Optimization of defect regulation parameters in CsPbI3 perovskite solar cells via machine learning-assisted response surface methodology

All-inorganic CsPbI3 perovskite solar cells (PSCs) have shown significant progress in power conversion efficiency (PCE). However, defects at grain boundaries and interfaces still limit their performance and stability. Traditional optimization methods, which focus on individual parameters, are time- and energy-consuming and often fail to achieve optimal multi-parameter performance. To address this challenge, we develop machine learning assisted response surface methodology (ML-assisted RSM) to optimize defect regulation parameters. The reliability and accuracy of the RSM-driven optimization can be validated by ML. Furan-2,5-dicarboxylic acid (FDCA) was found to be an effective interfacial defect regulation material on the CsPbI3 perovskite surface. Using FDCA as the model material to regulate defect, ML-assisted RSM identified optimal conditions (14.26 mM, 103 °C, 9 min), achieving a high PCE of 18.71 %, which aligned closely with the predicted value of 18.68 %. FDCA treatment, which contributed to the “defect healing” effect, improved perovskite quality and reduced non-radiative recombination by interacting -COOH bidentate functional groups with uncoordinated Pb2+ defects. This study demonstrates the value of ML-assisted RSM for fast and accurate multi-parameter optimization in PSCs, accelerating the development of high-performance and stable PSCs, and providing a robust framework for optimizing other photovoltaic devices.