Stacked Package Delamination
02/01/2003
Using a "fish bone" diagram to identify causes
BY T.Y. LIN
An advanced chip stacking technology has been developed recently in response to the requirements of the consumer electronics market.1,2
Conventional integrated circuit (IC) packaging process — die attach, wire bonding and molding — can produce the stacked die packages with high assembly yield. The major challenges are the backgrinding and sawing processes because die thickness must be reduced to the range of 50 to 125 µm3,4 to fit two or more ICs within the standard package heights. Gentle wafer handling is an additional requirement that arises because of thin wafers.4 The die-attach process can be implemented by epoxy paste or tape/film.5 For same-size die stacking, a silicon spacer or tape5 typically separates the bottom and top die. For wire bonding, the combination of normal and reversed bonding will be used for solving wire sweep issues.6 No critical issues have been found in the molding process in terms of wire sweep, coplanarity and mold voids.7
Warpage issues can appear during the die bonding process due to the thin, large die.4 Die-attach paste selection is critical because the epoxy bleed-out can cover the wire bonding pads and cause wire bondability problems5, and large errors of die placement can lead to electrical failures.3,8 Thermal management also can be a problem because at least one die-attach layer usually is electrically nonconductive, which typically correlates with a lower thermal conductivity.
 Figure 1. C-SAM results show delamination under the lead frame paddle in some units. |
null
In this article, the stacked die was constructed in 100-lead thin quad flat packs (100TQFPs) and performed by temperature cycles according to JEDEC level 3 (60°C/60%RH for 40 hours). The temperature cycle results showed that 100TQFPs with stacked die construction could pass electrical functionality tests after 1,000 temperature cycles. However, it was found that the full delamination occurred at 100 temperature cycles at the interface between the lead frame paddle and the mold compound.9-12 The evaluations described here used "fish bone" diagrams to understand the effects of the process, material and design on the results.
The main results revealed that die-attach paste voids initiated the small delamination at the die-attach edge and propagated down to the lead frame/mold compound interface due to the stress concentration and the weak adhesion strength (without a dimple structure at the backside of the lead frame).
Reliability evaluations and C-SAM imaging were performed on the packages with stacked die after 100 temperature cycles. C-SAM results showed that 16 out of 82 failed at the mold compound/lead frame paddle interface (Figure 1).
Delamination Analysis with "Fish Bone" Diagram
Figure 2 shows the "fish bone" diagram that can help identify the possible root causes inducing the delamination. The 100TQFP with a single die is quite mature, so it can be used as the controlled test vehicle.
A review of the materials involved showed that the die-attach paste was the only material that was different between the single-die control lot and the stacked-die components under evaluation. The mold compound and lead frame materials were the same.
From the design viewpoint, there were numerous differences between the control and experimental lot. In the stacked configuration, the top die was the same size as the die in the control lot, and the lower die in the stacked version was larger (5.0 x 5.0 mm vs. 3.17 x 3.80 mm). Also, the stacked die each were 0.4 mm thick, while the single chip was 0.5 mm. It could be assumed that the stiffness and stress distribution for single and stacked die packages, therefore, were significantly different. On the process side, the main difference was the insertion of a plasma clean step in the stacked die configuration just before wire bonding. The plasma cleaning may reduce the copper oxidation thickness, which could affect the adhesion strength between the mold compound and lead frame pad. As such, the plasma clean may contribute to the delamination.
 Figure 2. A "fish bone" diagram can identify possible root causes of mold compound delamination. |
null
Material and Process Evaluations
EP1 was used in the single-die package; EP2 was selected as the die-attach paste for the stacked die package. To compare the performance of EP1 and EP2 in the same test vehicle, both were dispensed onto the 100TQFP. The result of the reliability assessment is illustrated in Table 1. Further studies showed that the delamination was induced by die-attach paste voids after die-attach curing. Delamination can initiate at the voids and propagate to the surface under the hygrothermal loading.11,13
 |
Similar experiments were run with different mold compounds and plasma cleaning as a variable, and no delamination difference were noted.
 Figure 3. Finite element stress analysis for a TQFP with (a) a single die and (b) stacked die. |
null
Package Structure FEA
Finite element analysis (FEA) of the stresses in the different package constructions was performed. The thermal stress was simulated based on the temperature variations from 175° to 25°C using the commercial FEA code ANSYS. Figure 3 shows the principal stress of a cross-section of each package structure.
 |
It should be noted, however, that the stress simulation was only of the static thermal stress from 175° to 25°C, which does not really represent the stress state of the reflow and temperature cycle process. It could be very high and, ultimately, induce the delamination.
Conclusion
Based on the "fish bone" diagrams, the material, process and design aspects were evaluated by experiments and finite element analysis. The root causes of delamination at the lead frame paddle/mold compound interface were mainly die-attach paste voids, high-stress concentration and weak adhesion strength. Hence, the follow-up activities will concentrate on improving the die-attach paste voids and tooling up the "dimple" structure on the back of the lead frame so that the package becomes resistant to the high stresses at the interface. Also, a dynamic thermal stress model could be developed based on the static stress simulation, and a fracture mechanics methodology could be applied in the model to make the simulation more accurate.
Acknowledgements
The author thanks Philip Ho, Ph.D., Agere Systems, for performing failure analysis, Sumitomo Bakelite Singapore Pte. Ltd. and Ablestik Singapore Pte. Ltd., for providing the mold compound and die-attach paste, respectively, for this project.
For a complete set of references, contact T.Y. Lin, formerly of Agere Systems Singapore Pte. Ltd., e-mail at [email protected].
Illustration by Gregor Bernard