A Deep Learning Approach for Loss-Analysis from Luminescence Images

Abstract

Identifying and quantifying loss mechanisms in solar cells are key requirements for increasing cell efficiencies. In this study, we present a novel method based on luminescence images to identify and quantify losses in silicon cells using a state of art deep learning technique: generative adversarial networks. In addition to the common use of defect identification, we also use the images to isolate a specific defect and to quantify its impact on cell efficiency. This is achieved by reconstructing a defect-free luminescence image and comparing it to the original image to determine the performance shortfall. The large-scale loss-analysis powered by the proposed deep learning method has the potential to significantly improve the quantitative analysis of luminescence image data, both in research and development and in high volume manufacturing.