Study and implementation conditions of the multivariate outlier detection methods for screening of potential field failures

In semiconductor manufacturing for automotive, reliability constraints are strong: zero defect is expected by specific statistical techniques. In particular, distributions of the parametric electrical tests implemented at the probe or assembly steps, are controlled, test by test. In order to increase reliability in field by enhancement of potential failure screening, new techniques of test result distribution monitoring are studied with a multivariate approach, when all the test results are addressed simultaneously.