Machine Learning-assisted Multiobjective Optimization of Tin-based Perovskite Polycrystals

Thin film transistors (TFTs) possess a structural distinction from conventional silicon-based chips. Tin-based perovskites exhibit outstanding optoelectronic properties, convenient processing methods, and cost-effective production. Notably, their nearly negligible ion migration and high carrier mobility at room temperature position them as excellent candidate materials for TFTs. Nevertheless, reproducing high-quality tin-based films remains challenging due to the stringent preparation conditions and process parameters involved. We incorporated machine learning to aid in the optimization of film properties, utilizing Bayesian optimization to adjust HyperParameters. This approach maximizes the utilization of experimental data, narrowing the solution space and enabling simultaneous optimization of multiple indicators. Consequently, we successfully prepared high-quality all-inorganic tin-based perovskite films within a short experimental period. The crystal morphology and photoelectric properties of the films exhibited significant improvements compared to manually optimized results. This work offers a novel and efficient optimization strategy for exploring high-quality new semiconductor films.