Comparison of Machine Learning Techniques for Predictive Modeling of High-Speed Links

Abstract

We compare three different machine learning techniques for constructing predictive model for eye opening based on channel length and interconnect cross-sectional geometry. Surrogate model is constructed using sparse grids, support vector regression, and artificial neural networks. Models for training data are generated using quasi-TEM modeling of the interconnect, and eye opening training data is obtained from statistical high-speed link simulation using IBIS-AMI transmitter and receiver models. Numerical results illustrate that all three methods offer reasonable predictions of eye height, eye width and eye width at 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-12</sup> bit error rate.