Investigation of Short‐Circuit Current Density in Non‐Fullerene‐Based Ternary Organic Solar Cells by Incorporating Machine Learning Algorithms with Effective Descriptors

Non‐fullerene acceptor (NFA)‐based ternary organic solar cells (OSCs) are emerging as promising devices for converting sunlight into electricity, contributing to environmental solutions. However, selecting the third component remains a significant challenge, as it plays a critical role in achieving high short‐circuit current density ( J sc ) in NFA‐based ternary OSCs (comprising donors, acceptors, and the third component). Traditional trial‐and‐error experimental methods face substantial limitations, including high energy consumption, cost, and time demands, which may not be sufficient for investigating the quantitative relationships between material properties and J sc in ternary OSCs. In this study, we examine the effects of the highest occupied molecular orbital–lowest unoccupied molecular orbital (HOMO–LUMO) energy gap (ΔHOMO and ΔLUMO) between different organic materials, considering these as effective molecular descriptors, on the primary photovoltaic parameter ( J sc ) in NFA‐based ternary OSCs. The eXtreme Gradient Boosting (XGBoost) algorithm yields reasonable predictions, with an R 2 value of 0.76. Additionally, three NFA‐based ternary OSCs are fabricated and characterized experimentally to validate the predictions made by the proposed model. Using three different NFA‐based ternary OSCs as inputs, the model demonstrates good predictive accuracy for J sc values. The proposed interpretable model and effective molecular descriptors offer a practical machine‐learning approach for accelerating the development of NFA‐based ternary OSCs with targeted J sc values and can also be extended to other organic electronic applications.