Optimization and Scalability of Polymer-Modified PSCs Investigated by Machine Learning

Advanced engineering strategies are employed to optimize the performance of perovskite solar cells (PSCs). In this study, the introduction of polyvinylpyrrolidone (PVP) to the MAPbI₃ perovskite precursor results in PSCs presenting self-healing ability in a moisture environment and power conversion efficiency (PCE) of up to 20.35%. We utilize machine learning to correlate comprehensive J–V experimental data with corresponding photovoltaic parameters. We identify key factors and correlations of J_sc, FF, and V_oc that primarily influence the PCE and scalability of polymer-modified PSCs. The findings indicated that the correlation between PCE and active area (AE) drops from 40% in reference cells to approximately 1% in the modified cells with PVP, justifying the scale-up potential of the modified approach. This is not the case for untreated devices, where PCE is largely affected by shunt (R_sh) and series (R_s) resistances. We evaluated 25 different algorithms through cross-validation, with the Gaussian Process emerging as the best-performing model, achieving an R² of 0.94 and minimal errors. This model/algorithm was applied to optimize the fabrication process by predicting the optimal amount of PVP, which was determined to be 4.5 mg/L, and predicting the corresponding current–voltage (J–V) characteristics as well. This study offers a robust framework for systematically designing and optimizing durable and scalable polymer-modified PSCs, advancing the field of third-generation photovoltaic technology.