PSIJ Transfer Function Response Prediction via NARNET and KBNNs

In this paper, nonlinear autoregressive neural network is combined with knowledge-based neural network in order to develop an efficient method that further expands the bandwidth of the jitter transfer function. In the proposed hybrid approach, a knowledge-based neural network is developed using the training data from two types of models: fast-to-evaluate analytical model for jitter transfer function and computationally expensive circuit simulator generated accurate jitter transfer function response. Knowledge-based neural network can efficiently produce relatively accurate prediction of the jitter profile within the desired frequency range, using which a large number of data points is generated. In the next step, nonlinear autoregressive neural network is trained using data obtained from knowledge-based neural network. Proposed nonlinear autoregressive neural network ensures reasonable accuracy even beyond the frequency range of the original accurate data that is used in developing the knowledge-based neural network. A case study with 32nm CMOS technology is presented to demonstrate the validity of the proposed approach compared to a circuit simulator.