Statistical Techniques and Non-Destructive Testing Methods for Copper Wire Bond Reliability Investigation

Abstract

Microelectronic devices require packaging for mechanical protection and electrical interconnections. Reliability challenges in microelectronics packaging are becoming more severe, as applications demand smaller package sizes and operation in harsher environments, such as in automotive applications. At the same time, manufacturers are seeking to reduce production costs by using new materials, for example in wire bonding by replacing costly gold wire with more economical copper.

Because microelectronic devices are expected to function reliably for years or even decades, depending on the application, reliability testing is commonly accelerated, e.g. by using elevated temperature and/or humidity. Even so, testing is often time consuming, requiring weeks or months for product qualification. Furthermore, although standard test conditions exist, little guidance is available in the literature to indicate how long products passing these tests will survive in operation.

Non-destructive testing methods provide a great deal of information regarding product degradation and reliability. With proper statistical analysis, strong conclusions can be made about device reliability with relatively short test durations, since testing need not continue until all samples fail. However, data analysis techniques used in the electronics packaging literature are often limited, with statistical analyses and confidence bounds rarely presented. Analysis of incomplete or censored data requires specialized techniques from the field of survival analysis.

The contributions of this thesis can be divided in two topics. The first topic is the equipment and techniques used to obtain new reliability results, including a method for temperature calibration of the miniature ovens used, a modification of those ovens for use as environmental chambers with humidity control, and procedures for optimization of wire bonding processes.

Second, statistical techniques for analysis of reliability data are demonstrated, using accelerated failure time models to analyze resistance data from copper wire bonds in high temperature storage testing. In doing so, new information was provided to answer an important open question in the field of copper wire bonding, namely, the maximum temperature at which one can expect copper wire bonds on aluminum metallization to perform reliably.

In particular, ball bonds made from 25 µm diameter palladium-coated copper wire are estimated to be highly reliable up to at least 167 °C in a clean environment without encapsulation, with failure rate of only 1 ppm after 12000 h. PCC wires were more reliable than bare Cu wires when unencapsulated or when encapsulated in silicone. Conversely, bare Cu was more reliable than PCC when encapsulated in epoxy. The best-performing encapsulated bonds tested were bare Cu wire with a highly heat tolerant epoxy, which are estimated to survive 12000 h with 1 ppm failure probability at 159 °C.

Effects of several other factors on bond reliability were also investigated, namely the cleaning process, Al bond pad thickness, and the bonded ball size. Sample and environmental cleanliness were found to be critical to good reliability. Bond pad thickness and bonded ball size had only minor effects on reliability, suggesting that these factors can be safely chosen to satisfy other requirements such as bond pad pitch or current-carrying requirements.