ECE Seminar Series
Title: Multiprocessor Computing Systems Integrating Silicon Photonic Interconnects
Speaker: Mahdi Nikdast
Affiliation: Polytechnique Montreal/McGill University, Montreal, Canada
Day: Friday, February 24, 2017
Time: 11:00 am - 12:00 pm
Location: Engineering B214
Abstract: Computing systems play an important role in our lives today, from data centers to automobiles, cellphones, medical systems, etc. Such systems are continuously scaling by integrating an increasingly large number of processing cores to satisfy new application demands, such as higher computation and communication bandwidth requirements for emerging Big Data applications. As a result, the performance of computing systems is determined not only by the performance of different processing cores and memories but mainly by how efficiently they collaborate with one another. Conventional metallic interconnects cannot keep pace with the ever-growing communication requirements in computing systems since they impose a higher latency and power consumption with a limited bandwidth as the system scales. Addressing such issues, silicon photonics is introduced as a promising technology to realize high-performance interconnects for multiprocessor computing systems, paving the way for high-performance computing. Employing silicon photonic interconnects (SPIs) in computing systems requires novel system architecture designs and design methodologies. Furthermore, SPIs are sensitive to thermal and fabrication process variations and they intrinsically suffer from power loss and crosstalk noise. This interdisciplinary talk discusses different opportunities and challenges related to employing SPIs in multiprocessor computing systems. In particular, we comprehensively study the impact of crosstalk noise and fabrication process variations on the performance of such systems, and develop efficient design solutions, both at the system-level and device-level, to address those issues. Our studies contribute to developing automation design tools and system design methodologies, as well as realizing reliable silicon photonic devices for implementing high-performance computing systems integrating SPIs.