About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Four-Dimensional Scanning Transmission Electron Microscopy (4D STEM) for Quantitative Materials Characterization
|
| Presentation Title |
Latent‑Space Diffraction Mapping of Recrystallization Pathways in GaAs |
| Author(s) |
Ellis Rae Kennedy, Kwanghwi Je, Erik Thiede |
| On-Site Speaker (Planned) |
Ellis Rae Kennedy |
| Abstract Scope |
Recrystallization in amorphous GaAs proceeds through subtle, spatially heterogeneous structural rearrangements that are difficult to quantify with conventional diffraction. 4D-STEM captures this complexity, but its high-dimensional diffraction data require new strategies to extract pathway-level insight. We introduce a latent space diffraction mapping framework that couples a convolutional autoencoder, unsupervised clustering, and Markov‑style trajectory analysis to reconstruct nanoscale recrystallization pathways during in situ heating. The method identifies amorphous, crystalline, polycrystalline, twinned, and hybrid intermediates, revealing a hierarchy of partially ordered states.
Applied to GaAs, the approach uncovers a temperature-driven reorganization of the transformation landscape near 250 °C, where hybrids act as precursors to polycrystalline grains and dense {111} nanotwin networks. The latent representation also exposes weak symmetry amorphous motifs that precede ordering and are missed by scalar descriptors. This workflow establishes a general route for quantitative 4D‑STEM, enabling mapping of phase transformation pathways and metastability in complex materials. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Phase Transformations, Characterization, Machine Learning |