Microstructure Evolution in Six-Micro-Meter Microbumps

In the previous section, the concave-down curve indicates the formation of high-EM-resistance structure in 6-μm microbumps, which is very likely the IMC, Ni3Sn4. This inference could be proven by the SEM images (the cross-sectional view of the microbumps tested under 4.6 × 10⁴ A/cm² on the 150ºC hot plate) in Figure 3-8.

As can be seen, the images are marked as a1 to a3, b1 to b3, and c1 to c3, with the letter referring to the sample and the number denoting the microbump tested. Samples

“a,” “b,” and “c” were 6-μm microbumps tested for 49.8 hr, 321.6 hr, and 1961.8 hr, respectively; while microbumps “1,” “2,” and “3” were tested under no current stressing, upward electron flow, and downward electron flow, respectively. After EM testing, all the images obtained showed obvious Ni3Sn4 growth, which corresponded with the inference stated in section 3.2.1. The formation of Ni₃Sn₄ was a very important issue. The electrical resistivity of Ni3Sn4 is 28.5 μΩ-cm, which is the highest resistivity in the Cu-Ni-Sn reaction system. It is 2 times higher than that of Sn2.5Ag, 13.0 μΩ-cm, 4 times higher than that of Ni, 6.8 μΩ-cm, and almost 17 times higher than that of Cu, 1.7 μΩ-cm. Besides having high resistivity, Ni3Sn4 was also brittle, so the transformation may lower the reliability under structural testing.

Although the concave-down bump resistance curve revealed clearly the formation of high-EM-resistance structures, there were some unique phenomena worth mentioning.

After EM testing for 49.8 hr, some Sn2.5Ag was found in the microbump without current stressing, as shown in Figure 3-8 (a1). Ni UBMs on both chip side and interposer side were not seriously consumed, and the initial scallop-shaped Ni₃Sn₄ between solder and Ni became plate-shaped [2]. Unlike the non-stressed one, microbumps tested by the upward and downward electron flow showed obvious polarity effect. The cathode-side UBMs driven by EM were quickly dissolved into the

-78-

solder and formed IMC. In Figure 3-8 (a2) and (a3), over 80% of the solder was transformed into Ni₃Sn₄, which possessed higher EM resistance than the solder.

However, the bump resistances in Figure 3-6 took almost 500 hr to reach the long-time constant value, implying that the transformation from solder to Ni₃Sn₄ took around 500 hr to complete. It seemed that these two results, the fast phase transformation and the slow bump resistance increase, went against each other.

However, such was actually possible for two reasons. First, the distribution of Ni3Sn4 affects the increase in bump resistance; and second, the Sn grain orientation has a significant impact on the diffusivity of Ni in Sn. In Figure 3-8 (a2) and (a3), Ni3Sn4 was not found to be of plate-shaped. Some residual Sn2.5Ag was dispersed in the Ni3Sn4, preventing the bump resistance of 6-μm microbumps from immediate increase [77]. The bump resistance stopped increasing when the transformation completed. The Sn grain orientation affects more the diffusivity of Ni in Sn than the distribution of Ni₃Sn₄ [78]. At room temperature, the diffusivity of Ni in Sn with parallel grain orientation is 2.04 × 10⁶ times higher than that with perpendicular grain orientation (D║, RT = 1.41 × 10^-5 cm²/sec and D┴, RT = 6.91 × 10^-12 cm²/sec). At 100°C, the difference is still 1.20 × 10⁵ times (D║, RT = 5.84 × 10^-5 cm²/sec and D┴, RT = 4.85 × 10^-10 cm²/sec). In microbump samples, the pitch was lowered to 30 μm to increase the I/O density, so the diameter was only 18 μm. In this circumstance, only one to two grains could be found in a microbump. Therefore, the difference in diffusivity caused by Sn grain orientation affected the Ni diffusion in Sn and the consequent Ni3Sn4

growth.

Figure 3-8 (b) shows the microbumps tested for 321.6 hr. In Figure 3-8 (b2) and (b3), no residual Sn2.5Ag was observed in the microbumps at this stage of EM testing.

Similar to the polarity effect found in Figure 3-8(a2) and (a3), the cathode UBM

-79-

became non-uniform. The Ni atoms were driven by EM to the anode side so the anode-side IMC was thicker than the cathode-side, and the crack formed near the anode side. This phenomenon was also found in the sample tested for 1961.8 hr.

Whether the microbumps have upward or downward electron flow, cracks were observed between the IMC. However, the crack did not expand smoothly because the crack was not caused by external stresses. Instead, they were caused by the reaction between Ni and Sn. The molecular volume of Ni₃Sn₄ was smaller than the sum of Ni and Sn. The volume shrank by 10.5% after the reaction between Ni and Sn [79]. In view of the previous observations, it can be concluded that the microstructure evolution of 6-μm microbumps under EM testing involved the following four changes.

(1) In the initial stage of testing, the Ni atoms driven by EM quickly diffused into Sn2.5Ag and formed Ni₃Sn₄. The fast-growing Ni₃Sn₄ on the chip side and on the interposer side then came into contact, and the residual Sn2.5Ag dispersed between IMCs. (2) The residual Sn2.5Ag continued reacting with the Ni atoms and then formed voids between the chip-side IMC and the interposer-side IMC. Once several voids became connected, they looked like a crack between the IMCs. (3) Owing to the high EM resistance of Ni₃Sn₄, such status remained for a long time until the Ni cathode-side Ni ran out. (4) After Ni became insufficient, the cathode-side Cu joined the reaction and turned Ni₃Sn₄ into (Ni, Cu)₃Sn₄, (Cu, Ni)₆Sn₅, and Cu₃Sn. The resistivity of Cu6Sn5 and Cu3Sn are only 16.5 μΩ-cm and 8.5 μΩ-cm, which are only two-thirds and one-third that of Ni3Sn4, respectively. As a result, the bump resistance of sample 3 showed an obvious drop after being tested for 1000 hr. The volume shrinkage during Ni3Sn4 formation caused not only the crack between IMCs but also the big void beside the joint. Ouyang et al. reported some very similar results in 2012 [80]. In their study, some microbumps, after transforming into IMC bumps, lasted for

-80-

over 4000 hr under 8 × 10⁴ A/cm² at 150°C. However, they did not use Kelvin bump structures to monitor the bump resistance, so they did not observe the correlation between the concave-down bump resistance curve and the microstructure evolution under EM testing. This difference shows the power of Kelvin bump structures and the importance of monitoring the bump resistance of a single bump.

Under severe stressing condition, 9.2 × 10⁴ A/cm² on the 150°C hot plate, the correlation between the bump resistance and the microstructure evolution did not change. The bump resistance curves are shown in Figure 3-7, and the microstructures are shown in Figure 3-9. As can be seen, only one microbump, the bump with upward electron flow in sample 5, shows concave-down bump resistance. All the others showed concave-up bump resistance. Comparing the microstructures shown in Figure 3-9 reveals a series of voids found in most of the EM-tested microbumps, with the exception of the microbump with upward electron flow in sample 5.

-81-

Figure 3-8 6-μm microbumps in samples EM tested for (a) 49.8 hr, (b) 321.6 hr, and (c) 1961.8 hr by 0.12 A (4.6 × 10⁴ A/cm²) on 150°C hot plate.

-82-

Figure 3-9 6-μm microbumps in samples EM tested for (a) 141.3 hr and (b) 192.3 hr by 0.24 A (9.2 × 10⁴ A/cm²) on 150°C hot plate.

-83-

3.2.3. Bump Resistance at Different angles in Six-Micro-Meter