Machine learning-assisted prediction of organic solar cell efficiency from TCA triplelayer reflectance spectra

Organic Solar Cells (OSCs) are one of the most promising solar cells due to the possible for large-scale and lowcost printed production. Therefore, efficient manufacturing processes and optimization methods are crucial. Currently, the traditional trial-and-error method is mostly used to optimize device performance, which is complex and time-consuming. Previous machine learning (ML) methods can reduce the workload, but relay on multiple inputs. To accelerate the optimization process, a novel ML-based approach was proposed to predict the power conversion efficiency (PCE) of OSCs, utilizing the reflectance spectrum of the transparent electrode/ charge transport layer/active layer (TCA triplelayer). For this purpose, a dataset, containing PCEs of six types of OSCs with different active layer materials and reflectance spectra of TCA triplelayers, had been constructed by simulations via finite-difference time-Domain method. Based on the dataset, machine learning algorithms were employed to construct the regression models. Spectra pre-processing and feature extraction techniques were integrated to refine the predictive accuracy of these models. Consequently, the model based on Multilayer Perceptron Regression (MLPR) algorithm demonstrated the best performance, with coefficient of determination (R2) of 0.984 and root-mean-squared error of 0.408. These results underscore the potential to accurately predict the PCE of OSCs from the reflectance spectra of TCA triplelayer. Ultimately, a strategy was further proposed to utilize the developed regression model for real-time quality monitoring of TCA triplelayer during device fabrication. This offers a rapid way to evaluate the quality of TCA triplelayers and their influence on device performance.