Wireless Network On-Chips History-Based Traffic Prediction for Token Flow Control and Allocation

Abstract

Wireless network-on-chip (WiNoC) uses a wireless backbone on top of the traditional wired-based NoC which demonstrated high scalability. WiNoC introduces long-range single-hop link connecting distanced core and high bandwidth radio frequency interconnects that reduces multi-hop communication in conventional wired-based NoC. However, to ensure full benefits of WiNoC technology, there is a need for fair and efficient Medium Access Control (MAC) mechanism to enhance communication in the wireless Network-on-Chip. To adapt to the varying traffic demands from the applications running on a multicore environment, MAC mechanisms should dynamically adjust the transmission slots of the wireless interfaces (WIs), to ensure efficient utilization of the wireless medium in a WiNoC. This work presents a prediction model that improves MAC mechanism to predict the traffic demand of the WIs and respond accordingly by adjusting transmission slots of the WIs. This research aims to reduce token waiting time and inefficient decision making for radio hub-to-hub communication and congestion-aware routing in WiNoC to enhance end to end latency. Through system level simulation, we will show that the dynamic MAC using an History-based prediction mechanism can significantly improve the performance of a WiNoC in terms of latency and network throughput compared to the state-of-the-art dynamic MAC mechanisms.