Accelerated discovery of potential ferroelectric perovskite via active learning

Abstract

Ferroelectric materials inherently exhibit great memory effect through piezo- and pyroelectricity, which enables their utilization in many state-of-the-art applications. Here, we demonstrate a novel material screening platform for identifying, via machine learning and active learning, new inorganic ABO₃-type perovskite materials that potentially possess ferroelectric properties. First, the machine learning model for predicting the band gap and formation energy is constructed based on the initial database. Then, an active learning process is implemented to demonstrate its practical applicability to an initial database of less than 10% of the entire chemical space of materials. The proposed platform demonstrates its reliability by identifying already known ferroelectric materials that satisfy the band gap and formation energy criteria. Furthermore, with an exploration of only approximately 30% of the total database, more than 90% of the materials found after the active learning process are satisfactory. This study validates that utilization of machine learning, with optimization, can greatly accelerate the discovery of novel materials.