Special-purpose hardware [for cellular automata]

The simple structure of cellular automata makes it natural to think of implementing them with special-purpose hardware. And indeed from the 1950s on, a sequence of special-purpose machines have been built to implement 1D, 2D and sometimes 3D cellular automata. Two basic ideas have been used: parallelism and pipelines. Both ideas rely on the local nature of cellular automaton rules.

In the parallel approach, the machine has many separate processors, each dedicated to handling a single cell or a small group of cells. In the pipelined approach, there is just a single processor (or perhaps a few processors) through which the data on different cells is successively piped. The key point, however, is that at every stage it is easy to know what data will be needed, so this data can be prefetched, potentially through a specially built memory system.

In general, the speed increases that can be achieved depend on many details of memory and communications architecture. The increases have tended to become less significant over the years, as the on-chip memories of microprocessors have become larger, and the time necessary to send data from one chip to another has become proportionately more important.

In the future, however, new technologies may change the trade-offs, and indeed cellular automata are obvious candidates for early implementation in both nanotechnology and optical computing. (See also page 841.)