Behavioral Modeling of Steady-State Oscillators with Buffers Using Neural Networks

This paper describes a method to model the nonlinear time-domain steady-state behavior of oscillators with output buffers using neural networks. In the proposed model, an augmented neural network (AugNN) with a periodic unit is used to capture the periodicity of the oscillatory output waveform, where the nonlinear dynamic behavior of the output buffer is taken into account using recurrent neural networks (RNN). The model is trained using the data obtained from the simulation of transistor-level circuit models. The proposed model is compatible with Verilog-A. An example applied to a buffer-included ring oscillator demonstrates that the proposed modeling method offers good accuracy and significant simulation speed-up to facilitate time-domain analysis without compromising intellectual property (IP).