Abstract

Opto-electronic interfaces can provide large numbers (a few thousand) of high bandwidth interconnections to a single chip[1]. However, an optical relay system[2] is capable of handling substantially more channels: up to a few hundred thousand. If, for example, a 1-D, nearest-neighbor interconnect were to be implemented optically then, although the capacity of the optoelectronic interface may be fully utilized, the potential capacity of the optics will not. To harness this ‘extra’ capacity the outputs from, and inputs to, many opto-electronic chips can be spatially interleaved within a free-space optical system. This allows a high dimensional network topology to be embedded within a physical 3-D layout, opening up the possibility of realizing attractive, but previously impractical, interconnection networks for massively parallel processing systems.

© 1997 Optical Society of America