• • The framework leverages over 20,000 experimentally measured PSC samples and approximately 260 multi-scale features, enabling a highly generalizable performance prediction model with prediction error comparable to small-scale models, directly addressing the industrial bottleneck of data scarcity and poor model transferability across diverse device architectures.
• • Experimental validation on 8 types of PSCs with new feature values yielded an additional PCE improvement of 0.92% to 2.43% (absolute percentage points) over baseline fabricated devices, demonstrating the framework's capacity to generate actionable optimization suggestions for novel material compositions and device stacks.
• • The full-process AI framework integrates material selection, fabrication processes, and environmental factors (illumination, temperature, wind speed) into a unified optimization solution, overcoming the device-level oversimplification of prior ML models that failed to capture multi-scale complexity in PSC design.
• • The framework's expansion module successfully adapted to new data with sufficient learning and analytical capabilities, proving its universality and ability to keep pace with the rapid development of PSCs, thereby reducing the need for costly and time-consuming traditional experimental optimization or DFT calculations.