Neural network-based model for dual-junction solar cells

Design and development of solar cells can be substantially improved by using models which can provide accurate estimation of complex device characteristics. The artificial neural network (NN)-based models which learn from examples is an effective modeling technique that overcomes the deficiencies of conventional analytical techniques. In this paper, we propose NN-based modeling techniques for estimation of behavior of dual-junction (DJ) GaInP/GaAs solar cells involving complex phenomena, e.g., tunneling effect and complex interactions between the junctions. With extensive computer simulations we have compared performance of NN-based models with that of a sophisticated device simulator, ATLAS form Silvaco. We have shown that the NN-based models are able to estimate the solar cell characteristics close to that of the experimentally measured response. Compared with the response from ATLAS-based models, the NN-based models provide better results in estimation of tunneling phenomenon, determination of external quantum efficiency and I-V characteristics of DJ solar cells. Copyright (C) 2010 John Wiley & Sons, Ltd.