Bayesian Optimization and Prediction of the Durability of Wide Bandgap Perovskite Thin Films Under Light and Heat Stressors

Wide-bandgap perovskites are top contenders for tandem solar cells, but they suffer from instability when exposed to the necessary operational conditionals of elevated temperature and light. Here, Bayesian optimization (BO) of wide bandgap perovskite compositions is used to optimize the light and heat stability of perovskite films in the triple halide Cs-FA perovskite design space. To improve experimental power, we leverage robotic automation of perovskite thin film deposition and processing, which provides highly reproducible methods for robust comparisons across the composition space. We demonstrate that the BO framework also requires just 40% of the samples needed to extract the same learnings as a conventional grid search method.<br/><br/>From the large body of stability data, we develop a regression model that offers a robust prediction of perovskite stability under light and heat based on the composition and easily measurable film-level optical properties. The model predicts the amount of phase degradation after 700 hours of 1-sun, 85°C illumination with a mean absolute error of 0.5% across the full Cs-FA triple halide space. Furthermore, it links quick and straightforward optical measurements conducted in less than 5 minutes to the 100s of hours long, resource-intensive light and heat tests of ISOS-L-2.<br/><br/>Overall, integration of (1) reproducible automated processing, (2) experiments leveraging Bayesian optimization to learn faster, and (3) machine-learning to predict perovskite stability from early optoelectronic characterization together holds promise to accelerate the development of durable perovskite materials and devices.