Machine Learning for Screening Small Molecules as Passivation Materials for Enhanced Perovskite Solar Cells

Utilization of small molecules as passivation materials for perovskite solar cells (PSCs) has gained significant attention recently, with hundreds of small molecules demonstrating passivation effects. In this study, a high-accuracy machine learning model is established to identify the dominant molecular traits influencing passivation and efficiently screen excellent passivation materials among small molecules. To address the challenge of limited available dataset, a novel evaluation method called random-extracted and recoverable cross-validation (RE-RCV) is proposed, which ensures more precise model evaluation with reduced error. Among 31 examined features, dipole moment is identified, hydrogen bond acceptor count, and HOMO-LUMO gap as significant traits affecting passivation, offering valuable guidance for the selection of passivation molecules. The predictions are experimentally validate with three representative molecules: 4-aminobenzenesulfonamide, 4-Chloro-2-hydroxy-5-sulfamoylbenzoic acid, and Phenolsulfonphthalein, which exhibit capability to increase absolute efficiency values by over 2%, with a champion efficiency of 25.41%. This highlights its potential to expedite advancements in PSCs. A high-accuracy machine learning model is established to efficiently screen effective passivation small molecules, where random-extracted and recoverable cross-validation is introduced to enhance the model evaluation accuracy. This facilitated the identification of dominant molecular traits influencing passivation effects and the screening of excellent passivation materials. The consistency between predictions and experimental results confirmed the reliability of the machine learning model. image