Machine learning for perovskite solar cell design

As representatives of third-generation solar cells, perovskite solar cells (PSCs) have experienced rapid development. Suffering from inefficient traditional trial-and-error methods and huge search space, discovering superior performance of perovskite materials and high conversion efficiency and stability of PSCs is still a challenge. With the increased computational power and the establishment of large databases, data-driven machine learning (ML) is rapidly gaining momentum in the materials field. ML can predict the properties of potential perovskite materials as well as provide additional physical understanding to accelerate the advancement of PSCs. In this review, we first outline the basic steps and methods of ML. Then, we focus on recent advances in ML for perovskite property predictions and candidates screening, and research to find conditions for higher efficiency or stability in PSCs. We also analyzed the understanding provided by the ML approach and the relationship between the descriptors and the target properties. In addition, we summarize comments and opinions and discuss the current challenges and future opportunities in the field.