The network-on-Chip (NoC), as a new architectural trend, can enhance the bandwidth of metallic interconnects in multiprocessor systems-on-chip (MPSoCs). Nevertheless, as the number of possible integrated processing cores on a single die continues to increase, metallic interconnects are not able to satisfy the bandwidth and latency requirements within the package power budget. To overcome such problems, optical networks-on-chip (ONoCs) have been proposed as the proper substitute for electronic NoCs in MPSoCs. However, the basic photonic devices which are widely used in constructing ONoCs are imperfect and suffer from inevitable crosstalk noise and power loss. As a result, the small crosstalk noise from the basic photonic devices accumulates in large scale ONoCs and ultimately con...[ Read more ]

The network-on-Chip (NoC), as a new architectural trend, can enhance the bandwidth of metallic interconnects in multiprocessor systems-on-chip (MPSoCs). Nevertheless, as the number of possible integrated processing cores on a single die continues to increase, metallic interconnects are not able to satisfy the bandwidth and latency requirements within the package power budget. To overcome such problems, optical networks-on-chip (ONoCs) have been proposed as the proper substitute for electronic NoCs in MPSoCs. However, the basic photonic devices which are widely used in constructing ONoCs are imperfect and suffer from inevitable crosstalk noise and power loss. As a result, the small crosstalk noise from the basic photonic devices accumulates in large scale ONoCs and ultimately considerably diminishes the signal-to-noise ratio (SNR), causes severe performance degradation, and constrains the network scalability. The crosstalk noise analysis is dependent on the topological properties of ONoCs and hence is different in various ONoC architectures. An analysis and comparison of the crosstalk noise in different ONoC architectures help in choosing the most appropriate architecture within the allowed power budget, while satisfying the required performance. For the first time, we study and model the worst-case as well as the average crosstalk noise and SNR in three well-known ONoC architectures, mesh-based, folded-torus-based, and fat-tree-based ONoCs, at the system-level. Formal analytical models for the worst-case as well as the average signal power, crosstalk noise power, and SNR are presented. We consider a general optical router model to enable our proposed crosstalk noise and SNR analyses to be applicable to ONoCs using an arbitrary optical router. Utilizing the proposed general optical router model, the average and the worst-case SNR link candidates, which restrict the network scalability, are found. The analytical models are integrated into a newly developed crosstalk and loss analysis platform, called CLAP, to facilitate the crosstalk noise and SNR analyses in arbitrary ONoC architectures. Case studies of ONoCs using the optimized crossbar optical router, the Crux optical router, and the optical turnaround router (OTAR) using recent photonic device parameters are presented. We perform quantitative simulations of the worst-case as well as the average SNR in different ONoC architectures in CLAP. The quantitative simulation results show the critical behavior of crosstalk noise in large scale ONoCs. For example, we find that in folded-torus-based ONoCs using the Crux optical router and in the worst-case, the crosstalk noise power exceeds the signal power for network sizes larger than 12×12; when the network size is 20×20 and the injection signal power equals 0 dBm, the worst-case signal power and crosstalk noise power are -9.4 dBm and -6.1 dBm, respectively.