SPACE, an associative processor architecture, has been designed to allow experimentation with AI applications. The instruction set is small and highly orthogonal. All data bits in a word support parallel, maskable writes, allowing parallel logical and arithmetic transformations on stored data. Stored data bytes can be individually masked as don't cares. Processors can communicate with their neighbours in a bidirectional linear array to support multi-word data objects, and to allow parallel communication between neighbouring objects.
An implementation of SPACE containing 170496 processors has been
completed. This implementation used aggressive packaging techniques to
reduce the relative cost of associative storage. Performance has been
measured for a range of primitive operations.