2015 Volume 12 Issue 22 Pages 20150802
In this paper, we first propose two multicast partitioning methods named TBP (two block partitioning) and LBP (layer block partitioning), each of which has a different level of efficiency in 3D bufferless network on chip. In addition, we present a Recursive Partitioning (RP) method in which the network is recursively partitioned until all partitions contain comparable number of multicast destination nodes. By this method, the multicast destination nodes are distributed evenly and the network latency is significantly decreased. Simulation results illustrate that the RP scheme achieves 39%, 45% and 41% less latency on average than that of the TBP scheme and 8%, 17% and 12% less latency on average than that of the LBP scheme under three synthetic traffic patterns respectively.