Data-Driven Electron Microscopy
A single 4D-STEM scan or EELS map can generate a vast amount of data, often beyond the practical limits of conventional manual analysis. AEML develops computational workflows that integrate AI/ML into electron microscopy data processing to automate defect detection, phase segmentation, and spectral decomposition. In doing so, we extend the power of localized electron microscopy toward statistically meaningful analysis of large-scale datasets.
Detailed Research Topics
4D-STEM Data Processing
We use and contribute to open-source tools (py4DSTEM, pyXem, HyperSpy) for processing convergent beam electron diffraction datasets. Our workflows include strain mapping from nanobeam diffraction, virtual detector imaging, orientation mapping, and phase identification using unsupervised clustering of diffraction patterns.
AI/ML-Assisted Analysis
We apply deep learning and statistical methods to automate tasks that traditionally require expert judgment: defect detection in STEM images, phase segmentation, and spectral decomposition of EELS data. The goal is not to replace human expertise but to augment it — letting researchers focus on interpretation rather than repetitive measurements.
Multi-slice Electron Ptychography
With the upcoming Spectra Ultra and its 16-segment Panther detector, we are developing 3D phase retrieval algorithms based on multi-slice ptychography. This technique reconstructs atomic-resolution 3D structure from 4D-STEM data, with depth sensitivity that conventional imaging cannot achieve.
Automated EM Workflows
We envision a fully integrated pipeline: sample preparation → data acquisition → processing → report. By incorporating AI/ML at each step — from automated alignment and focus to intelligent data triage — we aim to increase throughput and reproducibility while reducing operator dependency.
Vision
The future of electron microscopy is not just about better hardware — it’s about better data. We believe that combining state-of-the-art instruments with modern computational approaches will unlock materials insights that are currently hidden in the noise.
Representative Publications
- J.-W. Cho, Y.-W. Byeon et al., “Next-generation analysis technologies of nano materials: based on electron microscopy,” Trends in Metals and Materials Engineering (2015)
- Y. Ko, Y.-W. Byeon et al., “Omics-enabled understanding of electric aircraft battery electrolytes,” Joule (2024)
Tools & Software
- py4DSTEM — 4D-STEM data analysis
- HyperSpy — Multi-dimensional data processing
- abTEM — Ab initio TEM simulation
- pyXem — Crystallographic orientation mapping
- Python / PyTorch — Custom AI/ML pipelines