A Statistical Quality Model for Delay Testing

Abstract

SUMMARY In this paper we introduce a statistical quality model fordelay testing that reflects fabrication process quality, design delay margin,and test timing accuracy. The model provides a measure that predicts thechip defect level that cause delay failure, including marginal small delay.We can therefore use the model to make test vectors that are effective interms of both testing cost and chip quality. The results of experiments us-ing ISCAS89 benchmark data and some large industrial design data reflectvarious characteristics of our statistical delay quality model. key words: delay testing, quality model, defect distribution 1. Introduction The progress of fabrication process and design technologyleads to marginal or parametric failures that are caused byresistive shorts or resistive vias [1]. They will occur moreoften and it will be more difficult to screen them effectivelyby the conventional testing methods.Delay testing is regarded as one of the key technologiesfor those problems and has been widely studied to improvetesting effectiveness. Many delay fault models were pro-posed. The transition delay fault model [2] considers thepropagation of lumped delay defects by logical transition tothe observation pins or flip-flops. It is widely used becauseof its high fault coverage, but it cannot detect defects caus-ing delays that are smaller than the test timing. The path de-lay fault model [2] considers the propagation of distributeddelay along a path. It can detect small delay defects, but itis hard to achieve high fault coverage because there is ex-ponential number of paths in a chip. Some longest pathsare selected and tested; however, their coverage is so smallthat the effectiveness is not quantified [2]. A statistical pathselection technique that considers process variation was pro-posed [3]. It is a reasonable approach for practical testing,but its effectiveness is still unknown.There are some papers that address small delay defects.Park et at.[4],[5] give the basic idea that evaluates small de-lay defects. They introduced the delay defect density func-