Research On Virtual Channel Allocation Algorithm For NoC

In recent years, NoC(Network on Chip) has gradually become the focus in the fieldof academy. With the development of NoC, the design requirement of NoC in powerloss has become increasingly stringent, and the communication on chip has becomemore sensitive to the transmission latency. Wormhole routing algorithm requires lessspace on the cache and lower latency, has been widely used. However, HoL(Head ofLine blocking) may be happened in the case of large traffic because of its routingmechanism, and latency will be increased, throughput will be decreased. Existingvirtual channel technology makes the original serial buffer resources parallelization, sothat the blocked packet is blocked only in the current virtual channel, the data packetswaiting behind will be transported in other free virtual channels, thus the HoL problemis effectively solved. However, how to allocate virtual channels can increase networkperformance but remain or decrease the chip area, become the issue to be considered inthe NoC application.In this paper, virtual channel technology theory of NoC both at home and abroad isstudied, a virtual channel allocation algorithm based on the redirect model(RMoVCA)is proposed through the analysis of the flow control mechanism, the virtual channelredirection model is designed for data packets which cannot be properly transmitted dueto physical link failure or node error can be cached and waits for redirection. Thus,RMoVCA not only solves the problem HoL, but also enhances the fault tolerance andimproves the throughput rate. The simulation results show that RMoVCA is improvedby12%on network performance. Through the analysis of NoC communication, avirtual channel allocation algorithm based on genetic simulated annealingalgorithm(GSAoVCA) is proposed. The initial population is operated by geneticalgorithm, the search is controlled by the annealing temperature, then the optimal NoCvirtual channel allocation scheme will be obtained. Thus, the resources allocation willbe optimized, and the network performance is improved by GSAoVCA. The simulationresults show that GSAoVCA were respectively increased by18.9%,64.6%thanGAoVCA and UniformVCA on network performance.The paper proposes two virtual channel allocation algorithms which have a largeimprovement in network performance. The next step of the research work will continueto the direction such as buffer distribution, node design and so on, thereby to constantly optimize NoC performance.