Inferring Material Parameters from Current-Voltage Curves in Organic Solar Cells via Neural-Network-Based Surrogate Models

Abstract

Machine learning has emerged as a promising approach for estimating material parameters in solar cells. Traditional methods for parameter extraction often rely on time-consuming numerical simulations that fail to capture the full complexity of the parameter space and discard valuable information from suboptimal simulations. In this study, we introduce a novel workflow for parameter estimation in organic solar cells based on a combination of numerical simulations and neural networks. The workflow begins with the selection of an appropriate experimental dataset, followed by the definition of a device model that accurately describes the experiment. To reduce computational complexity, the number of variable parameters is carefully selected, and reasonable ranges are set for each parameter. Instead of directly fitting the experimental data using a numerical model, a neural network was trained on a large dataset of simulated results, allowing for efficient exploration of the high-dimensional parameter space. This approach not only accelerates the parameter estimation process but also provides valuable insights into the likelihood and uncertainty of the estimated parameters. We demonstrate the effectiveness of this method on organic solar cells based on the PBDB-TF-T1:BTP-4F-12 material system, demonstrating the potential of machine learning for rapid and comprehensive characterization of emerging photovoltaic materials.