Hybrid Learning Enables Reproducible >24% Efficiency in Autonomously Fabricated Perovskites Solar Cells

ABSTRACT Achieving high‐performance perovskite solar cells (PSCs) with satisfactory reproducibility remains a major challenge due to their intrinsic susceptibility to processing variations and environmental fluctuations. To address this challenge, this study introduces an autonomous optimization framework that integrates hybrid machine learning and high‐throughput experimentation with modified gradient ascent methods to optimize fabrication processes and minimize experimental variances. The framework successfully maps the complex, non‐linear interdependencies between fabrication parameters and reveals the critical decoupling of photovoltaic metrics. Optimization across seven rounds and 144 parameter sets results in pronounced power conversion efficiency (PCE) and reproducibility enhancement on the platform. The optimized procedure delivers champion devices achieving PCEs exceeding 24%, surpassing the experience manual operator performance (20.6% PCE, CV >25%) and reducing the coefficient of variation (CV) to below 4.7%, with improvements consistently validated across independent trials. This work offers a practical strategy for improving PSC performance and reproducibility, while laying a foundation for scalable manufacturing and accelerated materials development.