A Weighted Ensemble Model for Screening Passivation Materials in High-Efficiency Perovskite Solar Cells
The commercialization of perovskite solar cells (PSCs) is hindered by stability issues primarily stemming from interfacial defects. This study employed a machine learning (ML) screening approach and constructed a learnable weighted ensemble model (LWEM) to enhance prediction robustness for identifying effective interface passivation materials. The ML model predicted that an imidazolium salt-based interface modifier, 1-benzyl-3-methylimidazolium tetrafluoroborate (BMT), is suitable for planar n-i-p PSCs. Subsequent experimental results demonstrated that BMT provides synergistic passivation via an 'ion-coordination dual-lock' mechanism that significantly suppresses non-radiative recombination, facilitates hole extraction, and improves the quality of the perovskite film. The BMT-modified devices achieve a significant increase in power conversion efficiency (PCE) from 22.45% to 24.89% under AM 1.5G illumination, and attain a high PCE of 41.31% under 1000 lux light emitting diode (LED) indoor lighting. Additionally, the modified devices exhibit outstanding stability under long-term storage and maximum power point tracking conditions. This work provides a strategy for developing high-performance and highly stable PSCs for both indoor and outdoor applications.