(Georgia Institute of Technology, 2020-09-08)
Coskun, Ayse
The design of today's leading-edge systems is fraught with power, thermal, and variability challenges. The applications in rapidly growing computing domains of cloud and HPC exhibit significant diversity and require an increasing number of threads and much larger data transfers compared to applications of the past. In tandem, power and thermal constraints limit the number of transistors that can be used simultaneously, which has led to the Dark Silicon problem. Thus, it is becoming increasingly difficult to harness the full potential of computer chips. This talk argues that there is a need for novel design and management approaches to push computing systems operation closer to their peak capacity and reclaim the dark silicon. Specifically, the talk will discuss how to use 2.5D integration technology with silicon photonic networks (PNoCs) to build (heterogeneous) computing systems that provide the desired parallelism, heterogeneity, and network bandwidth to handle the demands of the next-generation applications. At the core of this ambitious vision is designing modeling and optimization frameworks that are able to capture and tweak the complex cross-layer interactions among devices, architecture, applications, and their power/thermal characteristics. Specific methods that will be highlighted in the talk include runtime management of applications and PNoC wavelengths, EDA methods that optimize placement & routing of PNoC systems with strong power and thermal awareness, and new architectures built with PNoCs.