Routing Algorithm Performance Comparison for Network-on-Chip in Neuromorphic Computing
The neuromorphic computing requires a very-large-scale integration system to implement the spiking neural
network. However, traditional bus topology is not suitable for network interconnection due to its low scalability. To achieve
scalability, network-on-chip is applied in the neuromorphic computing. The routing algorithms used by the network-on-chip
affect the networks performance. Unlike the wormhole routing model, which has more than one flit in each packet, the
packet of the neuromorphic computing only contains one flit. However, performance analysis of different routing algorithms
completed in previous works is based on packets containing multiple flits. In this paper, the performance of the XY, oddeven,
and dynamically adaptive and deterministic routing algorithms (DYAD)are analyzed for the neuromorphic computing.
The simulation is carried out on the NIRGAM simulator.
Index Terms - Network-on-chip, neuromorphic computing, XY routing algorithm, odd-even routing algorithm, DYAD