Machine learning aided photovolatic property predictions, design and library generation of indeno-fluorene donors with lowest exciton bindings

Abstract

A comprehensive study leveraging Machine Learning (ML) and quantum chemical calculations is presented to predict and optimize photovoltaic (PV) properties of Indeno-Fluorene based electron donors. A dataset of 1631 organic donors was compiled from literature and broken down into molecular descriptors using RDKit. The Exciton Binding Energy (E_b) value (eV) were calculated and predicted using ML models, achieving remarkable R² values of 0.88–0.98. Donors with the lowest E_b values were further analyzed for electronic charge patterns, electronic excitations, and PV parameters, including Light Harvesting Efficiency (LHE), Open Circuit Voltage (V_oc), and Short Circuit Current Density (J_sc). The optimized donors exhibited impressive PV performance: LHE (43–95 %), V_oc (−0.42 to 3.90 V), and J_sc (9.42–70.95 mA/cm²). This work demonstrates the efficacy of ML-aided design and prediction of PV properties, facilitating the rapid identification of high-performance Indeno-Fluorene donors for dye-sensitized solar cells. The generated library and predictive models provide valuable resources for the development of efficient organic PV materials.