Networks and Interconnects
Networks of independently timed processors are very common because of the difficulty of linking the timing of two or more processors each of which is physically distinct from the others. The internet is an example of a network of processors in which vast amounts of data move between many millions of personal computers and servers, each of which operates with its own independent clock. It is impractical to link all the clocks in all the PCs in the world, so the data must be re-synchronized as it arrives at the input to each PC. This is not really a problem, as the data rates are relatively modest, usually less than 10 MHz, and any delays resulting from synchronization are small compared with the overall communication link delay. The development of integrated circuit fabrication technology has also produced networks on a much smaller scale on chip, where it is also difficult to distribute a single clock. Increasing chip size and reducing metal interconnect widths mean that the delays produced by a connection can be much greater than the delays produced by a gate, or even a processor. Not only are they large, but they are also difficult to predict at the design stage, so the design of a completely synchronous system on chip in which all the registers are controlled by the same clock becomes much more difficult.
8.1 COMMUNICATION ON CHIP
A system on silicon will normally consist of many processor modules connected by a system of communication links. The ...