Disentangling Environmental Effects on Perovskite Solar Cell Performance via Interpretable Machine Learning

Despite the rapid increase in efficiency of perovskite photovoltaics, poor reproducibility remains a barrier to their commercialization. Film processing and device performance are highly sensitive to environmental factors during fabrication, yet these interactions are not well understood. Here, we present a systematic methodology to investigate the direct and coupled effects of environmental variables on the perovskite solar cell performance. To do this, we developed an integrated fabrication platform to independently control solvent partial pressure, absolute humidity, and temperature during spin-coating and thermal-annealing of perovskite films and combined it with a closed-loop Bayesian optimization framework to efficiently explore the multidimensional processing space. Efficiency maps reveal coupled nonlinear effects of these variables on device performance, validated by in situ structural characterization, which showed that humidity–solvent interactions affect film crystallization. To overcome the limitations of conventional SHapley Additive exPlanations in disentangling strongly coupled variables, we distilled the knowledge of a Gaussian teacher regressor into multiple student models within an interpretable machine learning framework that employs Shapley interaction analysis to decipher these coupled interactions. This study demonstrates active learning with interpretable machine learning as a powerful tool to explore complex processing landscapes and highlights the importance of environmental control for robust and generalizable processing protocols to accelerate scalable, high-performance, and reproducible perovskite solar cell manufacturing.