ABSTRACT

The introduction of very large scale integration circuit technology has offered a promising scenario for processing large-scale data by multiprocessor arrays in a highly parallel mode. This chapter discusses a special-purpose network of processors; i.e., wavefront array, which was developed by S. Y. Kung et al. The important feature of this wavefront array processor is its computational notation which removes the need for global synchronization and also proves to be useful for programming the machine and describing the algorithm. Consequently, the temporal locality condition required in systolic arrays (i.e., it has to be at least one time-unit delay allotted, so that signal transactions from one node to the next can be completed) is no longer necessary, since there is no explicit timing reference in the wavefront arrays. The chapter concludes with remarks on the simulation of the phenomenon of the propagation of the waves and a further discussion about some improved issues for the quadrant interlocking factorization method.