About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Frontiers of Machine Learning on Materials Discovery
|
| Presentation Title |
Accelerating Defect Predictions in Semiconductors Using Crystal Graphs |
| Author(s) |
Arun Kumar Mannodi Kanakkithodi, Md Habibur Rahman |
| On-Site Speaker (Planned) |
Md Habibur Rahman |
| Abstract Scope |
Quick defect predictions are complicated by difficulties in assigning measured levels to specific defects and the expense of large-supercell computations involving charge corrections and advanced functionals. We address this issue by combining density functional theory (DFT) simulations and crystal Graph-based Neural Network (GNN) models to drive the accurate prediction of defect formation energies (DFE) of native defects, impurities/dopants, and defect complexes, in a variety of technologically-important binary and ternary semiconductors. Trained on a dynamically growing defect dataset of > 15,000 defective structures, models based on Atomistic Line Graph Neural Network (ALIGNN) give best DFE prediction accuracy of ~ 98%. We apply these models to screen across thousands of hypothetical single defects/dopants and complexes, leading to a library of low energy semiconductor defects. |