Data-mining process for ferroelectric materials with symmetry-breaking phase transformations by Nature Language Process (NLP). Credit: HKUST

Researchers from the Hong Kong University of Science and Technology (HKUST) and Tongji University have developed FerroAI, a deep learning model that can produce phase diagrams for ferroelectric materials in just 20 seconds.

Using FerroAI, the team discovered a novel ferroelectric material with an exceptionally high dielectric constant of 11,051, representing a significant advancement in AI-driven ferroelectric materials research. Their findings, published in the journal npj Computational Materials, offer an innovative approach to accelerating the design and discovery of functional materials.

This study, titled “FerroAI: a deep learning model for predicting phase diagrams of ferroelectric materials,” was jointly conducted by Prof. Sherry Chen Xian (co-corresponding author), Associate Professor of the Department of Mechanical and Aerospace Engineering, HKUST, and Dr. Zhang Chenbo (first author and co-corresponding author), Researcher of the School of Physics Science and Engineering and MOE Key Laboratory of Advanced Micro-Structured Materials, Tongji University, and 2019 Ph.D. graduate of HKUST.

Ferroelectric materials are extensively utilized in sensors, memory devices, and energy harvesting applications due to their distinctive electrical properties. Their performance is critically dependent on crystal structures, whose evolution is typically characterized through composition–temperature phase diagrams.

Demonstration of user-interface (UI) on how to predict the phase diagram with a given chemical compound by FerroAI. Credit: npj Computational Materials (2025). DOI: 10.1038/s41524-025-01778-0

Conventional approaches to constructing these diagrams demand the synthesis of numerous samples, followed by extensive experimental testing and computational analysis. Such a process usually spans months or years. More efficient strategies for phase diagram construction and materials prediction underlie the horizon of basic research in the ferroelectric community.

Previous machine learning approaches could predict phase transitions for specific material systems but lacked sufficient generalization capability to accurately capture latent features across diverse material families. To address this limitation, the research team extracted data from over 40,000 published papers to construct a comprehensive phase-transformation database encompassing thousands of ferroelectric systems. Through advanced data augmentation techniques and intelligent parameter optimization strategies, the team developed and trained the deep learning model FerroAI.

Experimental validation demonstrated that FerroAI achieves over 80% prediction accuracy across multiple crystal structures while providing insights into the critical roles of various dopant elements at specific lattice sites. Remarkably, tasks that traditionally require months of materials development work can now be completed by AI in just 20 seconds.

Statistics of classified ferroelectric phase transformation dataset. Credit: HKUST

Prof. Chen said, “This study demonstrates the importance of high-quality datasets. In addition, the success of phase diagram prediction suggests that the research paradigm of materials science is being reshaped by AI, which promotes the rapid discovery of new functional materials.”

This achievement represents important progress of the team in the field of intelligent materials. For the first time, a tight integration of phase transition prediction and experimental validation across material systems has been achieved. This showcases the huge potential of AI in accelerating both basic science and applied research.

More information: Chenbo Zhang et al, FerroAI: a deep learning model for predicting phase diagrams of ferroelectric materials, npj Computational Materials (2025). DOI: 10.1038/s41524-025-01778-0

Citation: Phase diagram map of ferroelectric properties unlocked with AI in seconds (2025, October 31) retrieved 31 October 2025 from https://phys.org/news/2025-10-phase-diagram-ferroelectric-properties-ai.html

This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.