주사투과전자현미경연구실

Machine learning and deep learning push past the limits of manual electron-microscopy analysis. From defect quantification in 2D TMDs and Pt nanoparticle morphology (Attention U-Net) to crystallography, tomographic reconstruction, and PEMFC electrode characterization, we automate and quantify the full EM workflow. Electromigration degrades Cu-Cu direct-bonding lines, yet its mechanism is hard to isolate because competing diffusion pathways act concurrently. Using correlative spatio-temporal mapping by in situ 4D‑STEM, we resolved a two-stage process. Mass transport is first governed by grain-boundary diffusion, then accelerates via surface diffusion once nanovoids nucleate at triple junctions.