Molecular structural descriptor‐assisted machine learning for organic photovoltaics with perylenediimide acceptors

다국어 초록 (Multilingual Abstract)

Although organic photovoltaics (OPVs) have evolved over the last two decades, the discovery of new materials and optimization of numerous considerations for high‐performance devices remain challenging. To reduce these laborious processes and expedite the advancement of OPVs, we constructed machine learning (ML) models that predict photovoltaic parameters. We designed a unique descriptor that divides the molecular structure into smaller units and translates them into a concise matrix. This allows the ML model to easily track structural units and understand which units are important for predicting target performance, enabling the ML model to prioritize crucial units. Therefore, without requiring additional data from measurements or calculations, the ML models can extract chemical properties from molecular structural information and accurately predict the photovoltaic parameters. The ML models that predict the photovoltaic parameters, including the open‐circuit voltage, short‐circuit current density, fill factor, and power conversion efficiency, all show remarkably superior prediction performance, with Pearson correlation coefficients exceeding 0.68. Consequently, in this article, we propose a highly precise and reliable predictive OPV‐ML platform that can robustly screen for unnecessary experiments and accelerate OPV development.

다국어 초록 (Multilingual Abstract)

Although organic photovoltaics (OPVs) have evolved over the last two decades, the discovery of new materials and optimization of numerous considerations for high-performance devices remain challenging. To reduce these laborious processes and expedite the advancement of OPVs, we constructed machine learning (ML) models that predict photovoltaic parameters. We designed a unique descriptor that divides the molecular structure into smaller units and translates them into a concise matrix. This allows the ML model to easily track structural units and understand which units are important for predicting target performance, enabling the ML model to prioritize crucial units. Therefore, without requiring additional data from measurements or calculations, the ML models can extract chemical properties from molecular structural information and accurately predict the photovoltaic parameters. The ML models that predict the photovoltaic parameters, including the open-circuit voltage, shortcircuit current density, fill factor, and power conversion efficiency, all show remarkably superior prediction performance, with Pearson correlation coefficients exceeding 0.68. Consequently, in this article, we propose a highly precise and reliable predictive OPV-ML platform that can robustly screen for unnecessary experiments and accelerate OPV development.