Machine learning-driven interface material design for high-performance perovskite solar cells with scalability and band-gap universality

Perovskite solar cells’ performance is critically governed by interfacial chemistry. Here, we combine correlation analysis, hierarchical clustering, and least absolute shrinkage and selection operator (LASSO) regression to optimize feature engineering, followed by employing LASSO and elastic net (ENET) regression to develop a screening model. A multifunctional buried interface molecule, (2-(1,3-dioxo-1 H -benzo[ de ]isoquinolin-2(3H)-yl)ethyl)phosphonic acid (BIPA), was successfully screened and synthesized. As expected, BIPA binds strongly with SnO 2 through Sn···O–P coordination, effectively passivating Sn 2+ dangling bonds and oxygen vacancies. The Fermi level was shifted upward, facilitating electron extraction. Besides, BIPA modulated crystallization dynamics and alleviated compressive strain through C=O···Pb interaction, enabling the growth of dense perovskite films. Consequently, the BIPA-modified device achieved an efficiency of 26.3%. Varied band-gap performance improvements (23.2% for 1.67 eV and 18.5% for 1.85 eV) further underscored the universality of this approach. Unencapsulated devices retained 91% and 92% of their initial efficiency after 1,600 h of International Summit on Organic Photovoltaic Stability (ISOS-L-1) protocol and 2,000 h of ISOS-D-2 protocol, respectively.