Publications

Mention the words ''stress voiding'', and everyone from technology engineer to manager to customer is likely to cringe. This IC failure mechanism elicits fear because it is insidious, capricious, and difficult to identify and arrest. There are reasons to believe that a damascene-copper future might be void-free. Nevertheless, engineers who continue to produce ICs with Al-alloy interconnects, or who assess the reliability of legacy ICs with long service life, need up-to-date insights and techniques to deal with stress voiding problems. Stress voiding need not be fearful. Not always predictable, neither is it inevitable. On the contrary, stress voids are caused by specific, avoidable processing errors. Analytical work, though often painful, can identify these errors when stress voiding occurs, and vigilance in monitoring the improved process can keep it from recurring. In this article, they show that a methodical, forensics approach to failure analysis can solve suspected cases of stress voiding. This approach uses new techniques, and patiently applies familiar ones, to develop evidence meeting strict standards of proof.

We have designed and manufactured a test chip devoted to the study of interconnect voiding. The test chip is suitable for evaluating theoretical models, acceleration recipes, and the effects of process variations. We describe the chip and a simple, stress-free packaging technique that eliminates any stress to the chip from die bonding or packaging thermal cycles. With this test chip, we can perform many necessary and desirable experiments: determining stress, observing or stimulating void growth, profiling hydrogen concentrations, and measuring excess current noise. We report here preliminary measurements of residual stress, observations of voids, and determinations of hydrogen concentrations of hydrogen concentration under variations in aluminum annealing and passivation. In agreement with observations elsewhere, we find that passivations which differ greatly in intrinsic stress do not differ much in the stress they impart to patterned metal; some workers have suggested instead that excess hydrogen in the aluminum contributes to voiding. Following this lead, we have used nuclear reaction analysis to profile the hydrogen concentration in passivation, metallization, barrier metal, and interlevel dielectric and present some preliminary measurements here. We conclude that passivated metallization may contain as much as 0.1 atomic % hydrogen. 10 refs.