Optical interconnection network for parallel access to multi-rank memory in future computing systems.

Area of Science:

• Computer Engineering
• Optical Networking
• High-Performance Computing

Background:

• Increasing core counts necessitate high-bandwidth, low-latency interconnects for core-to-memory communication.
• Traditional electronic bus systems face limitations in scaling and performance.

Purpose of the Study:

• To propose and investigate an optical interconnection network for simultaneous multi-rank memory access.
• To improve core-to-memory communication efficiency in modern computing architectures.

Main Methods:

• Design of an optical interconnection network with a novel wavelength assignment strategy.
• Development of a distributed, pipeline-mode memory controller architecture.
• Simulation using the PARSEC benchmark suite to evaluate performance.

Main Results:

• The proposed optical network enables simultaneous communication with different memory ranks.
• Efficient broadcast for flow control is achieved.
• Significant bandwidth enhancement and latency reduction compared to electronic bus systems were demonstrated.

Conclusions:

• The optical interconnection network offers a promising solution for high-performance core-to-memory communication.
• The designed architecture and wavelength strategy effectively address the demands of multi-core systems.
• The findings suggest a viable path towards more scalable and efficient computing interconnects.