xpipes: a Latency Insensitive Parameterized Network-on-chip Architecture For Multi-Processor SoCs

  • Authors:
  • Matteo Dall'Osso;Gianluca Biccari;Luca Giovannini;Davide Bertozzi;Luca Benini

  • Affiliations:
  • -;-;-;-;-

  • Venue:
  • ICCD '03 Proceedings of the 21st International Conference on Computer Design
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

The growing complexity of customizable embedded multi-processorarchitectures for digital media processing will soon require highlyscalable network-on-chip based communication infrastructures. Inthis paper, we propose xpipes, a scalable and high-performance NoCarchitecture for multi-processor SoCs, consisting of soft macrosthat can be turned into instance-specific network components atinstantiation time.The flexibility of its components allows our NoCto support both homogeneous and heterogeneous architectures. Theinterface with IP cores at the periphery of the network isstandardized (OCP-based). Links can be pipelined with a flexiblenumber of stages to decouple data introduction speed fromworst-case link delay. Switches are lightweight and supportreliable communication for arbitrary link pipeline depths (latencyinsensitive operation). xpipes has been described in synthesizableSystem C, at the cycle-accurate and signal-accurate level.