In polymer solar cells (PSCs), the electron donor (D) and electron acceptor (A) photoactive layers are the most crucial for the higher photoconversion efficiency (PCE). Non-fullerene acceptors (NFAs) have attracted attention as an alternative for fullerene-based acceptors due to the NFAs' tailor-made characteristics. NFA exhibits tunable optical and electrochemical properties, high solubility, and favorable intermolecular interactions that lead to compact packing in the active DA blend. In designing of high-performance NFA based PSCs, a proper understanding of the energy level alignment is crucial. In this study, we build predictive machine learning (ML) models to identify the features of potential D/A pairs. We constructed five different ML models using 475 D/A combinations where 147 different small molecules based NFAs were used. Before using ML algorithms, Scharber’s model was applied to compute open-circuit voltage (VOC), short-circuit current (JSC) and PCE using the energetic parameters obtained by DFT level calculations. All properties were correlated nicely when the same class of NFAs was used. We have included 13 different microscopic properties like reorganization energies, vertical transitions, etc obtained from quantum chemical calculations as a descriptor in the ML model and structural constituents and functional group descriptors. The best predictive capabilities were obtained by random forest (RF) and ridge regression (RR) approaches in the considered data set. Our learning models will be used to build new D/A pairs from their fragments and added to our data set to use for the virtual screening. The ML model will be used to predict the best D/A pairs based on the predicted PCE.

Machine learning for the high-performance donor/acceptor pairs for non-fullerene based organic solar cells

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.