Physics-guided characterization and optimization of solar cells using surrogate machine learning model

Characterization, material parameter extraction and subsequent optimization of solar cell devices is a highly time-consuming and complex procedure. In this work, we propose a method for quick extraction of limiting material parameters in solar cell devices using a surrogate, physics-embedded, neural network model. This surrogate model, implemented by an autoencoder architecture trained with a physical numerical model, allows to quickly extract the device parameters of interest at a certain process condition by using only a small number of illumination dependent current-voltage (JV) measurements. Our surrogate model adequately links material parameters at a certain process condition to final device efficiency. The model provides physical insights about the location of the best performing and robust processing conditions in solar cell devices. We test our approach with GaAs and CH3NH3PbI3 (MAPbI) perovskite solar cells. The model allows to find a set of processing conditions that maximize the performance of both GaAs and MAPbI solar cells, and analogous processing conditions that minimize solar cell variability.