Multi-objective AI-driven optimization guides the discovery of high-performance organic photovoltaics

Experience-driven optimization has been pivotal in advancing organic photovoltaics, where chemically intuitive and stepwise tuning of processing parameters can yield high-performance organic solar cells (OSCs). However, systematically locating optimal regions under high-dimensional, strongly coupled, and multi-objective constraints remains challenging across diverse scenarios. Here, we introduce a closed-loop workflow base on multi-objective Bayesian optimization for efficient process optimization of OSCs. Applied to a quaternary system comprising the widely used polymer donors PM6 and D18 and the non-fullerene acceptors BTP-eC9 and L8BO, the workflow navigates an eight-dimensional space jointly defined by composition and fabrication variables, encompassing 2.2 × 10 ¹⁴ possible combinations. Within five active-learning cycles, the search rapidly converges to a high-performance region, achieving a power conversion efficiency exceeding 20% while drastically reducing the time and effort compared to conventional trial-and-error approaches. The generalizability of the workflow is further demonstrated across different material systems and fabrication methods, consistently guiding the optimization toward high-performance regions in only a few iterations. This work establishes a generalizable, efficient, and scalable closed-loop framework for the rapid identification of optimal process windows in high-dimensional, multi-objective systems for organic photovoltaics and beyond.