AI-Driven Design of Dielectric/Ferroelectric Properties and Prediction of Electronic Structures

Abstract:

As integrated circuit semiconductor devices continue to shrink in size, hafnium-based oxides have emerged as a key candidate material for breaking through the performance limits of traditional dielectric/ferroelectric materials, thanks to their high dielectric constant, nanoscale ferroelectricity, and excellent compatibility with CMOS processes. Addressing the computational bottlenecks of traditional density functional theory in simulating material phase transition dynamics and interface effects, this report introduces the innovative application of artificial intelligence technology in hafnium-based oxide material design: first, by constructing a deep learning potential function that balances accuracy and efficiency, it precisely simulates the dielectric response characteristics of the (Hf,Zr)O₂ (HZO) system, overcoming the accuracy limitations of traditional force fields in dynamic polarization and phase stability simulations; further, based on a generalized potential function transfer learning framework, we achieve efficient simulation of ferroelectric phase transition dynamics from pure HfO₂ to multi-element (e.g., Si, La, Zr, etc.) doped systems, revealing the structure-property relationship between doping concentration, lattice distortion, and ferroelectric stability; Finally, based on the Transformer neural network architecture, we developed an atom position embedding Transformer model (APET) to achieve high-precision predictions of the electronic structure of multi-phase HZO systems and other materials, surpassing existing state-of-the-art methods in density of states (DOS) prediction.

This research provides a research tool for the development of high-performance storage and logic devices in the post-Moore's Law era, spanning from atomic-scale simulation to performance optimization.

Bio:

Wei Ren got his bachelor's degree from the national undergraduate program for physics at Shanxi University and his Ph.D. in physics from The University of Hong Kong. After postdoctoral and research assistant professor positions at The Hong Kong University of Science and Technology and the University of Arkansas respectively, he is now a professor at Shanghai University's Materials Genome Institute and Department of Physics. His research interests include computational condensed matter physics, quantum materials informatics, and materials artificial intelligence machine learning. He has published over 200 peer-reviewed papers, with many in top - tier journals like 13 in Physical Review Letters, 60 in Physical Review B, among others in Nature, Science, Advanced Materials, and Nano Letters, etc. He has made significant contributions such as predicting spin-chirality-driven monolayer multiferroicity in 2D vanadium halide, reporting sliding ferroelectricity in bilayer vanadium sulfide and its ferrovalley magnetoelectric coupling, and reporting the coexistence of Dirac and Weyl electron structures in a novel topological semimetal polar crystal material.