HOME > Profile > LU, Anh Khoa Augustin
External affiliations
- The University of Tokyo
Research
- Keywords
First-principles calculations, disordered materials, machine learning, charge-aware neural networks
First-principles calculations, graph neural networks, machine learning, two-dimensional materials, semiconductors.
PublicationsNIMS affiliated publications since 2004.
Society memberships
The Physical Society of Japan, The Japan Society of Vacuum and Surface Science
Awards
- Best Poster Prize at 788. WE Hereaeus Seminar (Bad Honnef, Germany) (2023)
Funds
- 科研費 基盤研究(C) (2026)
Research Center for Materials Nanoarchitectonics (MANA)
Title
Large-scale atomistic simulations of semiconductor interfaces
Keywords
Semiconductors, Graph neural networks, Unsupervised learning, Density Functional theory (DFT), Electric field
Overview
The development of semiconductor devices is driven by our ability to engineer their structure at the atomic scale and control their electronic properties. Recent advances in atomistic modeling have made large-scale simulations possible, allowing us to model imperfections that shape the properties of semiconductor interfaces, such as defects, impurities or charge transfer. This research combines three methods to develop this knowledge, namely large-scale linear-scaling density functional theory (DFT) calculations to calculate the electronic properties, charge-aware graph neural networks to perform simulations under an applied electric field, and unsupervised machine learning methods to analyze the resulting atomic structures. The combination of these approaches will unlock new materials and devices.
Novelty and originality
• Accurate predictions of dynamical charges (Born effective charges) with equivariant graph neural networks
• Large scale molecular dynamics simulations under electric field with charge-aware graph neural networks
• Structure analysis towards the understanding of the medium-range order in complex disordered materials
Details
Machine learning potentials have emerged as a cost-effective alternative to first-principles calculations with high accuracy and high simulation speed. However, atomic charges are typically neglected, thus limiting their field of applications when an electric field is applied, which is essential in semiconductor devices. Several methods have been proposed to describe dynamical charges, but they typically have to compromise between simulation speed and accuracy. By developing an optimized architecture for message-passing graph neural network with SevenNet-Polar, we train highly accurate force field while ensuring efficient scalability, enabling molecular dynamics simulations under electric field with millions of atoms.
Disordered structures and interface structures often require large cells to be representative of real systems. This makes their analysis challenging. By developing appropriate descriptors and unsupervised learning methods such as two-step locality preserving projections (TS-LPP), it becomes possible to detect subtle structural differences within a sample, without the need of prior knowledge. We are actively optimizing these methods to make it applicable to large complex systems.
Summary
• Development of highly scalable message-passing graph neural networks capable of predicting dynamical charges
• Development of unsupervised machine learning methods and descriptors to capture subtle structural differences in disordered materials
この機能は所内限定です。
この機能は所内限定です。



