In recent years, encouraging progress in non-fullerene acceptors-based organic solar cells (NFAs-OSCs) has been made in the field of clean-energy technologies. Nevertheless, achieving a high fill factor (FF) for NFAs-OSCs is a great challenge due to the FF value can be dramatically affected by various factors. An accurate prediction (based on empirical data) of how inherent characteristics of polymer-NFA combinations (e.g., frontier molecular orbitals (FMO) and charge-carrier mobilities) affects the FF in NFAs-OSCs is therefore urgently needed for the future commercialization prospect. The presented work demonstrates an outperforming predictive model using the optimized Gradient-boosting decision tree (GBDT) machine learning (ML) algorithm to accurately predict the FF value, which is based on a dataset consisting of > 180 unique donor/NFA blends with reported FF from previously published publications and experimental physical descriptors (FMO and charge-carrier mobilities). In addition, the SHapley Additive exPlanations (SHAP) strategy is further applied to not only gain the interpretability of the GBDT model prediction but also directly map the relationship between the FF value and experimental physical descriptors by systematic interpretable explanations (e.g., global and local). The SHAP analysis qualitatively reveals that the highest occupied molecular orbital (HOMO) of the donor and the HOMO of the acceptor are the two most important features associated with FF value prediction, which is in good agreement with previous experimental facts. Furthermore, the ML strategies explore the key requirements (e.g., small HOMO offset (<0.3 eV), high hole and electron mobilities (>1.0 × 10–4 cm2 V−1 s−1), and desired μh/μe ratio (in the range of 1.8–3.3)) for enhancing the FF values of NFAs-OSCs. This study provides novel insights into the possible physical mechanisms of FF enhancement, thus facilitating research efforts (e.g., rational design of energy level tuning in donor/NFA blend) for high-performance NFAs-OSCs.
Read full abstract