Gaussian-based optical networks-on-chip: Performance analysis and optimization

Abstract

Compared with optical networks-on-chip (ONoCs) based on traditional topology such as mesh, torus, and fat-tree, Gaussian-based ONoCs have significant topological advantages in the network diameter and average jump distance. However, the intrinsic loss and crosstalk noise, which are inevitably inherent elements in basic optical devices, can lead to severe degradation of network performance and even restrict the normal communication of ONoCs. Therefore, in this paper, a numerical analysis model for analyzing the worst-case crosstalk noise and optical signal-to-noise ratio (OSNR) in Gaussian-based ONoCs is proposed. According to the all-pass characteristic of Gaussian-based ONoCs, the 5-ports all-pass optical routers are selected, which make the proposed model suitable for Gaussian-based ONoCs using arbitrary 5-ports all-pass optical routers. In addition, a numerical simulation is presented, which uses the Cygnus and optimized crossbar optical routers to verify the feasibility of the proposed analytical model. The simulation results show that the OSNR of Gaussian-based ONoCs decreases sharply with the network scale enlargement, the insertion loss and crosstalk noise tremendously limits the network scale. Furthermore, in order to make the Gaussian-based ONoCs more available, an optimization method is also proposed for improving network performance. The optimization method choose the best optical link between two arbitrary communication points, which can effectively avoid the worst case and greatly improve the performance of Gaussian-based ONoCs.