An Algebra for VLSI Algorithm Design.

Abstract

Algorithms designed for VLSI implementation are usually parallel and two-dimensional in the sense that many processing elements laid out on a silicon surface can operate simultaneously. These algorithms have been typically described by graphs or networks where nodes represent processing elements or registers and edges represent wires. Although for many purposes these traditional representations are adequate for specifying VLSI algorithms, they are not suited for manipulating algorithms designs. In this paper an algebraic representation, together with a semantics, is proposed for VLSI algorithm designs. By algebraic transformations analogous to some typically used in linear algebra, alternative but equivalent designs satisfying desirable properties such as locality and regularity in data communication can be derived. This paper describes this powerful algebra for manipulating designs, and provides a mathematical foundation for the algebraic transformations. The algebraic framework is more suitable for supporting formal manipulation on designs than the network or graph-theoretic models, especially for complete designs. As an application of the proposed algebra, the paper demonstrates its use in the design and verification of systolic algorithms. (Author)

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Apr 01, 1983
Accession Number
ADA142796

Entities

People

  • H. T. Kung
  • W. T. Lin

Organizations

  • Carnegie Mellon University

Tags

Communities of Interest

  • Advanced Electronics
  • C4I

DTIC Thesaurus Topics

  • Algebra
  • Algorithms
  • Computations
  • Computer Science
  • Computing System Architectures
  • Digital Communications
  • Electrical Engineering
  • Fabrication
  • Image Processing
  • Integrated Circuits
  • Linear Algebra
  • Mathematics
  • Notation
  • Semantics
  • Signal Processing
  • Two Dimensional
  • Very Large Scale Integration

Readers

  • Artificial Intelligence
  • Parallel and Distributed Computing.