Ultra fast neural models for analysis of electro/optical interconnects

Abstract

The key to realize low-cost, low-power and high-frequency systems with shrinking design margins and expanding system complexities lies in the effective use of design automation tools in both architectural and detailed design stages. The need to reduce design iterations of such systems further demands that the tools be fast and reliable. A new modeling paradigm is presented to drastically speedup model evaluation and simulation of electrical and optical interconnects. Transmission line models are developed based on neural networks for signal integrity evaluation, allowing massive analysis of delay, crosstalk and switching noise on large numbers of signal tracks to be performed in a timely fashion. Neural models for laser diodes is also developed for analysis of optical systems. Such analysis can be used during architectural and detailed physical design stages to catch potential design errors early and more accurately, contributing to increased design efficiency.