Machine Learning‐Driven Vapor Passivation Enables Efficient Perovskite Solar Cells

ABSTRACT Perovskite solar cells (PSCs) have achieved remarkable power conversion efficiency (PCE) recently, yet the device instability induced by defects remains a bottleneck. Despite the effectiveness of passivation strategies in reducing defects, conventional solution‐based methods may compromise the perovskite crystallinity. Thus, vapor‐based molecules are expected to penetrate into crystals and reduce defects. Screening effective vapor‐assisted passivation materials usually takes substantial trial‐and‐error costs. Here, we propose a hybrid workflow integrating machine learning models with first‐principles calculations to screen the key functional groups for vapor‐assisted passivation molecules and provide physical insights. The phenyl and halogen atoms were identified by Shapley additive explanation (SHAP) analysis to be effective. Theoretical calculations also indicated that there was a strong cation‐ π interaction between Pb 2+ and benzene ring. Besides, halogenated benzene could intercalate into the iodine vacancies on perovskite surfaces. Such robust interactions synergistically contribute to the stability of [PbI 6 ] 4− octahedral framework, and thus the formation of defects and structural collapse could be effectively suppressed. Moreover, the thermal conductivity of modified perovskite films was remarkably enhanced. Finally, the device modified with iodobenzene vapor achieved the champion PCE of 26.94%. This workflow integrating data‐driven discovery and atomistic understandings, provides an innovative strategy to screen vapor‐assisted passivation materials.