Machine Learning‐Driven Molecule Additives for High‐Performance Perovskite Solar Cells with Bandgap Universality and Scalability

ABSTRACT Molecule passivation materials are critical for high‐performance perovskite solar cells (PSCs). However, experimental identification of effective additives remains costly and time‐intensive, while standard machine learning (ML) struggles with accurate predictions. This study employs an integrated analytical pipeline, combining correlation analysis and hierarchical clustering with LASSO regression for feature engineering optimization, followed by the application of LASSO and ENET regression to construct a predictive screening model. This not only enables efficient identification of key molecular descriptors and fingerprint features—in conjunction with correlation and clustering analyses, but also minimizes model complexity and counteracts overfitting from limited data, thereby significantly improving its predictive capability. The predicted molecule (DCNP) interacts with perovskite through multiple interactions, suppressed the nonradiative recombination, improved perovskite crystallinity, and optimized the band alignment. Consequently, the efficiency of DCNP‐treated device was enhanced from 25.21% to 26.64% (certified: 26.08%). Perovskite module (5 × 5 cm 2 ), 1.67 eV‐ and 1.85 eV‐PSCs with PCE of 22.02%, 23.13%, and 18.65% were successfully fabricated, confirming the advantage in large‐area and universality across varied bandgaps. Moreover, DCNP‐treated PSCs retain 90% of the initial PCE after 1600 h under continuous illumination (ISOS‐L‐1 protocol). This highlights great potential to accurately predict passivation molecules and accelerate the advancements in PSCs.