Short Communication
Array size anomaly of problem-size independent systolic arrays for matrix-vector multiplication*

https://doi.org/10.1016/S0167-8191(05)80153-XGet rights and content

Abstract

A simple method for obtaining problem-size independent systolic arrays from a certain type of problem-size dependent systolic arrays is presented. The resulting array is fully fault-tolerant in the sense that a faulty processing element can be bypassed and the same problem of the original size can be solved. Such an array may outperform another larger array of the same type under certain conditions; this phenomenon is called the array size anomaly. Matrix-vector multiplication is used as an example problem solved by the systolic arrays. Two systolic arrays of different sizes for computing a given convolution are then derived and compared.

References (17)

  • H.T. Kung et al.

    Wafer-scale integration and two-level pipelined implementation of systolic arrays

    J. Parallel Dist. Comput.

    (1984)
  • O. Ersoy

    Semisystolic array implementation of circular, skew circular, and linear convolutions

    IEEE Trans. Comput.

    (1985)
  • K. Hwang et al.

    Partitioned matrix algorithms for VLSI arithmetic systems

    IEEE Trans. Comput.

    (1982)
  • H.T. Kung

    Let's design algorithms for VLSI systems

  • H.T. Kung

    Notes on VLSI computation

  • H.T. Kung

    Why systolic architectures?

    Computer

    (1982)
  • H.T. Kung et al.

    Systolic arrays (for VLSI)

  • H.T. Kung et al.

    A two-level pipelined systolic array for convolutions

There are more references available in the full text version of this article.

Cited by (6)

  • Matrix-vector multiplication on a fixed-size linear systolic array

    2000, Computers and Mathematics with Applications
  • Fast parallel algorithms for forecasting

    1997, Computers and Mathematics with Applications
  • Systolic implementation of nonlinear transformation of two sequences

    2003, 6th International Conference on Telecommunications in Modern Satellite, Cable and Broadcasting Service, TELSIKS 2003 - Proceedings
  • A family of bidirectional systolic arrays for matrix-vector multiplication

    2002, Proceedings of the International Conference on Modern Problems of Radio Engineering, Telecommunications and Computer Science, TCSET 2002
*

This research was supported in part by the National Science Council of ROC under contract NSC-79-0408-E011-05.

View full text