A custom processor for use in a parallel computer system

Abstract

A 440000-transistor, full-custom CMOS processor that is used as the basis of a parallel computer system is described. The primary design goal was to produce a processor that performed roughly an order of magnitude faster than its predecessor. The author discusses the chip-level architecture of the processor, comparing it to the original design, and shows what was done architecturally to increase the performance by an order of magnitude