Behavioral Modeling of Tunable I/O Drivers with Pre-emphasis Using Neural Networks

Abstract

This paper addresses the development of nonlinear behavioral models of tunable digital input/output (I/O) drivers covering features such as drive strength and pre-emphasis. The proposed modeling approach relies on the use of parameterized state-aware weighting functions that control the driver's output stage, which enables the accurate modeling of pre-emphasis behavior of the driver. The state-aware weighting functions are implemented using feedforward neural networks (FFNNs). The dynamic memory characteristics of the driver output port are captured using recurrent neural networks (RNNs). To address the tunable features in the state-of-the-art driver circuit designs, a parameterized model that takes into account driver control parameters is presented. Test cases of practical industrial driver examples demonstrate that the proposed modeling method offers good accuracy, flexibility and significant simulation speed-up to facilitate signal integrity analysis without compromising intellectual property (IP).