Hie failure analysis lab has analyzed 'n' devices that all have the same failure signature (a failure mode plus other observable characteristics), and found that they all failed from the same mechanism. We wish to identify this mechanism with the failure signature so that future parts with the same signature may be assigned a mechanism without analysis, but by inference from historical data, thus saving lab time and resources. What is the risk of error? A probability model is presented that allows the analyst to calculate a confidence interval for the proportion of future devices with the same signature failing from the same mechanism. An X/Y criterion is defined: one is X% confident that greater than Y% of future devices with this signature will also have this mechanism. The model is presented for an 'on-going process' application and for a 'finite-population' application. Easy calculation methods are presented, and charts are given to illustrate and shortcut the calculations.

This paper deals with the basic concepts of Signature Analysis and the application of statistical models for its implementation. It develops a scheme for computing sample sizes when the failures are random. It also introduces statistical models that comprehend correlations among failures that fail due to the same failure mechanism. The idea of correlation is important because semiconductor chips are processed in batches. Also any risk assessment model should comprehend correlations over time. The statistical models developed will provide the required sample sizes for the Failure Analysis lab to state "We are A% confident that B% of future parts will fail due to the same signature." The paper provides tables and graphs for the evaluation of such a risk assessment. The implementation of Signature Analysis will achieve the dual objective of improved customer satisfaction and reduced cycle time. This paper will also highlight it's applicability as well as the essential elements that need to be in place for it to be effective. Different examples have been illustrated of how the concept is being used by Failure Analysis Operations (FA) and Customer Quality and Reliability Engineering groups.

A new method of signature analysis is presented and explained. This method of signature analysis can be based on either experiential knowledge of failure analysis, observed data, or a combination of both. The method can also be used on low numbers of failures or even single failures. It uses the Dempster-Shafer theory to calculate failure mechanism confidence. The model is developed in the paper and an example is given for its use.