Date of Award:

8-2018

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Electrical and Computer Engineering

Committee

Koushik Chakraborty

Abstract

Over the last decade, General Purpose Graphics Processing Units (GPGPUs) have garnered a substantial attention in the research community due to their extensive thread-level parallelism. GPGPUs provide a remarkable performance improvement over Central Processing Units (CPUs), for highly parallel applications. However, GPGPUs typically achieve this extensive thread-level parallelism at the cost of a large power consumption. Consequently, Near-Threshold Computing (NTC) provides a promising opportunity for designing energy-efficient GPGPUs (NTC-GPUs). However, NTC-GPUs suffer from a crucial Process Variation (PV)-inflicted performance bottleneck, which is called Choke Point. Choke Point is defined as one or small group of gates which is affected by PV. Choke Point is capable of varying the path-delay of circuit and causing different forms of timing violation.

In this work, a cross-layer design technique is proposed to tackle the performance impediments caused by choke points in NTC-GPUs.

Checksum

e4eada201a85ab0317f0fd660f78f61a

Share

COinS