A Full-Process Artificial Intelligence Framework for Perovskite Solar Cells
The development of high-efficiency perovskite solar cells (PSCs) demands comprehensive control of multi-scale factors influencing device performance. Artificial intelligence (AI), represented by machine learning (ML), has rapidly become a key tool for PSC design and optimization. However, current ML models often oversimplify PSC design at the device level, failing to capture multi-scale complexity. They are constrained by relatively small, specialized datasets, limiting generalizability across diverse architectures and fabrication methods. This work developed a full-process AI framework based on over 20,000 experimentally measured PSC samples and approximately 260 multi-scale features. The framework offers significant advantages in sample diversity and feature richness, combining material selection, fabrication processes, and environmental factors to provide accurate, comprehensive optimization solutions. Data diversity and heterogeneity challenges were addressed through feature engineering and model training, yielding a highly generalizable PSC performance prediction model with prediction error comparable to small-scale models. The framework enables precise optimization of specific features for any PSC and provides valuable insights for designing high-performance photovoltaic devices. Experimental validation fabricated 8 types of PSCs with new feature values; the framework searched corresponding optimization suggestions, resulting in an additional improvement of 0.92% to 2.43% in final power conversion efficiency (PCE) of fabricated devices. This demonstrates the framework's universality and sufficient learning and analytical capabilities for new data, adaptable to the rapid development of PSCs.