Approaches to Realize High Precision CMOS Bandgap Reference Based on Neural Network

Due to the offset voltage which introduced by the mismatch and the channel-length modulation, the bandgap reference circuit in CMOS technologies is usually nonlinearly, and the design of bandgap references with high precise is a challenging circuit design problem. In this paper, we propose a method of optimizing the parameters of the components to compensate the non-linearity offset voltage using the backpropagation based neural networks. The method for training the network is investigated based on the coefficients of the input and output voltage of the circuit which make the training computationally efficient. The experimental results show that with the network structure approximation, the non-linearity of offset voltage can be reduced markedly, and the accuracy of bandgap reference is improved 10 times.