数据驱动与机器学习辅助的铁电材料

As conventional semiconductor technology approaches its fundamental physical limits, ferroelectric materials, with their electrically switchable spontaneous polarization, have emerged as promising candidates for next-generation information storage and neuromorphic computing devices. However, optimizing material performance and elucidating the underlying physical mechanisms remain central challenges in this field. Rapid advances in machine learning and data science are driving a paradigm shift in ferroelectric materials research, transitioning from empirical trial-and-error approaches toward an intelligent, data-driven framework. This review systematically summarizes recent progress in applying machine learning to ferroelectric materials research. Through high-throughput screening, structure-property relationship modeling, and physics-informed approaches, machine learning has significantly accelerated the discovery and performance optimization of novel ferroelectric materials, enabling systematic breakthroughs ranging from compositional design to crystal structure prediction. Furthermore, by integrating machine learning interatomic potentials with physics-enhanced phase-field models, multiscale computational simulations have effectively revealed the microscopic mechanisms governing ferroelectric phase transitions and domain structure evolution, thereby bridging the knowledge gap between atomistic-scale simulations and macroscopic properties. Finally, this article discusses potential pathways for the deep integration of data-driven methods with physical models and provides an outlook on the future direction of intelligent research and development for functional materials.