Abstract

Multistage Interconnection Networks (MINs), based on free-space optical interconnects, were proposed to overcome the communications bottlenecks, skew, and crosstalk associated with long electronic interconnects.¹ Versions based on 2-D arrays of processing elements (PEs) were suggested that more fully utilize the third dimension for higher density and efficient use of optical space bandwidth product (SBWP).^2,3,4 Several of these concepts are based on off-axis imaging techniques which perform shuffle⁵ based permutations by optically interleaving equal sized sectors (such as quadrants) of the source array and overlaying the result on an identical array of detectors. The detected signals are then subjected to local exchange/bypass switching elements which operate on small groups of the array and route the signals to the next stage's source array. The various schemes offer trade-offs between the number of stages necessary for an arbitrary permutation and the complexity of the local switching elements. Examples of proposed optical networks include the 2-D separable perfect shuffle,² the folded perfect shuffle,³ and higher order k-shuffles.⁴ Figure 1 is an example of a 1-D perfect shuffle, based on off-axis imaging and interleaving of two halves of the array. The figure is also a side view of the 2-D folded perfect shuffle or 2-D separable shuffle, based on the off-axis imaging and interleaving of the four quadrants of the array. Figure 1 depicts the input and output arrays to be identical, as required for cascadability. The optical efficiency for those emitters located at the outer regions of a quadrant can be increased by adding a lens over each quadrant, as shown in the figure, to capture the cone of light from each pixel's emitter and center it on the associated shuffling lens.

© 1993 Optical Society of America