Crosstalk Analysis and Optimization of Gaussian Networks-on-Chip

Abstract

Gaussian-based optical networks-on-chips (ONoCs) have significant topological advantages over mesh and torus networks in terms of network diameter and average jump distance. However, the crosstalk noise that accumulates in the network can cause severe performance degradation and limit the scalability of the network. Therefore, this paper analyzes the three aspects of basic optical elements, optical router and network, and proposes a Gaussian-based ONoCs worst-case crosstalk noise and signal-to-noise ratio (SNR) analysis model. In addition, we use Cygnus and crossbar optical routers for numerical simulations to verify the feasibility of the proposed analytical model. In order to improve the availability of Gaussian networks, we propose an optimization method to improve network performance that can effectively avoid the worst case.