Advancing Circuit Transient Response Macromodeling: From Conventional Neural Networks to Siamese-LSTM

Abstract

Macromodeling is a crucial technique to enhance a circuit design. Machine learning and neural networks have been recognized as useful techniques to perform macromodeling. In this paper, we have shown that using neural networks to perform macromodeling is feasible. We have presented different neural network methods for modeling the transient response of a low-noise amplifier (LNA) circuit. LSTM-based method has proven to have a superior performance against multilayer perceptron (MLP) and simple recurrent neural network (RNN). Finally, we have proposed a Siamese-LSTM-based neural network (S-LSTM) that can model the LNA circuit with a range of input amplitudes. We have also shown that our proposed S-LSTM model can model the transient response on a power amplifier circuit.