Design Of NoC Based On2D-mesh Topology

With the continuous expansion of the system size, the continuous improvement of the system complexity and the growing number of IP core, integrated circuit design is facing more problems. Noc (network-on-chip) appears to adapt to the development of the IC industry. The most important problem of NoC design is how to improve system performance and reduce system power consumption. In order to get higher throughput, lower delay and lower system power consumption of the NoC system, a detailed analysis of the NoC topology, routing algorithm, the structures of communication nodes and resource nodes has been done. A NoC implanted of a4-7-13-1FF-ANN, based on4x42D-Mesh topology and OE algorithm, whose communication nodes’virtual channel FIFO buffer depth of input port is16has been designed. Simulation results show that the new system dramatically improves the system with a small increase in system power consumption.The main works completed by the author of the paper include:1. The working principles of two routing algorithm-XY routing algorithm and OE routing algorithm are described. The performance of the two routing algorithms is compared in a same NoC structure based on2D-Mesh topology. Results show that the OE algorithm provides higher throughput and smaller system delay in the architecture for the system.2. The performance of the FIFO buffer depth in communication nodes’virtual channel in a NoC system based on4x42D-Mesh topology has been explored. Results show that the system gets the maximum throughput and minimum delay without waste of system hardware when the FIFO buffer depth is6.3. The system architecture of a NoC implanted artificial neural network is described. Results show that the NoC system implanted hardware applications has better performance in throughput and latency.On the basis of the above work, a new NoC system has been proposed. The system was tested by NIRGAM emulator and the accessed data is reliable and effective. Test results confirmed the idea proposed by the author, and achieved the desired optimization purposes. It is feasible and deserves further research.