Machine Learning-Based Defect Identification Method at the c-Si/a-Si:H Interface

Generation of defects leads to performance degradation of silicon solar cells. However, ab initio simulations of defect structures are constrained by system size, while the representative simulations of amouphous silicon (aSi) require a large amount of computation because of the wide distribution of relevant local structures. We propose a machine learning-based approach to replace the quantum mechanical ab initio calculations to identify defect formation in aSi-related structures. We demonstrate that with a correlated set of descriptors and a representative set of training data, the machine learning method can identify localized defects in structures containing Si-H structures with high fidelity.