Exploring concentration and channel slicing in on-chip network router

Abstract

Sharing on-chip network resources efficiently is critical in the design of a cost-efficient network on-chip (NoC). Concentration has been proposed for on-chip networks but the trade-off in concentration implementation and performance has not been well understood. In this paper, we describe cost-efficient implementations of concentration and show how external concentration provides a significant reduction in complexity (47% and 36% reduction in area and energy, respectively) compared to previous assumed integrated (high-radix) concentration while degrading overall performance by only 10%. Hybrid implementations of concentration is also presented which provide additional tradeoff between complexity and performance. To further reduce the cost of NoC, we describe how channel slicing can be used together with concentration. We propose virtual concentration which further reduces the complexity -saving area and energy by 69% and 32% compared to baseline mesh and 88% and 35% over baseline concentrated mesh.

Holistic Routing Algorithm Design to Support Workload Consolidation in NoCs

“…A local strategy (LOCAL), NoP, RCA-1D 3 and DBSS are implemented for different size meshes. Finally, we extend the evaluation to concentrated meshes [2], [32].…”

“…Here, we extend the analysis to the concentrated mesh (CMesh) topology [2], [32]. As a case study, we use radix-4 CMeshes [2], [32]; four cores are concentrated around one router, with two cores in each dimension, as shown in Fig.…”

“…Each core has its own injection/ejection channels to the router. Based on a CMesh latency model, the network channel has a 2-cycle delay, while the injection/ejection channel has 1 cycle delay [32]. The router pipeline is the same as discussed in Sec.…”

Circuit-Switched Memory Access in Photonic Interconnection Networks for High-Performance Embedded Computing

“…They also use integrated concentration [24] of 4 cores per network gateway, for a total core count of 256.…”

“…External concentration [24] is used because of the relative difficulty of designing high-radix photonic switches, and to reduce the number of modulator/detector banks. Designs of 4×4 photonic switches in the context of networks have been explored in [5], but because a mesh router requires 5 ports (4 directions + processor core), we must reconsider the design of the photonic switches to minimize power and insertion loss.…”