An improved interconnect noise methodology for reduced pessimism using dAP model

Abstract

Noise simulators were never accurate estimators of interconnect noise values as the switching pattern, slew and delay in a cluster of adjacent nets were never addressed. In modern signal integrity flows, a very high level of pessimism is incorporated into the noise methodology to augment the lack of accuracy, such as lumped aggression, noise ramps and lumped slews. Desirably, such pessimistic assumptions are much easier to compute hence efficient on simulation runtime, but often than ever it will result in an overdesigned chip. This paper discusses a recipe to reduce pessimism in interconnect noise methodologies by effectively modeling the downstream degradation on the aggression line. We will be presenting an aggression slew estimation method derived from the Elmore delay, which has less than 0.8% error compared to SPICE simulations and is inclined towards the higher slew bound when compared to both Elmore's delay and SPICE. We also show that the dAP method reduces the average reported noise by 5% even when we account for victim-aggressor reverse aggression issues.