Stratified testing of multichip module systems under uneven known-good-yield

Abstract

In this paper, a stratified technique is proposed for testing multichip module (MCM) systems. Its advantages are an improvement in quality level (QL) and cost-effectiveness. The proposed testing approach is accomplished in the presence of uneven known-good-yield (KGY) for MCMs consisting of different sets (or strata) of chips. This approach referred to as the lowest yield-stratum first-testing (LYSFT) considers the unevenness of KGY between strata for testing the chips and improving the QL. For comparison purposes, exhaustive testing (ET), random testing (RT) and random stratified testing (RST) are also evaluated. Given the strata, KGY and the sample size of the chips under test, the proposed LYSFT approach allocates and tests the sampled chips in a greedy (first) fashion, while RT and RST select the chip (for RT) and the stratum (for RST) randomly. A Markov-chain model is developed to analyze these testing approaches and is solved analytically in O(SN/sup 3/) for the LYSFT (where S is the number of strata and N is the number of chips in the MCM). A cost model is proposed as figure of merit and shown to relate the defect level (DL) with the number of tests performed. Parametric results show that the LYSFT dramatically outperforms RT and RST for improving the QL. A considerable reduction in tests can be achieved by the LYSFT at a very small loss in QL compared with ET.