In Pair B case, the set of whole bumps were stressed at the current density of 1 × 104 A/cm2 until failure occurred. Afterwards, they were subject to cross-sectioning for failure analysis. Figure 3-8(a) shows the cross sectional SEM images of the solder bump along cross-section plane A. The bump failed after stressing for 140 h at 100℃, and was then ground, cut, and polished away approximately 1/3 volume of the solder bump. The Cu-Sn intermetallic compounds were accumulated at the anode/board side, which implies that the Copper atoms on the chip side and in the solder were caused to migrate by the electron flow to the board side.
To examine the microstructure beneath the cross-sectioned surface in figure 3-8(a), further polishing was performed. Figure 3-8(b) and (c) shows the microstructures after polishing away half the volume and two-thirds the volume of the solder bump.
Comparing both figures, crack appears to be larger as the bump was polished down.
Fig. 3-9 shows the cross-sectional SEM image of the solder bump along the cross-sectioned plane B. Solder bump failed after 42 h of current stressing at 150℃.
Because the test temperature was raised up to 150℃, the failure time decreased from 140 hours at 100℃ to 42 hours. Obviously, the voids at the right are much bigger than the ones on the left as the result of the current crowding effect.
(b)
Figure 3-8: Cross-sectional SEM image of the solder bump stressed at 100 ℃ for 140 hours, (a) after polishing to one-third of the bump volume, (b) after polishing half of the bump volume, (c) after polishing two-thirds of the bump volume.
Figure 3-9: Cross-sectional SEM image of the solder bump stressed at 150℃ for 42 hours.
3-3 Discussion
A. Current Crowding Effect on Bump Failure
In the case of the cross-sectioned bump, the solder bump failed after 408 hours of current stressing, as shown in Fig. 3-4(a) through (d). The corresponding schematic picture of the bump is shown in the left bump of pair A in Fig. 3-2(b), where the electrons crowded into the bump from the upper-left-back corner. At the early stage of current stressing, it was speculated that atoms around the corner were migrated to the board/anode side, and voids started to form at the point where current crowding occurred.
As voids aggregated to form cracks, the contact area on the chip side decreased, forming a vicious cycle and deteriorated the contact. As the cracks move toward the cross-sectioned surface, current density increased dramatically, heating up the bump locally because of Joule heating. It is believed that the local temperature at the failure site was ramped over the melting point of the solder before failure because there were tiny solder balls observed near the voids. Furthermore, the surface morphology of the void has a smooth appearance which also implies that the solder may have been in a molten state before the failure.
In Fig. 3-9, voids at the upper-right corner are much bigger than those at the upper-left corner. This may be the result of current crowding, where the electrons crowded into the bump from the upper-right corner of the bump. At the board side in Fig. 3-9, Cu-Ni-Sn IMC grew thicker at the lower-right corner, which was attributed by the current crowding on the board side.
B. Temperature Effect on Failure and Microstructure of the Bumps
The results for the measured resistance of the stressing circuit as a function of stressing time is shown in Fig. 3-10. Although the resistance included both the resistance of the conduction lines and that of the bump pair, it was still observed that the failure time depended on the dramatic increase of the resistance. As expected, the bump failed within 42 hours of stressing time at 150℃, and within 142 hours at 100℃
because the diffusivity of metal atoms are larger at high temperature. To examine the microstructure changes caused by thermal effect, reference bumps were employed to undergo the similar thermal history during current stressing. Figure 3-11 shows the microstructure changes resulting from the thermal effect only. The bump was kept at 150℃ for 42 hours, and no obvious microstructure change was observed .
C. Interface Analysis after Current Stressing
An enlarged image of the chip side of the bump in Fig. 3-8(c) is shown in fig. 3-12.
The crack was found at the solder/IMC interface. Above the crack, the EDS composition analysis at point 1 and point 2 indicated that there are Al, Ti, Cr, Cu and Sn in this region.
It is believed that Al, Ti, and Cr atoms migrated by the electron flow into IMC. Because Cu atoms were caused by the electron flow to migrate toward the board side to form a Sn-Cu compound, the analysis at point 3 and point 4 revealed that only Sn and Ag atoms were detected at this region. The composition of these points in the solder is shown in table 3-1.
Figure 3-10: Measured resistance of the stressing circuit as a function of stressing time.
Figure 3-11: Cross-sectional SEM image of the reference bump, the bump was kept at 150℃ for 42 hours.
Figure 3-12: Enlarged SEM image of chip side in Fig. 8(c).
Table 3-1. Composition of solder at point 1, point 2, point 3, and point 4.
3-4 Conclusions
The electromigration-induced failures in SnAg3.8Cu0.7 solder joints on Ti/Cr-Cu/Cu has been investigated under current density of 1 × 104 A/cm2 and 2 × 104 A/cm2 at 100℃ and 150℃. The bumps failed at chip/cathode side. Cracks occurred along the solder and UBM interface, which led to the open failure of the bump. The current crowding effect played an important role on the failure. However, the electron flow does not cause apparent damage at the board side because of a much lower current density there than that in the chip side.
Chapter 4: Electromigration in Pb-free SnAg
3.8Cu
0.7Solder Stripes
4-1 Sample preparation and experimental procedures
A similar process to fabricate Blech’s specimens of Al stripe is adopted to fabricate the solder specimens. The main difference here is that the solder Blech specimen was fabricated in a Si trench, in which the top Si surface served as a polishing stopper during the subsequent polishing process. A four-inch p-type Si wafer was cleaned by piranha solution (H2O2 and H2SO4 at the ratio of 1:7) for 10 minutes at 100℃. After cleaning, the silicon wafer was patterned by photolithography and deep reactive ion etching (DRIE) to form dumbbell-shaped trenches of 3.1 µm in depth. Then, a 1200 Å SiO2 insulating layer was grown on the wafer. Subsequently, titanium and copper films with thicknesses of 1700Å and 4000Å, respectively, were thermally evaporated onto the silicon wafer by an e-beam evaporator. Thereafter, the copper film inside the trench was patterned and selectively etched to form a short stripe, and two pads were also patterned on the Ti film, which served as a wetting metallization layer for the SnAgCu solder during the subsequent reflow process. Then the wafer was cut into small pieces, with a die in each piece. Solder paste of SnAg3.8Cu0.7 was applied to the Cu stripe at 230℃ for 2 seconds, in which the subscripts in SnAg3.8Cu0.7 stand for weight percent. After the reflow process, the thickness of the solder may be over 10 µm thick, and its shape was bump-like.
Therefore, a polishing procedure was needed to thin down the solder stripe. The thickness of the solder stripe was controlled by the thickness of the Si trench, since the top Si surface served as a polishing stop for the solder stripe.
Figures 4-1(a) and (b) illustrate the schematic diagrams of the tilted and cross-sectional views of the specimen, respectively. The two square pads at the two sides of the specimen were the electrodes, and the central solder stripe was the specimen to be studied. The stripe was 350 µm long and 80 µm wide. A layer of Cu6Sn5 intermetallic compound (IMC) grew between the Cu and the solder during the reflow, as shown in Figure 4-1(b). The thickness of the solder was about 1 to 2µm, and it varied from sample to sample due to the polishing process, which polished some of the Si stopper.
The samples were then stressed at various current densities and temperatures, and the direction of the electron flow is indicated by the arrows in Figure 4-1(a) and 4-1(b).
Transmission electron microscopy (TEM) and scanning electron microscopy (SEM) were employed to observe the microstructure of the solder stripes. Focused ion beam (FIB) was utilized to prepare cross-sectional TEM specimens. AFM was used to measure the depletion volume on the cathode side of the samples. Each specimen was scanned in AFM six times in order to measure the volume before and after the current stressing, the standard deviation was less than 1% compared with the average volume.
The temperature increment due to the Joule heating effect was monitored by an infrared microscope, which has 0.1℃ temperature resolution and 2µm spatial resolution.
4-2 Results
A. Microstructure of the SnAgCu stripe and temperature measurement
Figure 4-2(a) demonstrates the backscattered electron (BSE) SEM image of the fabricated solder stripe, formed inside the Si trench. Figure 4-2(b) is the enlarged (BSE) SEM image for one end of the stripe. Typically, the four corners of the Cu stripe could
Figure 4-1: (a) Tilted-view schematic of the solder stripe on a Ti film in a Si trench.
(b) Cross-sectional schematic of the solder Blech specimen. The direction
(a) e
Solder
Ti
Silicon
(b) e
SiO2Ti Cu6Sn5
Silicon
SnAgCu
Figure 4-2: (a) BSE SEM image of the fabricated dumbbell-shaped stripe. (b) Enlarged SEM image for one end of the stripe
Pad Pad
Si substrate
Solder stripe
100µm
(b)
not be wetted by molten solder paste during sample preparation. Figure 4-3(a) illustrates the cross-sectional TEM image of the specimen. A layer of scalloped Cu6Sn5 intermetallic compound (IMC) is formed between the SnAg3.8Cu0.7 solder and the Cu layer. The SiO2
and Ti layers can be clearly seen in the figure. The Cu film was consumed almost completely, but a few small Cu islands may be observed under some large IMCs. Since the Cu layer is not continuous, it is not considered when calculating the effective current density in the solder stripe. Figure 4-3(b) displays the enlarged cross-sectional TEM image of the area within the white circle in figure 4-3(a). The grain size was about 1 µm in diameter. We speculate that the minuscule grain size was due to the film thickness and the rapid cooling rate in the sample preparation process. Through theoretical calculations, it was found that about 80% of the applied current would flow inside the solder, whereas about 19% of the current drifted along the IMC layer, and only about 1% stayed in the Ti layer.
Figure 4-4 displays the temperature increment in the solder stripe as a function of current density at 80°C, 100°C, and 120°C measured by a infrared microscope. The highest temperature increment was merely 5°C when the specimen was stressed by 0.12 A, which corresponded approximately to 1×105 A/cm2 in the solder stripes. The Joule heating effect in the stripes was much lower than that in the flip-chip solder bump [48].
This effect may be attributed to the stripe geometry and the excellent heat conduction of silicon substrate.
B. Threshold current density of the SnAg3.8Cu0.7 solder
Figures 4-5(a) and (b) show the SEM images of the SnAg3.8Cu0.7 solder stripe in the
Figure 4-3: (a) Cross-sectional TEM image of the solder Blech specimen.
(b) Enlarged image of the white rectangular district in (a), with average grain size of about 1 µm.
Si (a)
1µm Solder
SiO
2Cu Ti Cu
6Sn
5(b)
0 20000 40000 60000 80000 100000 0
1 2 3 4 5
Applied current density (A/cm
2)
Figure 4-4: Measured temperature increment inside the solder stripe as a function of applied current for the three stressing temperatures.
80℃
100℃
120℃
T empe ratu re incr ement ( ) ℃
Figure 4-5: (a) Tilted SEM image at the cathode side before current stressing. (b) Tilted SEM image on the cathode side after current stressing by 8.67×104 A/cm2 at 80℃ for 65 hours. The SnAg3.8Cu0.7 solder was migrated by the electron flow, but the IMC remained intact. (c) Corresponding AFM image of (a). (d) Corresponding AFM image of (b).
(a) (b)
(c)
10 µm
10 µm
10 µm
10 µm
(d)
cathode end before and after the current stressing by 8.67×104 A/cm2 at 80℃ for 65 hours, respectively. After the current stressing, the SnAg3.8Cu0.7 solder near the cathode was depleted by the electron flow, and the intermetallic compound at the cathode side was exposed, as shown in Figure 4-5(b). Figures 4-5(c) and (d) show the corresponding 3-D AFM images for the solder stripes in figures 3-5(a) and (b), respectively. The AFM image in Figure 4-5(d) also shows the depletion of the cathode end, which demonstrates that the AFM could measure the depletion of the solder. The depletion volume for this specimen was estimated to be 799 µm3. On the other hand, hillocks were formed at the anode end of the stripe, as shown in figure 4-6. The composition of hillock is mainly Sn.
The average drift velocity of the solder stripe can be obtained by dividing the depletion volume (∆V) by the product of the average cross-sectional area and the stressing time. Figure 4-7 displays the average drift velocity as a function of applied current density, showing a linear relationship for the three temperatures. By extrapolating the fitting line to the zero drift velocity, the threshold current density can be obtained.
The estimated values are 4.3 × 104 A/cm2 at 80 °C, 3.2 × 104 A/cm2 at 100 °C, and 1.4 × 104 A/cm2 at 120 °C. These values represent the maximum current densities that the SnAg3.8Cu0.7 solder can carry without electromigration damage at the three stressing temperatures.
Figure 4-6: Plan-view BSE SEM image of the anode side (a) before the current stressing, and (b) after the current stressing at 80℃ for 65 hours. Hillocks are composed of almost pure Sn formed at the anode side.
(a)
(b)
Hillocks
Current density (A/cm
2)
Figure 4-7: Average drift velocity of the solder stripe as a function of applied current density. The threshold current densities were obtained by extrapolating the fitted lines to zero drift velocity.
0 30000 60000 90000 120000 0.0
0.5 1.0 1.5 2.0 2.5
3.0 80℃
100℃
120℃
Drift velocity (Å/c m
2)
C. Activation energy and effective charge number of SnAg3.8Cu0.7 solder
The average drift velocity due to electromigration, as given by Huntigton and Grone [38] is
where J is the atom flux, C is the density of metal ions, B is the mobility, k is Boltzmann’s constant, T is the absolute temperature, eZ* is the effective charge of the ions, ρ is the metal resistivity, j is the electrical current density, Ea is the activation energy of diffusion, and D0 is the prefactor of diffusion constant. Equation (1) can be rewritten as
) Taking the logarithm of both sides of equation (2)
k
Therefore, by measuring solder drift velocity as a function of reciprocal temperature, the activation energy Ea and the product of diffusivity and effective charge number DZ* can be obtained.
Fiigure 4-8 shows the plot of ln(vT/j) as a function of the reciprocal temperature.
The activation energy (Ea) can be determined from the slope of the fitted line, and its value is 0.41eV in the temperature range of 80℃ to 120℃. However, the temperature in the solder needs to be calibrated due to the Joule heating effect, as shown in Figure 4-4.
The real temperatures in the solder were higher than the ambient ones, and the activation energy was calculated to be 0.45 eV using the real temperatures.
In addition, the product of diffusion diffusivity and effective charge number, DZ*,
1/T (K)
Figure 4-8: Plot of the ln vT/j as a function of reciprocal temperature. The activation energy of 0.41 eV was obtained from the slope of the fitted line.
0.00252 0.00259 0.00266 0.00273 0.00280 0.00287 -24.4
-24.2 -24.0 -23.8 -23.6 -23.4 -23.2 -23.0 -22.8
ln(vT/j)
0.41 eV
can be calculated from equation (3). Table I summarizes the product of the diffusivity and effective charge number, DZ*, and its average values are -1.8×10-10 cm2/sec at 80℃, -5.0×10-10 cm2/sec at 100℃, and -7.2×10-10 cm2/sec at 120℃. In order to estimate the value of Z*, the diffusivity data for pure Sn were adopted [49].
The average values of Z* were -27 at 80℃, -33 at 100℃, and -23 at 120℃, which are reasonable for solder materials.
Table 4-1: The product of diffusivity and effective charge number (DZ*) for various stressing conditions.
Temperature ( )℃ Current density (A/cm2) DZ* (cm2/sec)
80℃ 1.0×105 -2.26×10-10
80℃ 8.4×104 -1.72×10-10
80℃ 7.1×104 -1.61×10-10
Ave. : -1.86×10-10
100℃ 9.0×104 -6.88×10-10
100℃ 6.2×104 -5.35×10-10
100℃ 4.4×104 -2.80×10-10
Ave. : -5.01×10-10
120℃ 7.5×104 -1.03×10-9
120℃ 5.7×104 -1.03×10-9
120℃ 4.0×104 -8.12×10-10
120℃ 2.0×104 -3.37×10-10
Ave. : -7.26×10-10
D. Effect of microstructure on electromigration in eutectic SnAg3.8Cu0.7 solder stripes
It has been recognized that electromigration lifetime is strongly correlated to the microstructure of the metal lines. A number of studies have been carried out to investigated the effect of microstructure on electromigration life time [51]-[63]. By annealing the metal lines at elevated temperature to induce grain growth, the electromigration lifetime was found to increase until grain growth was constrained by the metal line thickness.
To investigate the effect of microstructure on electromigration in eutectic SnAg3.8Cu0.7 solder, as-prepared eutectic SnAg3.8Cu0.7 solder stripes were pre-annealed on the hotplate at 150℃ for 72 hours in the atmosphere. Afterwards, the pre-annealing samples were ground and polished and were stressed at various current densities and temperatures.
Figure 3-10 illustrates the cross-sectional TEM image of the specimen after heat treatment at 150℃ for 72 hours. It is found that the solder grains grew bigger. Layers of scallop-type Cu6Sn5 IMC formed at the interface between the solder and Ti layer.
Specifically, a thin (Ti, Sn, Cu) mixed layer formed beneath the Cu6Sn5 intermetallic compound, and the thickness is about 30 nm. The average thickness of the IMC was 1.42 µm. Through theoretical calculations, it was found that about 80% of the applied current would flow inside the solder, whereas about 19% of the current drifted along the IMC layer, and only about 1% stayed in the Ti layer.
The microstructurs of the as-prepared and the pre-annealed eutectic SnAg3.8Cu0.7
solder are respectively shown in figures 3-11(a) and (b). The samples were etched in the HCl + CH3OH (at the ration of 1:50) solution for one second to reveal the grain size.
Ag3Sn particles precipitated in the matrix and decorated the grain boundary in the as-reflowed sample. After pre-annealed at 150℃ for 72 hours, coarsening of Ag3Sn particles and growing of (Sn) grains occurred. Ag3Sn particles grew from initial sub-micron size to about 1 µm, and the number of particles reduced. The grain size grew from 1 µm to 7-10 µm after the annealing.
Figure 3-12 displays the average drift velocity of the pre-annealed stripes as a function of applied current density, showing a linear relationship for the three temperatures. By extrapolating the fitting line to the zero drift velocity, the threshold current density can be obtained. The estimated values are 4.6 × 104 A/cm2 at 80 °C, 3.9 × 104 A/cm2 at 100 °C, and 2.2 × 104 A/cm2 at 120 °C.
Figure 3-10 Cross-sectional TEM image of the solder Blech specimen after heat treatment at 150℃ for 72 hours.
Figure 3-11. Plan-view SEM images of eutectic SnAg3.8Cu0.7 solder (a) as-prepared, and (b) annealed.
(a)
(b)
Figure 3-12. Average drift velocity of the pre-annealed solder stripe as a function of applied current density. The threshold current densities were obtained by extrapolating the fitted lines to zero drift velocity.
4-3 Discussion
A. Threshold current density of SnAg3.8Cu0.7 solder stripes
To verify if these extrapolated values of threshold current density are correct or not, some specimens were stressed at current densities below the threshold current density.
Specimens were stressed at the current density of 3.5 × 104 A/cm2 at 80°C, 2.6 × 104 A/cm2 at100 °C, and 1 × 104 A/cm2 at100 °C for 72 hrs, and no detectable volume change was found.
Various stressing conditions have been investigated to study the electromigration behavior of SnAgCu bumps, and a large variety of conditions were reported to have caused damage in the bumps. Wu et al. conducted a MTTF experiment for SnAg4.0Cu0.5 bumps with thin-film under-bump metallization (UBM) of Al/Ni(V)/Cu, and found that the MTTF was 1454 hours for the bumps stressed by 5.0 × 103 A/cm2 at 153°C [45]. In addition, Choi et al. reported that the eutectic SnAgCu bumps with Al/Ni(V)/Cu UBM failed after the current stressing at 2.25 × 104 A/cm2 at 140 °C for 132 hours [42].
However, Lin et al. investigated the current carrying capability of eutectic SnAgCu bumps with 6 µm Ni under bump metallization, and found that there was no obvious electromigration damage in the bumps after the stressing by 2.55 × 104 A/cm2 at 150 °C for 2338 hours [46].
The above discrepancies may be due to the serious current crowding in the
The above discrepancies may be due to the serious current crowding in the