Time-Division-Multiplexing–Wavelength-Division-Multiplexing-Based Architecture for ONoC

Abstract

Future many-core processors will require high-performance yet energy-efficient on-chip networks to provide a communication substrate for the continually increasing number of cores on one chip. Optical network-on-chip (ONoC) is employed as a promising candidate interconnection solution for its high bandwidth and low energy consumption. However, optical circuit-switching (OCS) based architectures face the problem of high network congestion, low network utilization, and an overhead in power dissipation under heavy loads. In this paper, we propose an ONoC architecture with time-division multiplexing (TDM) and wavelength-division multiplexing (WDM) technology to solve the network contention problem faced by OCS-based ONoC. The number of wavelength groups and timeslots is optimized by using a genetic algorithm. A new optical router is designed to realize our TDM–WDM communication technology. A detailed model is built to analyze insertion loss and crosstalk noise. The simulation results show that TDM–WDM-based ONoC has better performance compared with equivalent OCS-mesh ONoC under a uniform traffic pattern. Similar analysis can be drawn for the real science application based on the PARSEC benchmark.