Functional and Performance Analysis of Network-on-Chips Using Actor-based Modeling and Formal Verification

Abstract

Network on Chip (NoC) has emerged as a promising architecture paradigmfor todays many-core systems. As complexity grows in NoCs, functional verificationand performance prediction in the early stages of the design process are suggestedas ways to reduce the fabrication cost. Formal methods have gained moreattention as alternative ways for analyzing NoC designs. In this paper we propose amethod to model different characteristics of the system, and also verify various functionaland performance properties by generating the full state space of the model fordifferent scenarios. We present a formal model for two-dimensional mesh GloballyAsynchronous Locally Synchronous (GALS) NoCs with four-phase handshakecommunication protocol, using the actor-based modeling language Rebeca. Functionaland timing behaviors, routing algorithm and communication protocol are capturedin the model. Deadlock freedom, message arrival, and end-to-end packet latencyare checked. In order to analyze large NoCs we propose a scalable approachbased on compositional verification for estimating maximum end-to-end packet latency.The compositional approach is specific for the XY-routing algorithm. Resultsof verification are compared and matched to simulation results of HSPICE using32nm technology.