Hierarchical optimization of photovoltaic device performance using machine learning

We propose a new approach to utilize machine learning models for solar cell loss analysis and optimization study. Specifically, different machine learning models trained for predicting individual precursor/lifetime/optical sample parameters will be running in a pipelined manner to predict and optimize the full cell performance. Sentaurus TCAD simulation data is used for training these machine learning models, and different regression models and learning frameworks are compared. Compensation effects from parameter variation in prediction results and its correlation with solar cell device physics is discussed. A framework to deploy these models using python REST APIs is discussed.