Benzothiophene semiconductor polymer design by machine learning with low exciton binding energy: A vast chemical space generation for new structures

Abstract

The development of new organic semiconductors with low exciton binding energies (Eb) is crucial for improving the efficiency of organic photovoltaic (PV) devices. Here, we report the generation of a chemical space of benzothiophene (BDT)-based organic semiconductors with lowest Eb energies using machine learning (ML). Our study involves the design of over 500 organic semiconductor structures with low Eb energies and their synthetic accessibility scores. For this, we collect 1061 BDT based compounds from literature, calculated their Eb energies, and predicted them using ML with Random Forest (RF) regression, yielding the best results. Our analysis, using SHAP values, reveals that heavy atoms are the main factors in lowering Eb values. Furthermore, we tested new organic chromophore structures, which showed an efficient shift of their molecular charges. The UV-Vis spectra of these structures exhibits a redshift in the range of 358-667 nm, while their open-circuit voltage (Voc) and lightharvesting efficiency (LHE) ranges from 1.64 to 1.954 V and 52-91 %, respectively. Current study provides a valuable chemical space for the development of new organic semiconductors with improved efficiency.