Cornell University
Atomic-scale Analysis of Defects in Battery Materials through Bayesian-Optimized Electron Ptychography
Abstract
dc:description.abstractA long-standing challenge in battery research is understanding lithium diffusion mechanisms at the atomic scale. Direct imaging of lithium is difficult because it is a light element and highly sensitive to radiation damage. Yet, atomic-resolution imaging is essential for identifying local defects and structural variations that strongly influence battery performance. While some conventional scanning transmission electron microscopy (STEM) methods can resolve both light and heavy elements in such materials, they are generally limited to two-dimensional projections. In this work, we demonstrate that multislice electron ptychography (MEP) captures three-dimensional structural information in a dose-efficient manner, while also providing higher resolution imaging of both light and heavy elements in a lithium-ion cathode. Using the depth-sectioning capability of MEP, we directly visualize lithium vacancy clusters buried within the cathode at the level of individual lithium columns. This capability to track lithium distributions in three dimensions will be crucial for advancing our understanding of battery charging mechanisms and guiding future material improvements. While these results highlight the potential of MEP for studying battery materials, practical challenges remain in its application. For beam-sensitive battery materials, conventional tilting to crystallographic zone axes is taxing and often impractical, particularly when samples contain multiple grains or drift due to instabilities of cryogenic holders. Small mistilts can, in principle, be corrected during MEP reconstruction, but solving for tilt alongside hundreds of other reconstruction parameters frequently leads to convergence failures or requires extensive simulations. Because MEP reconstruction is already computationally intensive, this creates a significant bottleneck. To address this challenge, we developed a workflow that incorporates Bayesian optimization of composite functions (BOCF) to robustly estimate specimen tilt and thickness from position-averaged convergent beam electron diffraction (PACBED) patterns. BOCF reliably converges in the complex three-parameter space and achieves objective values that are orders of magnitude better than conventional BO strategies. Notably, it reduces the computational demand to only several tens of simulations, in contrast to convolutional neural networks that require thousands that are restricted to specific experimental conditions and material systems. We also demonstrate that the optimized parameters can indeed correct specimen mistilts in MEP reconstructions, resulting in clearly resolved atom columns. By making previously unusable tilted datasets accessible, the method enhances the overall throughput of MEP and relaxes experimental constraints not only for beam-sensitive materials but also for other systems in general.
Degree
thesis:*- Name thesis:degree_name
- Ph. D., Materials Science and Engineering
- Level thesis:degree_level
- Doctor of Philosophy
- Discipline thesis:degree_discipline
- Materials Science and Engineering
- Grantor
- Cornell University
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Yoon, Dasol
- Chair dc:contributor.chair
-
- Muller, David
- Committee members dc:contributor.committeemember
-
- Abruna, Hector
- Matteson, David
Subjects
dc:subject × 6Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Bibid
- 17423343
- Dc Identifier Other
-
ProQuest Submission ID: 15308
ProQuest Publication ID: 32277614