Machine Learning for Optimization of Mass-Produced Industrial Silicon Solar Cells

We present a methodology where we combine numerical TCAD device modeling, machine learning and advanced statistics for getting a deeper understanding of how process variations influence device performance in mass produced crystalline silicon solar cells. For this, we use seven model input parameters that affect the mainstream solar cell design (PERC) and its performance the most and perform about a couple of hundred numerical TCAD device simulations in an expected range of these parameters. As such detailed numerical simulations take long time, we train and validate machine learning models on these simulations, which serve to describe ten thousands of fabricated PERC cells. The method gives concrete information for improving PERC cells with a modest amount of numerical modeling and hence in a very short time. This approach is not limited to a specific solar cell design or product.