Here is the abstract you requested from the Wirebonding_2016 technical program page. This is the original abstract submitted by the author. Any changes to the technical content of the final manuscript published by IMAPS or the presentation that is given during the event is done by the author, not IMAPS.
|Analysis of wedge tool wear-out by machine data|
|Keywords: wedge wear-out, prediction model, machine data|
|Goal: Future applications will challenge the die bonding technologies with smaller sizes, more precision and higher complexity. In the European ENIAC project EPPL, Fraunhofer IISB and Infineon evaluated the potential of Advanced Process Control (APC) methods, to increase the precision and overall quality in Back-End processes. While the wire bonding process is appearing in 90% of the worldwide produced devices, the number of unsatisfying bond quality in manufacturing production is still high. One reason for a weak performance is worn-out wedge tools. Thus, a concept was developed to rate the appearance of wear-out and detect it by different metrology methods. The presented evaluation shows that in the selected use-case only a statistically low number of wedge tools are actually worn-out at the fixed end of their lifespan. The wedge bonding machine parameters, generator current and tool deformation, have been analyzed for their capability to predict the wedge tool wear-out (Predictive Maintenance, PdM). The achieved results indicate a financial benefit, when exchanging wedges only if their performance is weak. Method: The evaluation of the tool wear-out was performed with 10 v-grooved tungsten carbide wedges at standard equipment for 500 µm aluminum-aluminum wedge bonding. In production, the wedge tools are cleaned several times and finally replaced after a fixed number of bond counts (known as Preventive Maintenance). The bonding parameters time, power and force are most relevant for the bond formation . In the start regime of the bonding process (until approximately 50 ms) the power and force are increased by a linear acceleration rate and stay constant afterwards. The output trace data of the generator current and tool deformation was collected for every bond count. Both parameters were analyzed in detail by developed key numbers. Furthermore, external metrology data has been gathered before and after every cleaning process. The bonding quality was determined by pull tests, shear tests and picture analysis of the bonds and the wedges. Results: The analysis shows that the average wedge is actually not worn-out at the fixed end of its lifespan. Only 1 out of 10 wedges showed a weaker performance, revealed by picture analysis at the end of their lifespan. The performance drop seems to be caused by a wear-out of the inside margins of the v-grooved tool. The inside margins of 2 analyzed wedges showed notches orthogonal to the direction of wedge vibration. In the wedge with deep notches, this caused excessive bonding deformation with “ear” shaped residues. The regularly performed pull and shear tests were not able to detect the change in performance. Thus, the wedges with the notches are not noted in every-day production. Unfortunately, the performed data analysis by machine data could not clearly detect the phenomena as well. Nevertheless, the results show that the average wedge is not worn-out at all and that a higher level of process control is possible. The potentials to detect the wear-out by machine data can only be verified by reaching longer lifespans and characterize a truly bad performance, caused by the wedge wear-out.|
|Klingert, Felix, PhD student
Fraunhofer Institute for Integrated Systems and Device Technology IISB, Friedrich-Alexander-Universität Erlangen-Nürnberg