Abstract
CMOS designs have some unique properties that prevent existing test generators from computing a test vector for a fault when one might exist. The problem lies in the premises laid out on what it takes to detect a stuck-at fault. The basic premise that states that it is required to set a line to 0(1) in order to detect a stuck-at 1(0) fault, and then propagate the error to an observable point, needs to be re-examined. This is due to the existence of indeterminate states throughout the logic. The paper distinguishes between the traditional test vector (here called a hard-detect), and a potential test vector (here called a soft-detect). Our proposed test set is the union of hard and soft-detects. We also re-examine the issue of redundancy and show that it needs to be re-defined in order to comply with CMOS technology behavior.This paper shows several examples to illustrate the problem; describes what it takes in order to remedy it; proposes possible enhancements to existing test generation algorithms, and outlines the risks faced in the event that no correcting steps are taken.