| Abstract Scope |
Four-dimensional scanning transmission electron microscopy provides a common platform for diffraction-based mapping and phase-retrieval imaging across length scales. This overview will highlight two families of AI-assisted 4D-STEM methods. First, neural priors embedded within differentiable multislice ptychographic forward models improve robustness to noise, accelerate convergence, regularize depth, and reduce manual parameter tuning while jointly refining specimen and probe parameters. Second, model-based analysis of nanobeam and precession diffraction data uses machine-learning frameworks to fit the full diffraction pattern for improved strain precision and automated determination of crystal phase and orientation. We will discuss recent applications to atomic-resolution ptychography, quantitative strain mapping, and phase identification in complex polycrystalline materials, including hafnium zirconium oxide ferroelectric films. We will also describe ongoing efforts toward fully differentiable orientation and phase mapping with dynamical diffraction, enabling unified parameter refinement, uncertainty estimation, and more reproducible 4D-STEM analysis. |