A novel, to our knowledge, type of packet scheduler that could significantl
y outperform current state-of-the-art schedulers is presented. The operatio
n and the design of such a scheduler are discussed, and a fully operational
experimental implementation is described. The scheduler uses a neural netw
ork in a winner-take-all strategy to optimize decisions on the throughput o
f both a crossbar and a banyan switching fabric. The problems of high inter
connection density are solved by use of a free-space optical interconnect t
hat exploits diffractive optical techniques to generate the required interc
onnection patterns and weights. (C) 2000 Optical Society of America.