Machine learning assisted identification of the matched energy level of materials for high open circuit voltage in binary organic solar cells

With the application of new materials and the optimization of device structure, binary bulk heterojunction organic solar cells (OSCs) have exhibited the outstanding performance in recent years. However, the open-circuit voltage ( V oc ) of binary OSCs is normally below 1 V and the matched energy levels of the donor, acceptor and transport materials with high V oc in binary OSCs have been rarely proposed. Herein, four different machine learning (ML) algorithms are applied to investigate V oc in binary OSCs according to the energy level of donor, acceptor and transport materials. Among them, the eXtreme Gradient Boosting (XGBoost) model provides the best prediction ability. Its prediction accuracy and root mean square error reach 0.94 and 0.04, respectively. Therefore, SHapley Additive exPlanations of XGBoost is selected and showed that the highest occupied molecular orbital (HOMO) of the donor plays the most important role for the improvement of V oc in all the energy level of donor, acceptor and transport materials. More importantly the energy level matching strategy of binary OSC materials for high V oc is delivered by machine learning, where the HOMO of the donor is about −5.45 ± 0.1 eV, the lowest unoccupied molecular orbital (LUMO) of the acceptor is about −3.80 ± 0.1 eV, and the work functions of the matched electron and hole transport materials are about −3.6 ± 0.2 eV and −5.1 ± 0.1 eV, respectively. In addition, the experimental verification results display that the measured V oc just has a relatively low error compared with the predicted V oc . Likewise, the predicted V oc based on the XGBoost model of PTB7:PC 71 BM is 0.79 V, and the experimental value is 0.76 V. The relative error is only 3.95%, which indicates the reliability of the ML prediction for high V oc in binary OSCs. The effects of the materials' energy levels on the V oc in binary OSCs are analyzed and the energy level matching strategy of materials for high V oc is delivered by machine learning. Experimental results verify the reliability of this machine learning approach.