Accelerated composition optimization of hybrid perovskites via data-driven materials design, DFT calculations and synthesis

Hybrid perovskites show promise as low-cost photovoltaics. However, achieving high efficiency necessitates careful optimization of their composition. Traditionally, exploring the vast compositional space is inefficient and time-consuming. This study demonstrates machine learning (ML) techniques to effectively and rapidly map the relationships between perovskite composition and critical band gap energy ( E g ) , leading to the path of finding highly efficient perovskites for solar cell applications. Among these, Random Forest and Gradient Boosting Regression algorithms demonstrated exceptional predictive performance. Compositions with E g values between 1.2 and 2.0 eV were screened for further study. To validate data-driven approach, F A P b ( B r 0.375 I 0.625 ) 3 was selected from the screened data and further studied through Density Functional Theory (DFT) and experimental methods. To the best of our knowledge, no previous studies have analyzed organic-inorganic halide perovskites in such depth, focusing on their crystal structure, stability (offering a new perspective on organic components), and E g . These aspects were thoroughly investigated and validated using ML, DFT, and experimental synthesis with characterization. • ML prediction: Simple ML algorithms can accurately predict perovskite features, even when trained on limited but carefully curated datasets. • Theoretical confirmation: DFT confirmed the predicted crystal structures and E g , providing robust theoretical validation for the ML results. • Experimental validation: The synthesized perovskite properties align closely with ML-predicted and DFT-calculated results, bridging the gap. • Implications for materials design: The alignment between prediction, and synthesis confirms the reliability of ML in designing new materials.