Abstract

We describe the use and implications of parallel models of computation in the design of optical architectures. The models covered include shared memory models and graph/network models¹; these models can be universal machines. Shared memory models consist of a set of processing elements operating in parallel and communicating with a global memory in an extremely general, contention-free manner; these models provide for fast computation of parallel algorithms. An architecture derived from a shared memory model is introduced and includes a set of electronic or optical processors, an optical reconfigurable interconnection network that permits fan-in, and some form of optical memory that permits contention-free simultaneous reads. Graph/network models consist of a set of processing elements with fixed interprocessor connections. Fine grained systems are considered. Fundamental differences between electronic and optical implementations of graph/network models are discussed; these stem from differences in assumptions on the processors and on the network degree and topology. Extensions to reconfigurable interconnections are also considered.

© 1986 Optical Society of America