Advancing the Design of High-Efficiency Printable Hole-Conductor-Free Mesoscopic Perovskite Solar Cells Through Machine Learning

Breaking through the power conversion efficiency (PCE) limits of printable mesoscopic perovskite solar cells (p-MPSCs) with machine learning (ML) shows great potential, but has not yet been accomplished. This work establishes a reliable workflow by constructing a high-quality p-MPSCs database for ML model development, followed by strategy formulation for achieving high-performance p-MPSCs. In the 8 validation experiments, the stacking ML model demonstrates excellent performance, with the prediction error not exceeding 2.16%. Model interpretability analysis reveals key factors influencing device performance and enables the formulation of screening rules for high-quality precursor additives based on molecular fingerprinting. This validated framework guides the experimental realization of p-MPSCs with a notable PCE of 19.36%, while theoretical projections suggest a maximum achievable efficiency of 24.32% through optimized design space exploration. A novel paradigm for accelerated discovery of p-MPSCs is established through the synergistic integration of interpretable ML models and targeted experimental validation.