Automated Perovskite Film Fabrication with High-Resolution Data Acquisition and Machine Learning Analysis
Ryosuke Kado, Fumiya Iwai, Ryosuke Nishikubo, Fumitaka Ishiwari, Akinori SaekiAbstract
Lead halide perovskite solar cells have emerged as promising candidates for next-generation photovoltaics; nevertheless, fabrication of the perovskite photoactive layer remains highly intricate with film formation strongly modulated by various processing and environmental factors. Here, we report an environment-controlled, robot-assisted platform for automated thin-film fabrication designed to quantify and visualize perovskite crystallization in situ. Beyond precise control of each processing step, the system captures dynamic phenomena spanning subsecond color evolution and millisecond antisolvent droplet behavior while simultaneously recording slower environmental parameters (temperature, humidity/moisture, and concentrations of oxygen and volatile organic compounds). For optoelectronic assessment, we implemented time-resolved microwave conductivity (TRMC) measurements together with photoluminescence intensity quantification by using a complementary automated evaluation system. By combining machine-learning regression with SHAP (Shapley additive explanations) analysis, we identify the environmental parameters that most impact film properties and propose a feedback strategy to refine data and extract precise effects of process factors. A comparative study of a two-cation perovskite, (FAPbI3)0.87(MAPbBr3)0.13, and a three-cation perovskite, FA0.85MA0.10Cs0.05PbI3, further reveals markedly different sensitivities to environmental factors (FA, formamidinium; MA, methylammonium), in particular, temperature and resultant phase transition kinetics during thermal annealing. Based on the ML analysis and feedback, we tuned the antisolvent droplet velocity and obtained the maximized power conversion efficiency of 24.4%. Notably, this manual device fabrication was consistent with the results obtained by the automated system. This work establishes a practical route to stable, quantitative, and mechanistically informed investigation of solution-processed photovoltaic films.