Strategy to Screen Donor and Acceptor Pairs for Organic Solar Cells Through Machine Learning

Machine learning (ML) has been utilized in this study to screen optimal donor and acceptor counterparts of solar cell molecules based on the device efficiency. Almost 42 ML models are tested, among which random forest (RF) regression, light gradient boosting machine (LGBM), and Nu support vector regression (NuSVR) models appear to be the best models. The solar cell device performance defined by the device properties, i.e., photoconversion efficiency (PCE (%)), short‐circuit current ( J _sc (mA/cm ² )), open‐circuit voltage ( V _oc (V)), and donor and acceptor molecule's charge transfer (Δ N ) are predicted using best ML model selected based on its high R ² . Suitable resemblance of predicted and actual values is found for all those properties. Chemical reactivity parameters of donor and acceptor molecules have been utilized to screen the best donor and acceptor molecules based on their PCE (%) values. Synthetic accessibility assessment has also been considered as one of the parameters in the optimization process that signifies the ease of synthesis of the donor and acceptor molecules. This strategic ML framework is able to find the efficient donor and acceptor counterparts based on its chemical stability that directly influence solar cell performance, in short time and in an efficient way.