A machine learning prediction model for quantitative analyzing the influence of non-radiative voltage loss on non-fullerene organic solar cells

Abstract

The open-circuit voltage (Voc) in organic solar cells (OSCs) hardly exceeds 1.0 V because of the relatively high voltage loss caused by charge non-radiative recombination at the donor–acceptor interface. Herein, in this paper the machine learning (ML) prediction models are used to explore the relationship among the donor and acceptor structures, electronic properties, and the non-radiative voltage loss (△Vocnon-rad). Among the models, the prediction performance from the optimal random forest (RF) model has 13.48% enhancement compared with that of the support vector regression (SVR) model. A combination of correlation and importance is used to collaboratively screen out the key features of acceptor materials with low △Vocnon-rad in OSCs. The importance analysis indicates that the benzene-1,2-diamine, prop-2-en-1-imine and nitrogen sulfur bond are the important structures, which represents the electron-deficient unit (A') in the fused-ring core of non-fullerene acceptors (NFAs). It is worth mentioning that the selected key features also have good applicability in the small data with ternary OSCs, and its coefficient of determination (R2) is 0.704 in the testing set. In addition, the four new Y6 derivatives (Y6O, Y6B, Y18B, and Y18U) are designed by the screened key features. And quantum chemical calculations show that the introduction of benzene ring and branched side chain to the A' unit can make the HOMO and LUMO energy levels of the molecule tend to rise. More importantly, the HOMO-LUMO gap is 2.69 eV and the optical band gap is 1.80 eV in Y18B, which are smaller than those of Y6. Y18B also has the smallest electrostatic potential of 5.08 kcal/mol on the molecular surface. Significantly, it decreases the singlet–triplet energy gap and exciton binding energy of Y18B for effectively reducing the △Vocnon-rad in the device. This work provides an effective model to accelerate the exploration of new and highly efficient NFA-OSCs with the lower △Vocnon-rad.