A Behavioral Design Flow for Synthesis and Optimization of Asynchronous Systems Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 21, 2019
Creator
  • Hansen, John B.
    • Affiliation: College of Arts and Sciences, Department of Computer Science
Abstract
  • Asynchronous or clockless design is believed to hold the promise of alleviating many of the challenges currently facing microelectronic design. Distributing a high-speed clock signal across an entire chip is an increasing challenge, particularly as the number of transistors on chip continues to rise. With increasing heterogeneity in massively multi- core processors, the top-level system integration is already elastic in nature. Future computing technologies (e.g., nano, quantum, etc.) are expected to have unpredictable timing as well. Therefore, asynchronous design techniques are gaining relevance in mainstream design. Unfortunately, the field of asynchronous design lacks mature design tools for creating large-scale, high-performance or energy-efficient systems. This thesis attempts to fill the void by contributing a set of design methods and automated tools for synthesizing asynchronous systems from high-level specifications. In particular, this thesis provides methods and tools for: (i) generating high-speed pipelined implementations from behavioral specifications, (ii) sharing and scheduling resources to conserve area while providing high performance, and (iii) incorporating energy and power considerations into high-level design. These methods are incorporated into a comprehensive design flow that provides a choice of synthesis paths to the designer, and a mechanism to explore the spectrum between them. The first path specifically targets the highest-performance implementations using data-driven pipelined circuits. The second path provides an alternative approach that targets low-area implementations, providing for optimal resource sharing and optimal scheduling techniques to achieve performance targets. Finally, the third path through the design flow allows the entire spectrum between the two extremes to be explored. In particular, it is a hybrid approach that preserves a pipelined architecture but still allows sharing of resources. By varying performance targets, a wide range of designs can be realized. A variety of metrics are incorporated as constraints or cost functions: area, latency, cycle time, energy consumption, and peak power. Experimental results demonstrate the capability of the proposed design flow to quickly produce optimized specifications. By automating synthesis and optimization, this thesis shows that the designer effort necessary to produce a high-quality solution can be significantly reduced. It is hoped that this work provides a path towards more mature automation and design tools for asynchronous design.
Date of publication
Resource type
Rights statement
  • In Copyright
Note
  • ... in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Computer Science.
Advisor
  • Singh, Montek
Degree granting institution
  • University of North Carolina at Chapel Hill
Language
Parents:

This work has no parents.

Items