Contamination-Enhanced Growth of Au/Al Intermetallic and Horsting Voids
In 1972 C.W. Horsting published a paper relating what he and his colleagues at the RCA Corporation learned about how "bad" wire bonds can be driven to premature failure while "good" bonds remain strong2/. He was specifically looking at gold-to-aluminum bond interfaces. He reported that much of the literature of the time indicted "purple plague", or the presence of the purple colored species of gold-aluminum intermetallic, as the cause of weak aluminum-to-gold bonds. We now know that the occurrence of intermetallic species always occurs in these bonds and that these species are stronger than the original materials. New work at that time was finding that Kirkendall voiding, the creation of voids in intermetallics due to the different diffusion rates of the two materials with respect to each other, was causing weakened areas in the bonds when stressed with thermal cycling. The voiding was found to develop faster with elevated temperature.
Horsting's paper reports testing and analyses done after several lots of parts with gold plated posts and aluminum ultrasonic wire bonds, were showing bond lifts after long duration bakes (1000 hrs, 200°C). Some lots didn't have this problem. An experiment was done where bond wires were installed in electronic packages and bond pulls were done before and after a 1 hour bake at 360°C in a clean, nitrogen atmosphere. The distribution of the pull strengths showed clearly that the "good" and "bad" packages looked the same before the bake. After the bake the pull strength distributions for the "good" and "bad" packages moved to the left (lower strength). The "bad" packages showed a binomial distribution with one lobe of the distribution well into the failure range. Though the time and temperature were very effective test conditions, the paper reports that their selection was "somewhat arbitrary". What luck! Crossections were done of the "good" and "bad" bonds aged using the 200°C/1000 hr condition and the 360°C/1 hr condition. The "bad" ones looked the same following each test: purple intermetallic between the aluminum wire and the unconsumed gold on the post. The "good" ones looked the same before and after too: aluminum wire, purple intermetallic, white intermetallic, no unconsumed gold, nickel underplating.
Horsting reported that extensive testing was done on many packaging lots to eliminate bond process variations and to confirm the initial results. This follow-on testing proved a clear dependence of the bond strength on the amount of contamination in the gold plating. Neither stripping the top layers of the gold nor overplating the contaminated gold with "clean" gold improved the bond strength. Removal of contaminated gold and replating with "clean" gold brought back the reliable bonds. The underplating and thickness of the gold were not factors.
Horsting concluded that "Impurities in the gold are expelled and concentrated directly ahead of the moving diffusion front by a process similar to zone refining. When the concentration of these impurities rises above some critical level, precipitation occurs. The precipitate particles act as sinks for vacancies produced in the diffusion reaction and small voids develop, which eventually join to cause a separation. This separation acts increasingly as a diffusion barrier to the gold and eventually shuts off the gold supply to the diffusion zone...".
This concept agrees with other published work and aptly describes what we at GSFC have recently seen in two separate cases of gold-to-aluminum wire bond failures. A NASA JSC flight readiness report for STS-109 reported cases of bond lifts (aluminum pads, gold wires)/141. Destructive Physical Analysis records showed that out of 19 samples tested of one part number, one manufacturer, seven lot date codes between 1974 and 1981, the three parts that had bond pulls below 4 grams were parts that had chlorine contamination in the bond. Harman1/ describes an experiment where Bromine was able to contaminate the gold-aluminum intermetallic from a source outside of the bond area while the aluminum wire itself was not corroded by the free Bromine. Harman also describes several cases of Fluorine and Chlorine contributing to gold-aluminum bonds as well as . A more recent occurrence of Horsting voiding was the result of polyimide contamination at a wirebond interface (see Wire Bond Lifts on the Homepage herein or click here).
Note: Go to this Website to find an interesting interview done with Ernest Kirkendall, published by Hideo Nakajima in The Minerals, Metals & Materials Society's Journal of Metals, 49 (6) (1997), pp. 15-19