An independent overview of NoC can be found at http://en.wikipedia.org/wiki/Network_On_Chip
As chip complexity increases with each new semiconductor process node, the pictures of these chips look more and more similar to diagrams of distributed computer systems. These complex systems on a chip (SoCs) are composed of numbers of processing blocks that may be a mix of different types or in rare cases a collection of the same type of processing block. There are also a rising number of interface functions either for data communications between the chip and other functions in the overall system (off chip memory for example) or with physical devices that handle human interface functions. It is natural then to apply lessons learned from the evolution of networked computer systems to these complex chips, most notably in the autonomy of the individual blocks and their communications with each other and the world outside the chip. The general benefits that come from using networking for computer systems, now the default means of connecting these systems together, are also provided at the chip level when Network on Chip technology is used.
These benefits include:
- Autonomy and independence of any node in the system from the other interconnected nodes
- Scalability of the overall system is accomplished by adding standard system components with characteristics appropriate to service the needs of that portion of the network
- Transport of a wide variety of data types and formats with standard hardware and protocols through the use of packetization of the data being transported
- Network capacity for each node to node connection can be independently scaled, reducing the cost of the overall system to the aggregate of the individual node to node costs
- Interoperability of network nodes from different suppliers is guaranteed by adherence to standard networking protocols and physical interfaces
- Independent optimizations of components in the system is enabled by the use of standards and the separation of the computation (the function of a node) from the communications between nodes
- Testing of the whole system is simplified because each node can be tested in isolation; if aggregated performance requirements between nodes are within the capabilities of the node then the system will function as a whole
- Instrumentation at nodes simplifies overall system debug by providing a systematic means to divide the problem / determine where the failure exists
- Different services can be layered on top of network basics to provide a broad spectrum of use or support for new applications; voice being carried over a data network for example
- Specialty suppliers focused in the networking area offer superior cost performance systems while not competing with those suppliers offering best in class computing solutions.
Physical design considerations can be taken into account in advance of the full system being constructed. As with verification, if the units of the NoC have been designed with special physical design challenges already solved, then the overall system design can be dramatically simpler. For example, links that need to span long on-chip distances can be designed in advance to reduce the impacts of spanning distance while maintaining the desired performance characteristics.