A Q-routing based self-regulated routing scheme for network-on-chip

A Q-learning based congestion table dynamic adjusted NoC routing scheme is presented to reduce the impact of the packet transfer delay on system performance in condition of network congestion. First, according to the degree of congestion, each router selects between DyXY routing scheme and Q-learning based routing scheme. The proposed routing algorithm relies on transfer latency of each packet and adjacent routers traffic condition to updates congestion status table of router in real time and make an optimal selection for the packet output channel. Secondly, congestion table self-regulated technique is proposed to overcome the defect that staled value in table degrading robust of routing. Finally, we compare the presented routing method to DyXY routing algorithm through BookSim2.0 simulation platform. The result shows that our method outperforms DyXY method by 13.08% and 16.82% in uniform traffic and transpose traffic respectively.