High performance multiprocessor computing system using optical centralized shared bus

Abstract

The success of optical interconnects already emerged at the machine-to-machine level giving an example for replacing electrical interconnects. There are three basic board-level optical implementation techniques: waveguide, free-space, and substrate-guided optical interconnects. With appropriate design and integration of optical components, signal broadcast can be effectively implemented in an innovative architecture called optical centralized shared bus. Based on this architecture, we proposed in this paper to apply the optical centralized shared bus in the multiprocessing systems for performance improvement. With the focus on demonstrating the technical feasibility in a general scenario, a conceptual emulation of the centralized shared-memory scheme was carried out on a generic PCI subsystem that incorporated an optical centralized shared bus.