A multi-chip processor/memory arrangement replacing a large computer chip,
includes a number of modules each including processing elements,
registers, and/or memories interconnected by an optical interconnection
fabric providing an all-to-all interconnection between the chips, so that
the memory cells on each chip represent a portion of shared memory. The
optical interconnect fabric is responsible for transporting data between
the chips while processing elements on each chip dominate processing.
Each chip is manufactured in mass production so that the entire
processor/memory arrangement is fabricated in an inexpensive and
simplified technology process. The optical communication fabric is based
on waveguide technology and includes a number of waveguides, the layout
of which follows certain constraints. The waveguides can intersect each
other in the single plane, or alternatively, a double layer of waveguide
structures and bent over approach may be used. Specific layout patterns
of the optical waveguides are presented. The communication of data along
the optical communication channels is performed in highly pipelined
decentralized routing manner and is envisioned for XMT architecture
application.