layered oxide cathodes(11). We quantified dislocation glide and climb velocities and showed that dislocation activity mediates crack nucleation, phase transformation, and lattice reorientation during cycling. These results establish dislocations as key drivers of chemomechanical degradation in layered oxide cathodes. The project also advanced three-dimensional characterization of lithium metal nucleation. Using cryoSTEM and cryoET, we determined the morphology, crystallography, and interphase chemistry of single-crystalline lithium seeds grown on lithiophobic substrates(12). These results provided direct evidence for dense lithium deposition enabled by controlled nucleation and fast lithium transport, offering design principles for fast-charging lithium-metal batteries. In addition to imaging, we developed machine learning–based spectroscopy analysis tools. We created an EELS data synthesizer and trained deep neural networks for automated core-loss edge recognition(13, 14). We further developed a calibration-free and reference-free deep- learning model for quantitative decomposition of mixed Mn oxidation states from EELS and XAS L₂,₃ edges(15). This approach resolves a long-standing limitation in quantitative near-edge structure analysis and enables robust analysis across instruments and experimental conditions. Finally, we applied our machine learning–enabled phase mapping and imaging models to all- solid-state battery cathode–electrolyte interfaces(16). These studies revealed atomic-scale origins of chemomechanical failure that are distinct from liquid-electrolyte systems, including surface frustration, interlayer shear, and complex O1 phase formation. These findings bridge the understanding between degradation mechanisms in solid-state and conventional lithium-ion batteries. 3. Training and Professional Development This project provided extensive research and training opportunities for postdoctoral researchers, graduate students, and undergraduate students in advanced electron microscopy, machine learning, and battery materials science. Trainees gained experience in atomic-resolution imaging, cryogenic microscopy, in-situ TEM, spectroscopy, and data-driven analysis. The project also supported outreach and educational activities involving undergraduate and high-school students. 4. Dissemination of Results The results of this project have been disseminated through publications in high-impact journals, including Nature Nanotechnology, Nature Energy, Nature Materials, Nano Letters, Advanced Materials, ACS Energy Letters, Energy Storage Materials, Scientific Reports, and Patterns. Findings have been presented at major national and international conferences. Open-source software and training datasets developed in this project have been released to the research community. 5. Impact The methods developed in this project significantly improve the spatial, temporal, and dimensional resolution of transmission electron microscopy and enable quantitative analysis of beam-sensitive battery materials. These techniques are broadly applicable to other materials systems, including soft, biological, and structural materials. The project contributed to the development of human resources in electron microscopy and energy materials and established an AI-driven imaging platform that strengthens institutional research infrastructure. The insights gained advance the understanding of battery degradation and interfacial processes and support the development of next-generation energy storage technologies. 6. List of Publications 1. R. Lin, R. Zhang, C. Wang, X.-Q. Yang, H. L. Xin, TEMImageNet training library and AtomSegNet deep-learning models for high-precision atom segmentation, localization, denoising, and deblurring of atomic-resolution images. Scientific reports 11, 5386 (2021). 2. C. Wang, R. Zhang, K. Kisslinger, H. L. Xin, Atomic-scale observation of O1 faulted phase- induced deactivation of LiNiO2 at high voltage. Nano letters 21, 3657-3663 (2021). 3. C. Wang et al., Resolving complex intralayer transition motifs in high-Ni-content layered cathode materials for lithium-ion batteries. Nature materials 22, 235-241 (2023). 4. R. Zhang, C. Wang, M. Ge, H. L. Xin, Accelerated degradation in a quasi-single-crystalline layered oxide cathode for lithium-ion batteries caused by residual grain boundaries. Nano letters 22, 3818-3824 (2022). 5. C. Wang et al., Chemomechanically stable ultrahigh-Ni single-crystalline cathodes with improved oxygen retention and delayed phase degradations. Nano letters 21, 9797-9804 (2021). 6. R. Lin et al., Characterization of the structure and chemistry of the solid–electrolyte interface by cryo-EM leads to high-performance solid-state Li-metal batteries. Nature nanotechnology 17, 768-776 (2022). 7. C. Wang et al., Tension‐induced cavitation in Li‐metal stripping. Advanced materials 35, 2209091 (2023). 8. C. Wang et al., Direct observation of nucleation and growth behaviors of lithium by in situ electron microscopy. ACS Energy Letters 8, 1929-1935 (2023). 9. P. Zou, C. Wang, J. Qin, R. Zhang, H. L. Xin, A reactive wetting strategy improves lithium metal reversibility. Energy Storage Materials 58, 176-183 (2023). 10. C. Wang et al., Nanoscale Origin of the Soft-to-Hard Short-Circuit Transition in Inorganic Solid-State Electrolytes. Journal of the American Chemical Society 147, 19084-19092 (2025). 11. C. Wang, R. Zhang, J. Li, H. L. Xin, Resolving electrochemically triggered topological defect dynamics and structural degradation in layered oxides. Proceedings of the National Academy of Sciences 122, e2409494122 (2025). 12. Z. Wu et al., Growing single-crystalline seeds on lithiophobic substrates to enable fast- charging lithium-metal batteries. Nature Energy 8, 340-350 (2023). 13. L. Kong, Z. Ji, H. L. Xin, Electron energy loss spectroscopy database synthesis and automation of core-loss edge recognition by deep-learning neural networks. Scientific reports 12, 22183 (2022). 14. S. Zheng, C. Wang, X. Yuan, H. L. Xin, Super-compression of large electron microscopy time series by deep compressive sensing learning. Patterns 2, (2021). 15. Z. Ji, M. Hu, H. L. Xin, MnEdgeNet for accurate decomposition of mixed oxidation states for Mn XAS and EELS L2, 3 edges without reference and calibration. Scientific Reports 13, 14132 (2023). 16. C. Wang, Y. Jing, D. Zhu, H. L. Xin, Atomic origin of chemomechanical failure of layered cathodes in all-solid-state batteries. Journal of the American Chemical Society 146, 17712-17718 (2024).