Communicating process architecture: Transputers and occam

Abstract

A communicating process language such as occam can be used to design VLSI devices, and can be compiled into silicon layout. Some parts of the design process are still performed by hand (such as the final placement of the functional blocks), but this cannot introduce errors. It is therefore possible to design concurrent VLSI systems using occam, establish that the design behaves as intended using the formal semantics of occam (or in simple cases by experimental testing of the occam program), and finally compile the occam source into correct silicon layout.

In order to simplify the construction of the compiler, many issues have been ignored. For example, the synchronous communication system is only appropriate for local communication between devices sharing a common clock. This problem can be overcome by using a different link implementation for ‘long distance’ communication (for example, the link used in the transputer itself).

Expressing an application in a form which efficiently exploits silicon area involves careful consideration of the relative costs of memory, processing and communication. Concurrent algorithms which perform ‘redundant’ calculations can be faster and consume less area than sequential algorithms which store values. An important use of a silicon compiler is to aid in the evaluation of ‘silicon algorithms’.