Perovskite solar cells demonstrate promise as next-generation solar technology, owing to their high conversion efficiency, lightweight nature, flexibility, and cost-effective manufacturing potential. Since the properties of perovskite solar cells can be adjusted by combining different materials, various compositions of perovskite materials have been fabricated so far. However, exploring this extensive chemical space for highly efficient and stable perovskite materials demands a great deal of time and cost. There are many parameters to be optimized, including device fabrication conditions, and it is generally very difficult to solve optimization problems in such high-dimensional spaces.

In recent years, materials informatics (MI) and process informatics (PI) techniques, have been attracting attention for the design of new materials and process optimization. In the development of perovskite solar cells, optimization of composition and deposition processes using machine learning has also been reported [1], [2]. While machine learning has shown promise in optimizing certain aspects of perovskite solar cells, comprehensive optimization of both composition and deposition processes remains unexplored.

This study focuses on utilizing Bayesian optimization to comprehensively optimize both the composition and deposition process of perovskite materials. Bayesian optimization offers an efficient method capable of navigating high-dimensional spaces with minimal trials, ideal for adjusting multiple parameters efficiently. Specifically, this study aimed to optimize the precursor solution's composition and key deposition process parameters using Bayesian optimization for fabricating CsFAPbI3-based perovskite solar cells. Four parameters were chosen for optimization: lead (II) iodide and methylamine hydrochloride concentrations in the precursor solution, perovskite layer spin-coating rotation speed, and perovskite layer heating temperature. 20 different conditions were used to fabricate perovskite solar cells, which served as initial data. These cycles involved a continuous search for experimental conditions, model evaluation, and updates to maximize conversion efficiency. An automated coating system was employed to minimize performance variations resulting from human factors during the deposition process.

The results indicate a significant improvement in the conversion efficiency of perovskite solar cells, surpassing 20% compared to an initial efficiency of approximately 12%. This study concludes the utility of employing Bayesian optimization through machine learning for the comprehensive optimization of both composition and deposition processes in perovskite solar cell fabrication.