Abstract

The single-instruction-multiple-data (SIMD) processing environment requires that in each time cycle an identical processing or interconnect be performed. Various optical free-space interconnect schemes for a SIMD processor array have been proposed. A common feature of these schemes is that the processors are distributed on a rectangular grid. A major problem with rectangular array topologies is that their optical implementations require the use of both space-invariant and space-variant optical elements. For example, the nearest-neighbor and the barrel-shifter networks require linearspace-invariantoperations for their center elements and spacevariant (wrap-around) operations for their edge and corner elements. In addition, because of the different geometrical distances between the rectangularly distributed processing elements, the latency problem, in which signals transmitted on the network over different paths undergo different delays, may seriously limit the processing rate of the system.

© 1990 Optical Society of America