New Research Reveals High-Performance Organic Solar Cells Surpassing AI Predictions
Published in the April 2026 issue of Adv. Energy Mater., the study achieves 19.67% PEC with eco-friendly manufacturing.
Abstract Descriptor-based artificial intelligence (AI) has emerged as a paradigm for molecular design in organic solar cells (OSCs); However, it inherently overlooks collective effects governed by bond hybridization, intermolecular coupling, and aggregation thermodynamics. Such effects are encoded at the solution stage, where pre-aggregation of photoactive materials dictates nucleation pathways, phase separation, and molecular ordering during film formation. Herein, we introduce a YBOV non-fullerene acceptor featuring sp2-hybridized branched side chains that exhibit an unprecedentedly strong solution-state pre-aggregation propensity. This behavior translates into highly ordered solid films with a densely packed crystalline microstructure, enabled by a thermodynamically stabilized core–terminal dimer. As a result, incorporation of YBOV into OSCs not only outperforms the benchmark L8-BO-based device, but also confers an effective nucleation seeding-agent function across diverse host OSC platforms, delivering efficiencies of up to 19.67% via green-solvent processing by alleviating the intrinsic current–voltage trade-off. Machine-learning predictions largely match experimental photovoltaic parameters with a slight upward bias, except for open-circuit voltage, which exhibits anomalous behavior driven by pre-aggregation–driven seeding effects beyond descriptor-based AI. This work establishes sp2-hybridized branched side chains as a new molecular design principle, introducing pre-aggregation-enabled seeding effects beyond AI prediction and providing a universal strategy for high-performance OSCs. Researchers from UNIST and Sungkyunkwan University have created a new type of organic solar cell (OSC) that outperforms existing predictions and highlights a hidden factor in device performance. The secret lies in how molecules clump together in solution—a phenomenon that traditional AI models overlook. Professor Changduk Yang from the School of Energy and Chemical Engineering, in collaboration with Professor Doo-Hyun Ko from Sungkyunkwan University, reported a new OSC with a power conversion efficiency (PEC) of 19.67% through a clean, environmentally friendly process. OSCs are lightweight, flexible, and capable of covering large surfaces. They can be integrated into building facades, windows, and wearable devices. The manufacturing process involves dissolving organic materials in solvents and coating them onto substrates—an approach that's simple and scalable. The researchers designed a new molecule, named YBOV, with branched side chains that promote strong pre-aggregation in solution. This aggregation acts like a seed during film formation, guiding molecules to pack more orderly. The result is a crystalline, densely packed active layer that improves charge flow and boosts efficiency. Notably, devices made with YBOV achieved high performance even when produced with eco-friendly ortho-xylene solvent, avoiding toxic chlorinated options. YBOV also proved versatile. When added to different donor materials or used as an acceptor in various blends, it consistently increased device efficiency. Its aggregation behavior enhances performance across a range of formulations. However, this clustering effect escapes prediction by standard AI models. When trained on 750 device measurements, the models underestimated the open-circuit voltage for YBOV-based cells. This shows that AI, which predicts based on molecular structure alone, cannot fully capture the collective behaviors that influence real-world performance. “Our work introduces a new design principle,” said the research team. “It considers how molecules behave in solution—something AI cannot currently predict. Combining this insight with eco-friendly processing opens new paths for commercializing high-performance, sustainable organic solar cells.” Supported by the National Research Foundation of Korea (NRF) and the InnoCore program of the Ministry of Science and ICT (MSIT), the study involved Seokhwan Jeong, Donghoo Won, and Zhe Sun from UNIST who contributed equally. The findings of this research were published in Advanced Energy Materials on April 20, 2026. Journal Reference Seokhwan Jeong, Donghoo Won, Zhe Sun, et al ., “Beyond Descriptor-Based AI Design: Sp2-Hybridized Branched Side Chains Enable Pre-Aggregation–Driven Seeding Effects in Green-Solvent-Processed Organic Solar Cells,” Adv. Energy Mater., (2026).
- 2026-05-27
- JooHyeon Heo
- 251