PhD Thesis Presentation
Silicon Photonics-Based Optical Networks-on-Chip for Multiprocessor Systems: Architecture Designs and System-Level Thermal Analysis

by Miss Yaoyao YE

 :  01 Nov 2013 (Fri)
 :  10am
Venue  :  Room 4480, 4/F (Lifts 25-26), HKUST

Examination Committee
Prof Qian ZHANG, CSE/HKUST (Chairman)
Prof Jiang XU, ECE/HKUST (Thesis Supervisor)
Prof Sudhakar YALAMANCHILI, The School of Electrical and Computer Engineering, Georgia Institute of Technology (External Examiner)
Prof Chin-Tau LEA, ECE/HKUST
The energy-efficient multiprocessor system is a natural platform for high-performance computing and embedded systems. Silicon photonics-based optical networks-on-chip (ONoCs) are emerging on-chip communication architectures that can potentially offer high bandwidth and energy efficiency.

In this thesis, we propose a torus-based hybrid optical-electronic NoC and a 3D mesh-based ONoC for multiprocessor systems, and show their performance and energy efficiency under real applications and uniform traffic. New low-cost 4x4, 5x5, 6x6 and 7x7 on-chip optical routers are proposed to reduce cost and optical power loss. In addition, new techniques of adaptive power control mechanism, floorplan optimization, and low-latency control protocols are used to further enhance the energy efficiency and performance.

The second part of the thesis focuses on the system-level thermal modeling and analysis for ONoCs. Thermal sensitivity of photonic devices is one of major concerns for the optical on-chip interconnect, which will result in additional optical power loss under on-chip temperature variations. System-level ONoC thermal models are required to fully understand these challenges and help further develop the optical on-chip interconnect technology. We systematically model the thermal effects in WDM-based ONoCs as well as in single-wavelength ONoCs, and find the optimal device settings to minimize the impacts of thermal effects in ONoCs. Based on the thermal models, we reveal important factors regarding ONoC energy efficiency under temperature variations, including the initial setting of photonic devices, the number of switching stages in the ONoC architecture, and the bandwidth of the optical switching elements. We develop OTemp, an optical thermal effect modeling platform for both WDM-based and single-wavelength optical links in ONoCs. OTemp can be used to analyze the optical power loss and power consumption for optical links under temperature variations. Finally, we use case studies to quantitatively analyze the thermal-aware power consumption of ONoCs with different combinations of low-temperature-dependence techniques.
*** ALL ARE WELCOME !! ***