A load balanced mapping for spiking neural network

Network-on-Chip (NoC) provides a scalable and packet-based inter-connected architecture for spiking neural networks (SNNs). However, existing neural mapping strategies just distribute all neurons in a population to an on-chip network core or nearby cores sequentially. The neurons within a population take a huge time cost for handling spikes, which results to uneven workload distribution among different on chip network nodes. This paper presents a NoC-based SNN mapping that makes workload balance among different nodes, aiming to accelerate application execution time. The experimental results show that proposed mapping strategy reduces application execution time by average 24%.