About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI/ML/Data Informatics for Materials Discovery: Bridging Experiment, Theory, and Modeling
|
| Presentation Title |
A “DFT-ML” Framework for Rational Design of Next-Generation Semiconductors |
| Author(s) |
Maitreyo Biswas, Arun Kumar Mannodi Kanakkithodi, Habibur Rahman, Rushik Desai |
| On-Site Speaker (Planned) |
Maitreyo Biswas |
| Abstract Scope |
We developed a rational virtual materials design strategy powered by high-throughput density functional theory (DFT) computations and surrogate machine learning (ML) models to perform multi-objective optimization and discovery of novel crystalline semiconductors. This strategy involved compiling massive DFT datasets of relevant properties within a multi-fidelity active learning framework and training predictive models to accurately obtain any property of interest as well as synthesis likelihood directly from the semiconductor composition or structure. “DFT-ML” models were trained for both bulk and defect properties, especially using graph neural network (GNN)-based interatomic potentials. The positive and unlabeled learning approach was applied to obtain synthesis probability scores by training on a combined dataset of DFT computations and known experimental results from the literature. Using these “DFT-ML” strategies, we successfully designed novel stable and synthesizable halide and chalcogenide compounds with suitable electronic, optical, and defect properties for a variety of optoelectronic applications. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Modeling and Simulation, Machine Learning, Computational Materials Science & Engineering |