Route Packets, Not Wires: On-Chip Interconnection Networks

Route Packets, Not Wires: On-Chip Interconnection Networks,10.1145/378239.379048,William J. Dally,Brian Towles

Route Packets, Not Wires: On-Chip Interconnection Networks   (Citations: 1383)
BibTex | RIS | RefWorks Download
Using on-chip interconnection networks in place of ad-hoc glo- bal wiring structures the top level wires on a chip and facilitates modular design. With this approach, system modules (processors, memories, peripherals, etc...) communicate by sending packets to one another over the network. The structured network wiring gives well-controlled electrical parameters that eliminate timing iterations and enable the use of high-performance circuits to reduce latency and increase bandwidth. The area overhead required to implement an on-chip network is modest, we estimate 6.6%. This paper introduces the concept of on-chip networks, sketches a simple network, and discusses some challenges in the architecture and design of these networks. 1I ntroduction We propose replacing design-specific global on-chip wiring with a general-purpose on-chip interconnection network. As shown in Figure 1, a chip employing an on-chip network is com- posed of a number of network clients: processors, DSPs, memo- ries, peripheral controllers, gateways to networks on other chips, and custom logic. Instead of connecting these top-level modules by routing dedicated wires, they are connected to a network that routes packets between them. Each client is placed in a rectangu- lar tile on the chip and communicates with all other clients, not just its neighbors, via the network. The network logic occupies a small amount of area (we estimate 6.6%) in each tile and makes use of a portion of the upper two wiring layers. Using a network to replace global wiring has advantages of structure, performance, and modularity. The on-chip network structures the global wires so that their electrical properties are optimized and well controlled. These controlled electrical parame- ters, in particular low and predictable cross-talk, enable the use of aggressive signaling circuits that can reduce power dissipation by a factor of ten and increase propagation velocity by three times (3). Sharing the wiring resources between many communication flows makes more efficient use of the wires: when one client is idle, other clients continue to make use of the network resources.
Conference: Design Automation Conference - DAC , pp. 684-689, 2001
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.
Sort by: