Abstract

Wireless network on chip (WNoC) can be used as an alternative to bus technology in high-core chips in which the
multi-hop paths between far apart cores are replaced with a wireless single-hop link. The main reason for using
wireless communication is to reduce latency as well as power consumption. According to the limitation of
resources, the performance of the WNoC is sensitive to the routing algorithm. While an appropriate routing
algorithm reduces latency, it should avoid deadlock. In this paper, we propose a novel routing algorithm using Qlearning, which is one of the reinforcement learning methods for balancing wireless network traffic on the chip.
Using such an algorithm, the nodes can make decisions based on congestion conditions in the network when
transferring flits from the source node to the destination one. The simulation results show that using the proposed
reinforcement learning for routing the packets considerably improves the performance of the network, more
precisely, the system performance is improved by 8% compared with the previous related works.

Key words: Wireless Network on Chip (WNoC), Q-Learning Algorithm, Reinforcement Learning (RL), Routing Algorithm, Deadlock.