Multiple-Input Neural Networks for Improved Photovoltaic Device Parameter Predictions

With the recent surge in AI popularity, neural networks have been used in a number of solar cell applications, including for materials discovery, processing optimization, and high-throughput characterization. One such implementation involves using neural networks to predict the physical parameters of solar cells. Researchers have employed simple device parameters, such as current-voltage characteristics, to estimate more complex materials parameters, such as carrier mobility. In this study, we propose a novel approach to data input for neural networks. Compared to the conventional practice of utilizing a single current-voltage curve as the predictive input, our novel approach involves the integration and layering of multiple experimental curves. This introduces heightened dimensionality to the data and thereby improves the overall performance of predictive models for solar cell optimization and non-destructive rapid characterization.