Predicting Organic-Inorganic Halide Perovskite Photovoltaic Performance from Optical Properties of Constituent Films

The photovoltaic performance of organic-inorganic halide perovskite solar cells largely depends on the perovskite composition and fabrication process. Most laboratory-scale fabrication methods are not transferable to industrial scale technologies such as slot-die coating or roll-to-roll coating, etc. Hence, it is advantageous to predict solar cell performance before completing the device. Most of the prevalent perovskite solar cell physical models aim to interpret devices’ interlayer recombination, band-bending, and charge injection, while the number of free parameters in these models require extensive measurements and fitting models that hinders their effectiveness as a prediction tool. Here, we present a machine learning (ML) method to take multiple measured device optical spectrums as input and directly predict the device JV curve output (Voc, Jsc and FF), thereby predicting the solar cell efficiency. These Blackbox models allow us to predict the performance of full devices with data that might contain many physical processes which could be too complex to predict accurately with a physics-based model. The model has achieved an averaged prediction percent error less than 5%. By analyzing the prediction weights of each parameter, we identify the optical transmission spectrum and photoluminescence spectral peak information are the most significant parameters that influence the algorithm results. This demonstrated work is the first work in the field to use ML to predict device photovoltaic properties solely from the optical properties of constituent materials and further establishes the potential of small-data-driven ML methods for guiding new designs of perovskite solar cell structures.