This paper presents a delay computing-model for a 2D-mesh wormhole based NoC architecture that is a widely used topology structure in NoC design. The model captures the core's message sending probability, packet length and the contention of link in the communication. The different solutions of core's mapping onto NoC architecture will cause different average delay and a genetic algorithm, which is based on the delay model, can automatically provide an approximately optimal mapping for large scale NoCs. The algorithm aims to achieve a minimum NoC average delay. Experimental results for random traffics and various NoC sizes show that an average approximately 20% reductions in the execution time than random mapping.
展开▼
