Diagnosis of clustered faults and wafer testing

Abstract

A probabilistic diagnosis algorithm is presented for constant degree structures. The performance of the algorithm is analyzed under a negative binomial failure distribution to account for fault clustering. It is shown that the algorithm can correctly identify almost all units even when the yield is low (much lower than 50%) and when faults are clustered. A wafer test structure is proposed, which utilizes the test access port of each die to perform comparison tests on its neighbors and incorporates a localized version of the diagnosis algorithm to determine the status of each die. Both the test time and the diagnosis time are invariant with respect to the number of dies on the wafer. The saving of test costs could be significant as compared with probe testing, because with probe testing dies are probed one at a time while they are tested in parallel with this scheme. The scheme is unique in that it is shown to work well when faults are clustered and when the yield is low.