Statistical techniques for planning type I singly censored reliability experiments with two stress factors

Abstract

A crucial consequence of the ongoing miniaturization in the microtechnology is that producing reliable integrated circuits (IC's) gets more and more difficult. The subject of this thesis can be situated in the world of testing the reliability of such IC's. These tests are important because the competitiveness of manufacturers strongly depends on the demonstrable reliability of the produced IC's. The major building blocks of an IC can be subdivided into off-chip and on-chip elements. At the on-chip level, active components as well as so-called back-end elements have to be considered. Active components, mainly transistors and diodes, care for the information contained in an IC. Back-end elements can be defined as elements taking care for the interconnection between the active components. This work concerns reliability tests on on-chip elements. One of the major failure mechanisms of on-chip interconnects is electromigration [SC09I]. The basic principle of electromigration is known: as a consequence of an electrical current, atoms in a conductor migrate in the direction of the electrons. The major failure mechanism of a transistor is hot-carrier degradation [VUI98], while capacitors mainly age due to time-dependent dielectric breakdown [MAR98]. One of the crucial problems when testing the reliability of new components is that their. lifetime under normal operating conditions is always extremely long (in the order of years). For that reason, the physical mechanisms that are responsible for a component to fail are studied and methods are established for accelerating these mechanisms. The electromigration mechanism, for example, can be accelerated by subjecting the interconnects to elevated temperatures and current densities. Then, the failure times of the test components operating under these accelerated conditions are measured and models are developed for extrapolating these data to real life conditions. The final step of a reliability experiment is the statistical analysis of the results. The problem this work deals with can best be defined as: "How to plan a reliability experiment so that the total number of available test components and the total amount of vailable measurement time can be optimally used for predicting the reliability of the components under consideration?". In this thesis, restriction will be made for reliability experiments with two stress factors.