AI-guided design of efficient perovskite solar cells operationally stable at 100°C

Operationally stable perovskite solar cells (PSCs) have been sought after and debated since first being demonstrated. Here, we report a four-agent collaborative artificial intelligence (AI) to guide rational design of light absorbers, ultraviolet-resistant hole transport materials, and robust heterointerfaces for stable perovskite photovoltaics. Validated through thermodynamically driven single-crystal growth and thin-film experimental characterizations, the multiagent framework identified a highly stable formamidinium-cesium lead iodide perovskite, FA_0.92Cs_0.08PbI₃. AI-driven insights further enabled the design of a customized hole transport molecule, (4'-(3,6-dimethoxy-9H-carbazol-9-yl)-[1,1'-biphenyl]-4-yl)phosphonic acid, with superior ultraviolet resilience, alongside dual-side metal oxide layer incorporation into the device configuration. The designed PSC can retain 97% of initial efficiency after 1000 hours of continuous operation at 100°C. This success demonstrates an accessible and promising full-chain AI route to accelerate the application of PSCs.