Accelerated exploration of efficient ternary solar cells with PTB7:PCBM:SMPV1 using machine-learning methods.

Machine learning (ML) provides an efficient tool for predicting the photoelectric conversion efficiency (PCE) of organic solar cells (OSCs). In this paper, random forest (RF), K-nearest neighbors, and support vector machine are used to predict the PCE for ternary OSCs with PCBM. The results of ML show that RF has the best PCE prediction accuracy. Therefore, RF is chosen to predict the champion PCE of ternary OSCs with PTB7:PCBM:SMPV1, which is around 8.01% in ternary OSCs with a doping ratio of around 6 wt% of SMPV1. To check the prediction, ternary OSCs with PTB7:PCBM:SMPV1 were fabricated, and the experimental results show that the best PCE of 8.83% is obtained in ternary OSCs with 7.5 wt% of SMPV1 introduced. The experiments verify the feasibility of ML in predicting the PCE of ternary OSCs, and its great potential in predicting the doping concentration of the third component for ternary OSCs. Moreover, the working mechanism of the performance enhancement in the ternary OSCs is further researched and demonstrated as the following: (i) an increase in photon capture in the visible light spectrum to enhance the short circuit current density (); (ii) high priority charge transport to boost the fill factor and .