Spatial computing on self-timed cellular automata

The huge computational power required to support paradigms of brain-like and evolvable computers in the 21st century will likely be delivered by computers with very homogeneous structures like cellular automata. It is expected that such computers can be manufactured cost-effectively in nanotechnology and will be at least ten orders of magnitude more powerful than current computers. Attempts to do general-purpose computations on cellular automata, however, have not gone beyond simulating Turing Machines on them. This paper proposes more effective ways to exploit the universal computing power of cellular automata. First, by using self-timing of cells, as opposed to synchronous timing, different parts of a cellular automaton can work more independently from each other, so is important to program them in a modular way. Secondly, by laying out programs for cellular automata spatially, the von Neumann communication bottleneck between processing elements and memory elements virtually disappears. Thirdly, by organizing programs as self-timed hierarchical modules, the ease of programming is greatly enhanced. Experiments with hierarchical modular programming are shortly reported.