About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
2D Materials – Preparation, Properties, Modeling & Applications
|
| Presentation Title |
Machine Learning Interatomic Potentials for Graphene and Silicon/Graphene Interface Systems |
| Author(s) |
Rumana Hasan, Dibakar Datta |
| On-Site Speaker (Planned) |
Rumana Hasan |
| Abstract Scope |
Silicon/graphene heterostructures hold great promise for electronics, optoelectronics, and energy storage by leveraging graphene’s high conductivity alongside silicon semiconductor. However, accurate atomistic modeling of such interfaces presents significant computational challenges. Machine learning interatomic potentials (MLIPs) offer an efficient and accurate alternative to quantum mechanical methods for large-scale simulations. Herein, we develop MLIPs for graphene and silicon/graphene heterostructures using high-dimensional neural networks (HDNNs) trained on DFT-based ab initio molecular dynamics datasets with atom-centered symmetry function (ACSF) descriptors. For silicon/graphene interfaces, including pristine, bilayer, and defective graphene with amorphous silicon, the HDNN achieved RMSEs of 1.95 meV/atom (energy) and 0.41 eV/Å (force). The graphene-only model achieved even lower RMSEs: 1.73 meV/atom and 0.21 eV/Å. These MLIPs enable energy and force predictions several orders of magnitude faster than DFT, offering a scalable framework for simulating interface dynamics and accelerating materials discovery. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Modeling and Simulation, Thin Films and Interfaces |