scispace - formally typeset
Open Access

A systolic 2-D convolution chip

H T Kung, +1 more
TLDR
A chip for performing the 2-D (two-dimensional) convolution in signal and image processing that consists of essentially only one type of simple cells, which are mesh-interconnected in a regular and modular way, and achieves high performance through extensive concurrent and pipelined use of these cells.
Abstract
: This paper describes a chip for performing the 2-D (two-dimensional) convolution in signal and image processing. The chip, based on a systolic design, consists of essentially only one type of simple cells, which are mesh-interconnected in a regular and modular way, and achieves high performance through extensive concurrent and pipelined use of these cells. Denoting by u the cycle time of the basic cell, the chip allows convolving a kxk window with an nxn image in O(sq m)(u/k) time, using a total of cu k basic cells. The total number of cells is optimal in the sense that the usual sequential algorithm takes O(sq m)(sq k)(u) time. Furthermore, because of the modularity of the design, the number of cells used by the chip can be easily adjusted to achieve any desirable balance between I/O and computation speeds. (Author)

read more

Content maybe subject to copyright    Report

Citations
More filters
Book

Models of Computation: Exploring the Power of Computing

TL;DR: In Models of Computation, John Savage re-examines theoretical computer science, offering a fresh approach that gives priority to resource tradeoffs and complexity classifications over the structure of machines and their relationships to languages.
Journal ArticleDOI

Parallel computer architectures for image processing

TL;DR: The fundamental problems of developing an effective MSIMD system are discussed and a simple SIMD/MIMD computational model for comparison with such systems is proposed.
Journal ArticleDOI

Parallel 2-D Convolution on a Mesh Connected Array Processor

TL;DR: The basic idea of the proposed scheme is to apply the 1-D systolic concept to 2-D convolution on a mesh structure to make the scheme suitable for VLSI implementation.
Book ChapterDOI

A Two-Level Pipelined Systolic Array for Convolutions

TL;DR: A two-level pipelined systolic array that is capable of performing convolutions of any dimension and the designs take full advantages of the pipelining assumed to be available at each cell are described.
Book ChapterDOI

Systolic Algorithms for Running Order Statistics in Signal and Image Processing

TL;DR: The running order statistics (ROS) problem is defined, a generalization of median smoothing, and algorithms designed for VLSI implementation are presented which solve the ROS problem and are efficient with respect to hardware resources, computation time, and communication bandwidth.