In recent years, due to the trend of miniaturization and the high performance of electronic products, the dimensions of integrated circuit shrink and the current that the circuit needs to carry keeps increasing, which increases the current density in the circuit dramatically, and so do the solder bumps. Therefore, there are numerous electromigration studies related to solder bumps. The damage mechanism of the solder bumps by electromigration is very similar with that of aluminum circuit. In 1999, Liu et al investigated electromigrationi in Sn-Pb solder bumps by sandwich structure.6 They found that there were voids at the cathode side after stressing under the current density of 1×105 A/cm2 at room temperature for 19 days. Besides, there were some hillocks formed at the anode side, as shown in figure
FIG. 2.3. (a) schematic illustration of Sandwich structure (b) SEM image of Sn-Pb solder stressed at 1×105 A/cm2 at room temperature for 19 days.
anode side, as shown in figure 2.3.
Because most of the solders constitute at least 2 different metals, the diffusion mechanism of solder bumps is something different from that of the circuit, made of pure copper or pure aluminum. Figure 2.4 shows the SEM images of hillocks formed in the pure Sn, Sn80Pb20, Sn70Pb30, eutectic Sn63Pb37, Sn40Pb60, Sn5Pb95, which are stressed under the current density of 105 A/cm2 for 40 h.22
FIG. 2.4. The hillocks formed in the alloy, which has different Sn, Pb composition, stressed at 105 A/cm2 for 40 h (a) pure Sn; (b) Sn80Pb20; (c) Sn70Pb30; (d) eutectic Sn63Pb37; (e) Sn40Pb60; (a) Sn5Pb95.
In SEM images, we can find that hillocks grew more easily at grains. Besides, we also find that the diffusion behavior is more active when the alloy is of eutectic composition, because they provide more diffusion interface in this region, so hillocks are more easily found in the eutectic phase region in Sn-rich alloy, as shown in figure 2.4 (b) and (c).
If we compare figures 2.4 (a) to (f), it is easy to find that the Pb-rich alloy possesses more resistance to EM in comparison with Sn-rich does. That is to say, for the Sn-Pb alloy, although Sn is the dominative factor for diffusion, the resistance to electromigration still depends on the microstructure of the solder.
Besides, Lee et al. investigated the migration of the Pb-phase in eutectic SnPb and Sn3.8Ag0.7Cu solders for flip chip solder bumps.
Figure 2.5 is the cross-sectional view of the eutectic SnPb solder stressed under the current of 1.5 Amp amd 120℃. One can see that Pb-rich component moves to the anode side and leaves voids at the cathode side due to electromigration and the Ni layer in UBM would melt due to the increase of temperature by joule heating. The reason is that when temperature is above 100℃, Pb is the dominative migration species. For lead-free solders, such as Sn3.8Ag0.7Cu, Lee et al. found that the extrusion was formed at the anode side, as shown in figure 2.6. Below the extrusion, Lee also found that Cu UBM would generate oxidation particles after current stressing.23
FIG. 2.5. Cross-sectional-view of SEM images of eutectic SnPb solder bumps stressed under the current of 1.5 Amp and 120℃. (a) o h; (b)20 h; (c)30 h; (d)39.5 h.
FIG. 2.6. SEM images of the cross-sectional view of Sn3.8Ag0.7Cu solder bumps stressed under the current of 1.5 Amp and 120℃.
(a) o h; (b)20 h; (c)110 h; (d)200 h.
2.6 The effects of electromigration on the interface
When soldering, the low-melting-point solder bumps would melt and react with the high-melting-point metal on the board side to form the joint, so the IMCs formed at the metal/solder interface becomes an important issue. Many papers reported that the IMCs at the interface affect directly or indirectly on the properties of the device, such as chemical properties, electric properties, etc.24-26
Huynh et al. designed a V-groove with two segment of Cu wire at the two sides, as shown in figure 2.7.27 He found that the eutectic SnPb sold
FIG. 2.7. (a) the structure of V-groove; (b) the cross-sectional view of the V-groove.
solder in theV-groove has different surface morphology evolution after stressing at different temperatures due to electromigration. As shown in figure 2.8, which is the SEM image of the solder bumps stressed for 8 days, there are hillocks formed at the anode side, but there are no depletion phenomena at the cathode side. However, after polishing to remove 10-µm-thick solder bump, it is easy to observe the voids formed at the cathode side, as shown in figure 2.8 (b). However, when the solder is tested under the room temperature, no voids and hillocks can be seen at the cathode side and the anode side, respectively, as shown in figure 2.8 (c).27
As to IMCs, they found that electromigration promotes the growth of IMCs. As shown in figures 2.9 (a) to (b), figure 2.9 (b) exhibits thicker Cu6Sn5 at the cathode side compared with the solder which is not stressed by current. For the interface at the anode side, the growth of the intermetallic compound is not as fast as the cathode side after stressing for 10 h.28, 29 From what has been discussed above, we may conclude that electromigration not only changes the distribution of the composition in the solder, but also changes the growth of the interetallic compound at the interface.
FIG. 2.8. (a) Surface morphology of V-groove after stressing (a) 2.8 × 104 A/cm2 at 150℃ for 8 days; (b) the void at the cathode side of V-groove; (c) 5.7 × 104 A/cm2 at room temperature for 12 days.
FIG. 2.9. The intermetallic compound formed at the cathode side and the anode side after stressing at 5.7 × 104 A/cm2 at 180℃ (a) o h at anode side; (b) 10 h at anode side; (c) o h at the cathode side; (d) o h at cathode.
2.7 The melting phenomenon in solder bumps
In general, the failure usually starts at the interface of the solder bumps and the circuit. Due to this phenomenon, Yeh attributed the failure mechanism to the current crowding. So, the current crowding phenomenon is another key issue to determine how the device is affected by the electromigration. Current crowding phenomenon usually happens at the corner when electrons flow from the circuit to the solder bumps.
That means the current density at this point increases dramatically, due to current crowding phenomenon.
Figure 2.10 shows the Cu UBM on the top of the solder bump has asymmetric melting phenomenon after current stressing.30 Hu pointed out that the Cu UBM would melt into solder bumps when stressing under the current of 1.27 A at 100℃, due to the local joule heating at the contact window. The Cu UBM on the chip side would be pushed into solder bumps to form IMCs after stressing for 15 mins, as shown in figure 2.10.
Few minutes later, the Cu layer of the UBM would be consumed gradually and the solder replaced the location which is occupied by Cu originally. Finally, the solder filled in the Cu circuit and the device failure occurred.
Hu also found that the Cu atoms are pushed from the cathode side to the anode side by electrons and reacted with Sn to form a large amount of Cu6Sn5 IMCs at the anode side, which is in deeper color in the solder bumps shown in the figure 2.10. On the other hand, the average dissolved rate of Cu atoms is calculated to be 1 µm/min by the relation between the length of the dissolved Cu trace and the stressing time, as shown in figure 2.11.
FIG. 2.10. The Cu UBM melted asymmetrically due to current crowding.
FIG. 2.11. The plot of length of the dissolved Cu trace against time of stressing time.
Chapter 3 Experimental
3.1 Fabrication of samples
An n-type, 4-inch Si (100) wafer was utilized as the substrate. After standard cleaning process, a 5000Å-thick SiO2 film was grown on the silicon wafer by wet oxidation method as the insulator. A Ti film of 1200 Å in thickness was deposited on the Si substrate by e-beam evaporation, followed by the deposition of a 5000 Å thick Sn film without breaking vacuum. Afterwards, the first-level mask was used to define Sn stripes and the selectively etching solution was FeCl3 + H2O at the ratio of 1:10.
The second-level mask was used to define Ti pads and the selective etching solution was NH4OH + H2O2 at the ratio of 1:5. As shown in figure 3.1, the three left-sided pictures and three right-sided pictures are plan view and cross-sectional view of specimens, respectively.
The schematic of the test samples is shown in figure 3.2, in which the Sn films on the left and right ends served as probing pads, and the center Sn film may deplete in the cathode end and grow hillocks in the anode end.
Sn
FIG. 3.1. The three left pictures and three right pictures are the procedure to prepare the testing samples, which are plan view and cross-sectional view, respectively.
FIG. 3.2. Schematic illustration of the cross-sectional view of the Sn Blech specimen. The direction of the electron flow is indicated by the arrows in the figure.
3.2 Sample analysis
The control of the applied current density is important in the electromigration study. Because of the rough surface of the evaporated Sn films, AFM (atomic force microscope) was used to measure the cross-sectional area of the cathode side of the Sn stripes to make sure that the applied current density in the Sn films was accurate, as shown in figure 3.3. The SEM (scanning electron microscope) image for the Sn test samples is shown in figures 3.4. Electric current was applied from the left side (anode) to the right side (cathode). The depletion area was measured by a commercial computer software. The power supply used in this study was a Keithley 2400 I-V source meter, which has a current resolution of 500 nA.
To measure the increase of temperature in the Sn stripe due to the current stressing, temperature measurement by infrared technique was performed by recording the temperature distribution (map) after the temperature reached a stable state.31 The temperatures in the Sn stripe during current stressing were mapped by Quantum Focus Instruments (QFI) thermal infrared microscopy, which has 0.1 ℃ temperature resolution and a spatial resolution of 2 µm.
um
Sn Ti
um
Sn Ti
FIG. 3.3. AFM (atomic force microscope) was used to measure the cross-sectional area of the cathode side of the Sn stripes to make sure that the applied current density was accurate.
FIG. 3.4. Plan-view SEM image of the fabricated sample. The Sn films on both ends served as probing pads, and the center Sn film was used to investigate the electromigration behavior.
Chapter4 Results and Discussion
Pure Sn stripes on Ti film were stressed under different current densities from 2.5 × 104 A/cm2 to 1.5 × 105 A/cm2 at room temperature (27℃~32℃ , 50 , 75 , and ) ℃ ℃ 100℃. The microstructure evolution, electromigration rate, and some important electromigration parameters, such as activation energy, and DZ* values were obtained in the experiment. Besides, we compared the phenomena observed in the study with those in other literatures.
4.1 Microstructure evolution
When electrons flowed from the cathode side to the anode side, Sn atoms migrated gradually in the same direction. Figures 4.1 (a) to 4.1 (f) show the depletion at the cathode side of the Sn stripes under different current densities at room temperature. We can clearly see that as the current density goes up, shorter time is required to form the depletion at the cathode side. We also found that the faster drift velocity (defined by depletion length divided by stressing time) we would get upon increasing the current density. The drift velocities for the current densities of 2.5 × 104, 5 × 104, 7.5 × 104, 1 × 105, 1.25 × 105, 1.5 × 105 A/cm2 are 0.013, 0.031, 0.070, 0.100, 0.136, 0.146 nm/s, respectively. (The depletion area was calculated by a commercial computer software “Inspector”) It is reasonable to explain that when more electrons were supplied from the cathode side of the Sn stripes under higher current density, more Sn atoms received momentum transferred form the electrons. At this time,
more
FIG. 4.1. Plan-view SEM images of the cathode side stressed at R.T. (a) 2.5 × 104 A/cm2 for 166 h; (b) 5 × 104 A/cm2 for 92 h; (c) 7.5 × 104 A/cm2 for 63.5 h; (d) 1 × 105 A/cm2 for 54 h; (e) 1.25 × 105 A/cm2 for 37 h; (f) 1.5 × 105 A/cm2 for 18 h.
more Sn atoms were pushed by the electrons and migrated from the cathode side of the stripe to the anode side.
At 50℃, as shown in figures 4.2 (a) to 4.2 (f), same phenomenon was found. At higher current density, shorter time is required to form the depletion and the drift velocity increases gradually. The drift velocities for the current densities of 2.5 × 104, 5 × 104, 7.5 × 104, 1 × 105, 1.25 ×
We have to mention that the depletion area in figure 4.3 (d) seems larger than the area shown in figures 4.3 (e) and 4.3 (f), but the time required for figure 4.3 (e) and 4.3 (f) was shorter than that for figure 4.3 (d). After calculation, the drift velocities for figures 4.3 (e) and 4.3 (f) are faster than that in figure 4.3 (d). Besides, the reason why we didn’t show the data of 1.5 × 105 A/cm2 at 100℃ is because the high current density might generate sufficient Joule heat at 100℃ that thermal migration starts to contribute to deplete the atoms at the cathode side. Since it is another issue, we discuss it elsewhere.
FIG. 4.2. Plan-view SEM images of the cathode side stressed at 50℃ (a) 2.5 × 104 A/cm2 for 91.5 h; (b) 5 × 104 A/cm2 for 63 h; (c) 7.5 × 104 A/cm2 for 40 h; (d) 1 × 105 A/cm2 for 28 h; (e) 1.25 × 105 A/cm2 for 24h; (f) 1.5 × 105 A/cm2 for 20 h.
FIG. 4.3. Plan-view SEM images of the cathode side stressed at 75℃ (a) 2.5 × 104 A/cm2 for 43 h; (b) 5 × 104 A/cm2 for 25.5 h; (c) 7.5 × 104 A/cm2 for 21 h; (d) 1 × 105 A/cm2 for 19 h; (e) 1.25 × 105 A/cm2 for 15h; (f) 1.5 × 105 A/cm2 for 12 h.
FIG. 4.4. Plan-view SEM images of the cathode side stressed at 100℃ (a) 2.5 × 104 A/cm2 for 25 h; (b) 5 × 104 A/cm2 for 12 h; (c) 7.5 × 104 A/cm2 for 12 h; (d) 1.25 × 105 A/cm2 for 6h.
Comparing the cathode of the Sn stripes stressed under the same current density but at different testing temperatures, we found that as the testing temperature goes up, faster drift velocity was obtain, as shown in figures 4.5 (a) to 4.5 (d). Figure 4.5 shows that the depletion of the cathode side of Sn stripes stressed under the current density of 1.25 × 105 A/cm2 at R.T., 50℃, 75℃, and 100℃. The stressing time for figures 4.5 (a) to 4.5 (d ) are 37 h, 25 h, 15 h, 6 h, respectively. It was found that the higher the stressing temperature, the faster the drift velocity. This phenomenon is reasonable and easy to explain. When we increased the testing temperature, Sn atoms had higher thermal fluctuation that they could overcome the barrier and move to the anode side than they were under lower testing temperatures. So, we can easily find that the drift velocities were faster, when Sn atoms were stressed under higher temperatures. The plot of the drift velocity against the applied current density at different temperatures is shown in figure 4.6. The drift velocity increased linearly with the increase of applied current density.
On the anode side, two types of whiskers, the hillock-type and needle-type ones, were observed. Figures 4.7 (a) to 4.7 (d) show the morphology on the anode side stressed under the current density of 5 × 104 A/cm2. It was found that the hillock-type whiskers could be grown at room temperature up to 100 ℃. However, the needle-type whisker can be only observed frequently on the Sn stripes stressed at room temperature, as shown in figure 4.7 (a). Only one needle-type whiskers was found on the samples stressed at 50 ℃, and no needle-type whiskers were observed on the samples stressed at 75 ℃ and at 100 ℃ As mentioned in . introduction, the needle-type whiskers grew when the Sn stripes were
stressed at room temperature
FIG. 4.5. Plan-view SEM images of the cathode side after the current stressing by 1.25 × 105 A/cm2 (a) ar RT for 37 h with a depletion area of 800 µm2. (b) at 50℃ for 28 h with a depletion area of 1164 µm2. (c) at 75℃ for 15 h with a depletion area of 15700 µm2. and (d) at 100℃ for 6 h with a depletion area of 960 µm2.
FIG. 4.6. Plan-view SEM images of the cathode side after the current stressing by 1.25 × 105 A/cm2 Threshold current densities can be obtained by extrapolating from the fitting curves to the zero drift velocity.
FIG. 4.7. Plan-view SEM images of the anode side after the current stressing by 5 × 104 A/cm2 (a) at R.T. for 37 h, both hillock-type and needle-type whiskers are formed; (b) at 50 ℃ for 28 h, hillock-type whiskers are formed (needle-type whiskers may be observed when the film was stressed longer);
(c) at 75 ℃ for 15 h, only hillock-type whiskers are formed.
(d) at 100 ℃ for 12 h, only hillock-type whiskers are formed.
stressed at room temperature. However, in our previous study, we found that the growth rate of whiskers was faster when the stripes were stressed at 50 ℃.16 The discrepancy may be owing to different stressing times.
The stressing times in our previous study were longer than 90 h. However, in this study, the stressing times were shorter than 91 h. Pure Sn stripes with longer stressing time may accumulate higher stress for breaking the surface oxide of Sn, and then start to grow whiskers. Moreover, when stressed at 75 ℃ and 100 ℃ no needle-type whiskers were observed , even when the stripes were stressed at 1.5 A/cm2 for 150 hours. The reason for that is not clear at this moment. This may be attributed to the softer surface oxide, resulting in the formation of the hillock-type whiskers at higher temperatures.32, 33
4.2 Threshold current density (Jc)
When applied current density is too small, we could not find any depletion at the cathode side due to electromigration. The reason for this phenomenon is because the low current density cannot supply atoms sufficient momentum to overcome the barrier to move toward the anode side. This parameter is so important that once the value is obtained, we would know that as long as the applied current density is below the threshold current density, the circuit will never be broken even if it were stressed for a long period of times.
We can obtain the threshold current density by extrapolating the four curves in figure 4.6 to zero drift velocity.12 As shown in figure 4.6, the four threshold current densities were 1.93 × 104, 9.65 × 103, 9.57 × 103 and 7.93 × 103 A/cm2 for R.T., 50 ℃, 75 ℃ and 100 ℃, respectively.
These values represent the maximum current density that the Sn film can carry without any electromigration damage. The higher the stressing temperature, the lower the threshold current.
4.3 Activation energy ( Ea ) and DZ*
To measure the activation energy, the increase in temperature due to Joule heating was first measured. Figure 4.8 shows the temperature increase as a function of applied current density measured on a hotplate at 75 ℃ and 100 ℃. It is found that the Joule heating caused an increase in temperature of approximately 1.3 ℃, when the sample was placed on a hotplate of 100 ℃ stressed by 1.5 × 105 A/cm2. Because there was no enough infrared emitted at R.T. and 50 ℃ for the temperature measurement, the temperature increase under current stressing at these
two
FIG. 4.8. The increase in temperature due to the applied current density detected by the Infrared technique.
two temperatures were assumed to be the same as that stressed at 75 ℃. By using the calibrated temperatures, the activation energy can be obtained by the slope in the plot of ⎟⎟ figure 4.9. Its value was calculated to be 0.32 eV (7366 cal/mole). In addition, the measured average values of DZ* were 1.95 × 10-10, 4.84 × 10-10, 1.27 × 10-9 and 1.99 × 10-9 cm2/s for R.T., 50 ℃, 75 ℃ and 100 ℃, respectively. These values are listed explicitly in Table I.
Sun and Ohring measured the product of DZ*, activation energy and other parameters for 16000-Å-thick Sn films by tracer-diffusion method.15 The values of activation energy Ea of 0.46 eV and DZ* ≈ 3.5 × 10-9 to 1.8 × 10-8 cm2/s were obtained when the Sn films were stressed by 1 × 104 A/cm2 at 142 ℃ to 213 ℃. The lower value of activation in the present work may be due to smaller grain size and lower stressing temperatures. The grain size of Sn film in the present work is expected to be smaller since its thickness was only 5000 Å. Blech observed a larger drift velocity in thinner Al film.12 Therefore, grain boundary diffusion may dominate in the thin Sn film, resulting in a smaller activation energy.
FIG. 4.9. The plot of ln ( VT / j ) against 1/T. Activation energy can be obtained by the slope of the fitting curve.
TABLE I. Testing currents, temperatures, corresponding drift velocities, and the
values of Z*.
4.4 Critical length
When the atoms at the cathode side are pushed to the anode side
When the atoms at the cathode side are pushed to the anode side