Machine-Learning-Based Interatomic Potentials for Group IIB to VIA Semiconductors - Towards a Universal Model

Aug 3, 2026

Rapid advancements in machine-learning enable interatomic potentials that retain ab initio accuracy while scaling to large systems. The work addresses the lack of truly universal atomic models by developing a unified deep learning interatomic potential, DPA-Semi, covering 19 group IIB–VIA semiconductors including Si, Ge, SiC, GaN, GaAs, CdTe, and CdS. Training uses density functional theory with numerical atomic orbitals to reduce cost. Comparisons across solid and liquid phases show GGA-level quality, positioning DPA-Semi as a pretrained universal model without parameter retuning.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages32

Reading Time1 h 21 min

AccessFree to download

Introduction

Background: Semiconductors and simulation needs

DFT and its limitations for large-scale industrial systems

Machine-learning interatomic potentials and deep potential models

Method and model scope

Unified DPA-Semi model for group IIB to VIA semiconductors

DFT-based training data and computational setup

Comparative evaluation

Solid and liquid phase property comparisons

Performance assessment vs deep potential baselines

Conclusion

Machine-Learning-Based Interatomic Potentials for Group IIB to VIA Semiconductors: Towards a Universal Model Jianchuan Liu a, 􀁧, Xingchen Zhang b, 􀁧 , Tao Chen c, Yuzhi Zhang d, Duo Zhang d, e, Linfeng Zhang d, f,and Mohan Chen*,b,f a. School of Electrical Engineering and Electronic Information, Xihua University, Chengdu, 610039, P. R. China b.HEDPS, CAPT, College of Engineering, Peking University, Beijing, 100871, P. R. China c HEDPS, CAPT, School of Physics, Peking University, Beijing, 100871, P. R. China d.DP Technology, Beijing 100080, P. R. China e.Center for Machine Learning Research, Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, 100871, P. R. China f. AI for Science Institute, Beijing 100080, P. R. China ∗ [Corresponding author. Email: mohanchen@pku.edu.cn](Corresponding author. Email: mohanchen@pku.edu.cn) Abstract Rapid advancements in machine-learning methods have led to the emergence of machinelearning-based interatomic potentials as a new cutting-edge tool for simulating large systems with ab initio accuracy. Still, the community awaits universal inter-atomic models that can be applied toa wide range of materials without tuning neural network parameters. We develop a unified deeplearning inter-atomic potential (the DPA-Semi model) for 19 semiconductors ranging from group IIB to VIA, including Si, Ge, SiC, BAs, BN, AlN, AlP, AlAs, InP, InAs, InSb, GaN, GaP, GaAs, CdTe, InTe, CdSe, ZnS, and CdS. In addition, independent deep potential models for each semiconductor are prepared for detailed comparison. The training data are obtained by performing density functional theory calculations with numerical atomic orbitals basis sets to reduce the computational costs. We systematically compare various properties of the solid and liquid phases of semiconductors between different machine-learning models. We conclude that the DPA-Semi model achieves GGA exchange-correlation functional quality accuracy and can be regarded as a pretrained model towards a universal model to study group IIB to VIA semiconductors. 1. Introduction Semiconductor materials play a crucial role in the development of modern society. In particular, the IIB to VIA group semiconductors, which are compound semiconductors composed of elements from groups IIB to VIA of the periodic table, possess excellent optoelectronic properties and are widely used in photovoltaic, optoelectronics, thermoelectrics, and other energy conversion fields. 1, 2, 3, 4 For example, silicon carbide (SiC) has found widespread industrial applications because of its excellent wear resistance, corrosion resistance, elevated temperature strength, as well as its high thermal conductivity and wide band gap.5, 6, 7, 8 Boron arsenide (BAs) was initially synthesized in 1958 9 but was recently confirmed to possess high charge carrier mobility and thermal conductivity. 10, 11, 12 Therefore, there is a promising prospect of utilizing this material to alleviate the current bottleneck issue in chip cooling. On the other hand, state-of-the-art simulation tools can complement experiments by elucidating experimental phenomena or predicting experimental outcomes, thereby providing invaluable information or better design principles. Among the simulation tools, the atomistic-level simulation tools can describe interactions of semiconductors from a microscopic perspective, providing valuable insights into the fundamental processes governing the behavior of semiconductor materials. In particular, quantum-mechanicsbased first-principles methods are able to predict the properties of semiconductors without reliance on experimental data. Among them, density functional theory (DFT) 13, 14 is one of the most widely used methods that can predict various properties of semiconductor materials. 15, 16, 17 Taking SiC asan example, it exists in various polytypes such as the 3C, 2H, 4H, and 6H structures, etc. Among them, the 3C structure is a cubic crystal lattice

FAQs

What problem does the DPA-Semi work target for group IIB–VIA semiconductors?

It targets the absence of universal interatomic models that work across many semiconductors without needing neural network parameter tuning.

How is the DPA-Semi model trained to balance accuracy and computational cost?

Training data come from density functional theory calculations using numerical atomic orbitals basis sets to reduce computational costs while maintaining first-principles quality.

What materials and phase types are included in the evaluation?

The unified model covers 19 semiconductors spanning group IIB to VIA, and properties are systematically compared for both solid and liquid phases.

Aug 3, 2026

Upload by Theodore