Prediction of electronic parameters of compensated multi-crystalline solar-grade silicon using artificial neural networks

Because of economic and energy-consumption considerations, multicrystalline solar grade silicon (mc-SoG-Si), instead of expensive electronic-grade Si, is being considered in photovoltaic (PV) industry for production of solar modules. These materials usually contain a comparable amount of acceptors (e.g., Boron) and donors (e.g., Phosphorus) and are therefore called compensated mc-SoG-Si. The three main electronic parameters, e.g., majority carrier mobility (μ), majority carrier density (p) and resistivity (ρ), of compensated mc-SoG-Si vary nonlinearly with temperature due to several complex mechanisms. In this paper, we propose two artificial neural network (ANN)-based models to predict these electronic parameters of mc-SoG-Si material. Using a limited amount of measurement data, we have shown that the first ANN-based model can predict the three electronic parameters of a given sample without accounting for the compensation ratio over a wide temperature range of 70-400 K. Whereas, the second ANN model can predict these electronic parameters of a given sample with unknown compensation ratio over the same temperature range. With extensive simulation results we have shown that these models can predict the three parameters with a maximum error of ±10%.