Flip Chip Solder Joint Failure Modes
High Electric-Current Density Testing
BY CEMAL BASARAN, H. YE, D.C. HOPKINS, D. FREAR, and J.K. LIN
Electromigration in solder joints under high direct-current density is a reliability concern for future high-density microelectronic packaging and power electronics packaging. The trend in flip chips to increase I/O count forces interconnecting solder joints to be smaller in size and carry higher current density. Current density will increase further as chip voltage decreases and absolute current levels increase. Research on electromigration and thermomigration in solder joints is in the early stages. In the following experiments, 20 test vehicle flip chip modules were subjected to DC electric-current stressing. The stressing current level ranges from 0.5 to 1.5 A and leads to a current density in the solder joint from 0.4 to 1.2 × 104 A/cm2, depending on the cross-section area of solder joint. Two test modules were subjected to DC pulse current stressing at a level of 3~10 A. Fourteen test modules failed due to current stressing; four test modules were damaged due to re-polishing after the nano-indentation tests; and two test modules survived more than 3,000 hours of stressing. Table 1 shows the test vehicle number and applied current levels for each module tested.
Table 1. Test matrix of flip chip modules.
Motorola Corp. provided test vehicle flip chip modules. The test module has a dummy silicon die with an aluminum (Al) conductor trace. The silicon die is attached on a FR4 PCB through eutectic Pb37/Sn63 solder joints. Copper plates on the PCB provide the wetting surface for the solder joints. Under-bump metallization (UBM) on silicon die side is electroless Ni. Voids between the solder joints are filled with underfill between the silicon die and PCB substrate. The thickness of the Al trace is about 1 μm, the width is about 150 μm, and the height of the solder joint is 100 μm. The test module was cross-sectioned and finely polished toward the center of the solder joints before current stressing.
We tested two solder joints on each module. The solder joints on each test module are named in such a way that current always flows from the copper trace through solder joint A into the Al trace on the silicon die side, then through solder joint B out another copper trace. During the course of current stressing, the test modules were taken off-circuit for scanning electron microscopy (SEM) analysis. Because it is difficult to measure the temperature on a 100-μm solder joint directly, the temperature of the silicon die was measured during current stressing with a fine-tipped thermal-couple thermometer. In a coupled thermal-electrical finite element simulation, the temperature in the solder joint is close to the temperature on the silicon die in this test module.
Observed Failure Modes
During the experiments, 14 modules failed due to electrical current stressing, four modules failed during re-polishing after nano-indentation tests, and two modules survived more than 3,000 hours of current stressing without failing. Three types of failures modes were observed among the first group: 1. failure in solder joint due to high temperatures; 2. failure in the Al trace due to high temperature; and 3. failure in solder joint due to void nucleation and growth in the solder joint during current stressing.
The reason for the first type of failure is obvious as the Pb/Sn eutectic solder has a low melting point of 183°C. M31, M33, and M53 experienced this type of failure. M31 and M33 failed after 30 minutes of current stressing with a measured Si die temperature more than 200°C. The solder joints in these two test modules melted (Figure 1). Joule heating was generated in the Al trace because the solder joints have good wetting with both Ni UBM on the Si die side and Cu plate on the FR4 side. Therefore, Al trace contributed to most of the resistance. For M31 and M33, electromigration and thermomigration should not contribute to module failure. M53 survived 22.5 hours of current stressing. Initial stressing temperatures measured on the Si die were 150°C, then rose gradually to over 180°C.
Figure 1. Secondary SEM of M33 after failure: solder joint B.
M4, M5, M7, and M54 exhibited Type-2 failures, among which M4 and M7 were subjected to pulsed direct current (PDC) stressing. A curve tracer was used for pulse stressing. The pulse frequency is 120 Hz with a pulse width of ~80 µm seconds. Pulse shape depends on wiring impedance, but resembles a rectangular wave. The duty factor (defined as the ratio of time of the on-period to that of the whole pulse period) is calculated to be 0.96%. When M4 was subjected to 10-A peak PDC stressing, its resistance increased to infinity immediately. A SEM image shows that the damage of the module was in the Al trace and silicon die (Figure 2a). M7 was first subjected to 3 A of PDC stressing for 50 hours, then 5 A of PDC stressing for 57 hours, and finally 7 A of PDC stressing for 23 hours. SEM images taken after stressing show that there was no microstructural change in the solder joints. When M7 was subjected to 10 A of PDC stressing, it failed immediately. Figure 2b shows that the Al trace was the actual cause of module failure.
Figure 2. Failure in the Al trace and Si die; (a) M4, solder joint A; and (b) M7, solder joint A.
With 7 A of PDC, peak current density in the solder joints was 5 ~ 7 × 104 A/cm2, which is much higher than that applied in the DC current-stressing experiments. The effect of PDC on electromigration has been shown to depend on the frequency and duty factor. At low frequency, electromigration acts as if it were DC for the time “on,” and back diffusion may occur during the time off. In our experiment, PDC frequency is within the low-frequency regime. The reason that we did not observed any damage in the solder joint during PDC stressing when high-current density was applied might be due to the low-duty factor (0.96%) of PDC. On the other hand, high-peak current of PDC generates a great amount of heat in the Al trace, which leads to its failure when 10-A PDC was applied.
M5 and M54 also experienced Type-2 failures, although they were subjected to DC stressing. For example, the applied current on M5 was 1.5 A and the module failed immediately. SEM secondary images of the solder joint B before and after stressing are shown in Figure 3a, while Figure 3b shows that the Si die separated from the solder joint and underfill along the Al trace where tremendous heat was generated.
Figure 3. Secondary SEM of M5, solder joint B; (a) initial; and (b) after failure.
Modules M6, M14, M34, M41, M42, M51, and M56 experienced Type-3 failure. This mode of failure happened due to void nucleation and growth in the solder joint under current stressing. In the modules that experienced Type-3 failure, severe void nucleation was observed in the solder joints. The fact that void nucleation was on the Si die side and mass accumulation on the Cu plate side on the cross-sectioned solder surface indicates that the failure process in the solder joint is a combined effect of electromigration and thermomigration. In these modules, solder joint A (where the direction of electromigration is the same as that of thermomigration) always had more severe void nucleation than solder joint B (where the direction of electromigration is opposite to that of thermo-migration). Therefore, the degradation of solder joint A caused the ultimate failure of the module.
Void Nucleation in Solder Joints
To understand Type-3 failures, which are due to mass migration, both electro and thermo, it is important to analyze the void nucleation modes in solder joints during current stressing, and their relationship with the failure of these test modules. Besides the modules that experienced Type-3 failures, there were other modules that also underwent void nucleation, but never failed after 3,000 hours - when the test was terminated. Four void nucleation modes were observed in these solder joints: mode 1. voids nucleate and grow in the Ni UBM - solder interface; 2. voids nucleate and grow in the region near Ni UBM - solder interface; 3. growth of pre-existing voids; and 4. no void growth after 3,000 hours of testing.
For Mode 1, voids were observed to nucleate and grow on the Ni UBM-solder interface for the majority of the solder joints. This interface was the favorite site for void nucleation and growth. Combined electromigration and thermomigration effect leads to an atomic-flux divergence in this region for both solder joint A and B, meaning the depletion of mass in the region, since Ni UBM acts as a barrier layer for the diffusion of solder joint. The direction of overall diffusion due to the combined effect of thermomigration and electromigration is from Ni UBM-solder interface to Cu plate in both solder joint A and B. Theoretical electromigration analysis indicates that maximum tensile spherical stress will be generated in this region and vacancy condensation will also occur in this region. The driving force for void nucleation and growth is proportional to the tensile stress. When numerically calculated, the void nucleation rate in passivated interconnect lines due to electromigration and thermal stress are based on vacancy condensation theory. It has been suggested that void nucleation by vacancy condensation is extremely slow and would not be expected to occur in reality. One expert proposed the possibility of contaminants at the metal/passivation interface acting as void nucleation sites in passivated metal lines. Others analyzed the effect of contaminants on void nucleation, and found that void formation at a flaw at the interface would require a considerably smaller stress than that in classical void nucleation theory. It was further concluded that voids would grow only at the intersection of the grain boundary with the passivation layer due to the large difference between the grain boundary and lattice diffusivities. For void growth to occur, atoms must be removed from the void surface and a grain boundary acts as an extremely fast path for material removal relative to the lattice. It was demonstrated that heterogeneous nucleation at the triple junction of a second phase particle and a grain boundary was the most probable. Based on these discussions, it is clear that the Ni UBM-solder interface is the naturally preferable site for void nucleation and growth, as observed in our experiments. Voids would nucleate at the interface of Ni layer-solder joint intermetallic compound. Contaminants lodged in this interface during the manufacturing process make this a preferred location for void nucleation. This interface has maximum atomic divergence, which is favorable for void growth. Figures 4 and 5 show examples of Mode-1 void nucleation and growth on the Ni UBM-solder joint interface.
Figure 4. Secondary SEM of M12, solder joint A: 36 hours.
Figure 5. Secondary SEM of M42, solder joint A: 178 hours.
Only three solder joints were observed to have Mode-2 void nucleation and growth. Figure 6a shows the void nucleation in the region near Ni UBM-solder joint interface in M34, solder joint A after 268 hours of current stressing. Hillocks were observed to build up near the solder-Cu plate interface. Figure 6b shows that void growth and severe depressions in the region near Ni UBM-solder interface developed after 444 hours of current stressing. One unexpected observation was that new voids nucleated in the region of the hillocks. The origin of these new voids is unclear because they were in the downwind region of thermomigration and electromigration, into which atoms diffuse and the material experiences compressive stress. The theory of void nucleation and growth under tensile hydrostatic stress does not apply in this region. One possibility of this behavior is that the void nucleation and growth occurred under shear stress. Shear bands are the preferred sites for nucleation, growth, and coalescence of voids, which are precursors to failure in titanium and Ti-6Al-4V alloys. In the hillocks region of solder joint A in M34, the material was subjected to biaxial compression stress according to the electromigration theory. However, in the direction perpendicular to the solder surface, the normal stress is zero. Therefore, the material in this region is also subjected to shear stress, which might be the cause of the new void nucleation.
Figure 6. Secondary SEM of M34, solder joint A; (a) 268 hours; and (b) 444 hours.
Voids nucleated in the hillocks regions shrunk further after current stressing. This might indicate that after the voids in the hillocks region nucleate and grow to a certain extent, the stresses triggering this void growth are counterbalanced by mass accumulation. Because the atoms diffuse into the hillocks region, healing of previous voids was observed. On the other hand, the depression near the UBM-solder interface continued to grow. The SEM image after failure indicates that the direct cause of failure was the severe depression and void growth in the UBM-solder interface region. It is worth noting that M34, solder joint A was the only solder joint with void nucleation in the hillocks region. Void nucleation and growth in the region near Ni UBM-solder interface was also found in solder joint B of M41 and solder joint B of M56. However, they were less severe compared to solder joint A of M34 because the direction of thermomigration was opposite that of electromigration in these solder joints. Of all the tested solder joints, only three had Mode-2 void nucleation and growth behavior, indicating that void nucleation in the region near UBM-solder interface is less favorable than that on the UBM-solder interface itself.
Some of the solder joints we tested had pre-existing voids, which are produced generally during the reflow process. Some of these pre-existing voids lead to Mode-3 void growth where the growth of pre-existing voids causes the ultimate failure of the test module. Experiment results show that whether or not these pre-existing voids would grow depends on their location - if pre-existing voids are located in the region near Ni UBM-solder interface out of which atoms diffuse due to the combined effects of thermomigration and electromigration, they likely will grow. If the voids are not located in the UBM-solder interface region, they are unlikely to grow.
There were several pre-existing voids on the cross-sectioned surface of solder joint A of module M56. One small void with irregular shape was located in the region near UBM-solder interface; and several others were located near the Cu-solder interface - two of them with round shape and others with irregular shapes. Only the pre-existing void in the region near the UBM-solder interface grew dramatically to form a large crack in that area. On the other hand, larger voids near the Cu-solder interface did not grow much, no matter their initial shapes. They did, however, change shape slightly, possibly due to local stress build-up and surface diffusion. Despite severe void growth in the region near the UBM-solder joint interface, two hillocks were built up gradually on the surface, and a depression area was formed between these hillocks.
This observation indicates that the diffusion process was not homogeneous within the solder joint. Careful examination of the phase structure reveals that the Pb-rich phase in the depression region was not equiaxially shaped and had a preferred orientation compared to that in the hillocks region. This preferred Pb-phase orientation was formed during manufacturing and preserved during current stressing. Experts believe that the effective boundary diffusion coefficient, Da, equals δDgb/d, where δ is the grain boundary width, Dgb is the grain boundary diffusivity, and d is the average grain size. Although the Pb-phase size is not the actual grain size, there should be a proportional relationship between this phase size and grain size - the larger the phase size, the larger the grain size. The orientation in-equiaxed phase structure indicates an orientation in-equiaxed grain structure, which leads to the difference of the average grain size in different directions. This means that the effective diffusivity may not be isotropic in this region; and therefore, the diffusivity in the entire solder joint is heterogenous. This observation suggests that the phase structure of eutectic Sn/Pb solder joint affects its diffusion property, and the failure process under current stressing.
Figure 7a shows an example of the growth of pre-existing voids. Pre-existing voids in solder joint A of M41 were located near to the Ni UBM-solder interface, and were produced during manufacturing. Pre-existing voids were observed to grow rapidly toward the UBM-solder interface and lead to the ultimate failure of the module (Figure 7b). Among the modules tested, some never experienced void nucleation and growth after 3,000 hours of current stressing.
Figure 7. Secondary SEM of M41 solder A; (a) initial; and (b) failure after 61 hours.
M8, M15, and M26 were subjected to 0.5 A of DC-current stressing. Relatively low current levels lead to low-current density in the solder joints, as well as smaller joule heating within the Al trace. Therefore, both the electromigration and thermomigration were less severe in these modules than solder joints in other modules where higher levels of current were applied. Current density is calculated based on the estimated cross-section area in the solder joints of M8 and M26 range from 0.57 ~ 0.75 × 104 A/cm2. Pb-phase growth was observed in the solder joints in both modules. Pb-phase coarsening indicates that electromigration and thermomigration were operative in these solder joints during current stressing. Lower current density and lower joule heating led to longer incubation times for void nucleation. Therefore, the voids in these solder joints did not have enough time to nucleate to an observable size. Conversely, when current density and stressing temperatures are low, the effect of migration may become almost invisible, even when the modules are stressed for an extremely long period of time. This was the case in M15, where the estimated current density was 0.4 × 104 A/cm2 and the stressing temperature was 40°C. No micro-structural coarsening was observed in solder joint A of M15 after 3,000 hours of current stressing. This indicates the possible existence of a threshold current density under which no electromigration failure would occur. This observation also agrees with earlier findings from electromigration experiments of thin pure-metal film.
Our experiments detected three different failure modes: one was due to the melting of the Al trace in the Si die; another was due to solder joint melting. Both of these failure modes resulted from joule heating and a lack of cooling in the specimen. These failure modes happen quickly, and the influence of electromigration and thermomigration is negligible. Most modules experienced electromigration- and thermomigration-induced failures. In this mode of failure, thermomigration dominates the failure process significantly. When the direction of the thermomigration is the same as that of the electromigration, the damage is more severe. When the direction of the electromigration and thermomigration are opposite each other, the thermomigration forces were larger. The Ni UBM-solder joint interface was the preferred site of void nucleation and growth. It is believed that the contaminants in the interface also accelerate the void nucleation process. The effects of pre-existing voids on the failure process of a solder joint depend on their location.
This project was sponsored by a grant from the U.S. Navy, Office of Naval Research, Advanced Electrical Power Systems, under the supervision of Terry Ericsen. For a complete list of references and figures, please contact the authors.
CEMAL BASARAN, PH.D., professor and director, may be contacted at University at Buffalo, SUNY, Electronic Packaging Laboratory, 243 Ketter Hall, North Campus, Buffalo, NY 14260; 716/645-2114 Ext. 2429; e-mail: email@example.com.