NoC: Network or Chip?

Abstract

Summary form only given. The concept of a communication network emerged, many times in the past, for connecting a large number of systems, replacing dedicated point-to-point connection and other small-scale interconnection mechanisms. Each network needs to provide a cost effective solution for a large number of possibly conflicting requirements such as flexibility, scalability, reliability and performance. Therefore, the task of network architects and designers is to solve multiple instances of a complex constrained optimization problem resulting in numerous and diverse network solutions. For example, there are different standards and architectures associated with interconnection networks, home networks, LANs, MANs, WANs and wireless networks. In this paper, we map the common lessons and concepts from the networking research to the emerging NoC field. We argue that the NoC optimization problem consists of several distinguished types that should lead to multiple diverse solutions. NoC network layer architectures pose new challenges in exploring solutions to traditional networking problems such as routing, quality-of-service, flow and congestion control and reliability. The unique characteristics of silicon chips require new solutions to these classical problems, and define a new set of NoC specific problems, such as automatic network design process, power and area optimization and specialized system functionalities. We speculate which class of solutions is likely to fit the different NoC types