Machine learning assisted identification of homobilayer sliding ferroelectrics with large out-of-plane polarization and low sliding energy barriers
Xian Wang, Yifan Li, Ying Zhang, Peng Wang, Yi-Ming Zhao, Jun Zhou, Jie Yang, Xuesen Wang, Lei Shen
Abstract
Two-dimensional (2D) ferroelectric materials hold great promise for ultrahigh-density data storage and ultrafast memory applications in flexible or wearable electronic devices. Because of depolarizing field effects and high-energy barriers for switching, only a handful of 2D ferroelectrics have been reported experimentally, such as ${\mathrm{In}}_{2}{\mathrm{Se}}_{3}$ and ${\mathrm{CuInP}}_{2}{\mathrm{S}}_{6}$. Through high-throughput calculations we identified a catalog of 25 synthesizable sliding ferroelectrics with large out-of-plane polarization and low sliding energy barriers. The high-throughput process started with 6351 monolayer materials in 2dmatpedia, narrowed down to 79 monolayer semiconductors with honeycomb structures, leading to the construction of 474 homobilayers sliding ferroelectrics, and ultimately identifying 25 high-performance 2D sliding ferroelectrics. Using big-data analysis and machine learning, we also revealed strong correlations between polarization and key physical properties, especially a hidden factor of effective van der Waals radius, which has been overlooked in previous fewer-sample studies. Based on these mechanistic insights, we proposed machine learning descriptors for predicting sliding ferroelectric properties.