Date of Award

Spring 1999

Project Type

Dissertation

Program or Major

Engineering

Degree Name

Doctor of Philosophy

First Advisor

Peter Arato

Abstract

Interfacing hardware-oriented high-level synthesis to software development is a computationally hard problem for which no general solution exists. Under special conditions, the hardware-software codesign (system-level synthesis) problem may be analyzed with traditional tools and efficient heuristics. This dissertation introduces a new alternative to the currently used heuristic methods. The new approach combines the results of top-down hardware development with existing basic hardware units (bottom-up libraries) and compiler generation tools. The optimization goal is to maximize operating frequency or minimize cost with reasonable tradeoffs in other properties.

The dissertation research provides a unified approach to hardware-software codesign. The improvements over previously existing design methodologies are presented in the frame-work of an academic CAD environment (PIPE). This CAD environment implements a sufficient subset of functions of commercial microelectronics CAD packages. The results may be generalized for other general-purpose algorithms or environments.

Reference benchmarks are used to validate the new approach. Most of the well-known benchmarks are based on discrete-time numerical simulations, digital filtering applications, and cryptography (an emerging field in benchmarking). As there is a need for high-performance applications, an additional requirement for this dissertation is to investigate pipelined hardware-software systems' performance and design methods. The results demonstrate that the quality of existing heuristics does not change in the enhanced, hardware-software environment.

Share

COinS