Abstract:Agile development is considered as an emerging trend in hardware design, and NoC-based multi-IP integrated system has become the mainstream solution. The most critical challenge of this solution is to design a common efficient NoC with ultra-high throughput and low-cost routers to meet the communication characteristics of varied applications. Usually, NoC can not achieve both the gains from high throughput and low cost because high-throughput router architectures need complex virtual channel control logic to ensure that the power consumption and area of buffers will not be too large. Meanwhile, the low-cost router architecture is difficult to achieve higher throughput by increasing the number of buffers due to its simple control circuits. Moreover, most routers in NoC are optimized only for specific applications. To address the problem, this paper presents a “viaduct theory” for throughput optimization and realizes a common NoC router architecture for different scenario requirements (CRAFT router). Firstly, we improve the top boundary of the router throughput with the fully-connected crossbar switch, which is an expensive solution for high throughput. Secondly, we use a method similar to the ring-decouple theory in low-cost routers to reduce the number of router buffers while maintaining high throughput. The experimental results show that the throughput for the low-cost program is 15.4% higher than the conventional router with the same overhead. For the high-performance program, our router saves 50% hardware overhead compared with fully-connected routers with the same throughput; The maximum throughput that can be achieved is 26.3% higher than conventional routers, and 14.2% higher than the other state-of-the-art designs.

Lock-free and Zero-copy Framework for Packet Forwarding Based on Multi-core Processor

A multi-path routing method with shared buffer on NoC

A High Performance Router Based On Network Processor And Ternary-Cam

CRAFT: Common Router Architecture for Throughput Optimization.

A 2d Mesh Noc With Self-Configurable And Shared-Fifos Routers

TrafficLite: A Configurable On-Chip Interconnect Router Microarchitecture

An Efficient Scheduling Mechanism With Flow-Based Packet Reordering In A High-Speed Network Processor

Green Router: Power-Efficient Router Design

GreenRouter: Reducing Power by Innovating Router's Architecture.

Power-proportional Router: Architectural Design and Experimental Evaluation

Design of a High-Throughput NoC Router with Neighbor Flow Regulation

A Fast and Efficient Global Router for Congestion Optimization

Implementation of Packet-Fair Queuing Algorithms in High Speed Routers Based on Discrete Rates and Packet Lengths Group Architectures

Per-flow Queuing Memory Management without Per-flow Queues in High Speed Routers

Research of Packet Forwarding among Virtual Routers in Identical Logic Network

Scalable Hardware Content Router: Architecture, Modeling and Performance

Caesar: High-Speed and Memory-Efficient Forwarding Engine for Future Internet Architecture

Study on a Packet Scheduling Algorithm in High Performance Routers

High-performance IPv6 Forwarding Algorithm for Multi-Core and Multithreaded Network Processor

Data Path Processing in Fast Programmable Routers

Critical-Path Driven Routers for On-Chip Networks