Abstract

A single-instruction machine is generally considered to be impractical in electronics because it may lead to excessively complex programs. In digital optical cellular image processors (DOCIPs), the single-instruction machine is well suited to the system architecture. The reason for this is that the optical parallelism and global interconnection capabilities of DOCIP also contribute to parallel optical control signal flow. We show that two DOCIP architectures, DOCIP array and DOCIP hypercube, have a simple organization, low cell complexity, and (with external memory) the ability to simulate random access machines and Turing machines. Both of them have a simple control structure: (1) a single instruction that includes fetch, execute, store, and pipelines the three fundamental operations of binary image algebra¹; (2) a single control level that has only one higher control level for interpretation between program and data. The DOCIP array, a nearest-neighbor connected cellular array, has an extremely simple organization due to its O(1) interconnectivity. The DOCIP hypercube increases the interconnectivity to O(log N) and combines features of conventional cellular logic and conventional hypercube architectures for 2-D computational cells. The programming and instruction set of the DOCIP array and the DOCIP hypercube are illustrated and compared.

© 1987 Optical Society of America