3.1. The Growth of Interface IMC during Aging
This section mainly discusses the morphology and evolution of IMC after reflow and aging with different surface finishes. As mentioned above, much literature [
21,
22,
23,
28] has studied the IMC composition of standard ENEPIG and OSP. In this section, the composition changes in IMC of different surface finish substrates are calculated by EDS analysis (TESCAN, Brno Czech Republic) and summarized in the cross-section of the IMC evolution process. This paper focuses on the changes in IMC composition during the interface reaction of the ultrathin Ni(P) layer ENEPIG, which has previously been studied less.
For OSP substrates, because there is no Ni layer barrier during the reflow process, copper atoms react with Sn in the solder at the interface to form IMC. A part of the copper involved in the reaction derives from the solder, and the other part derives from the diffusion of copper in the pad. The IMC interface reaction process of the OSP substrates is shown in
Figure 3.
Firstly, the uneven Cu6Sn5 intermetallic compound was formed at the interface. With the extension of aging time, the thickness of Cu6Sn5 increased, the morphology changed from scallop to plane layer and a new IMC Cu3Sn layer was gradually formed below Cu6Sn5. The interface of the Cu6Sn5/Cu3Sn/Cu sandwich structure was formed, which is in line with the results of previous experiments. As Cu3Sn possesses great brittleness compared with solder and other types of IMC, the influence of Cu3Sn on the properties of solder joints will be discussed in subsequent chapters.
The reaction process of the interface IMC layer during aging of ultrathin ENEPIG is shown in
Figure 4 and
Figure 5. As the gold layer and palladium layer on the ENEPIG substrate will rapidly diffuse into the solder during reflow to form dispersed IMC, the main factors affecting the growth of IMC at the interface are the diffusion of Sn-Cu and the barrier effect of the Ni(P) layer.
The IMC on the ultrathin ENEPIG substrate presents fibrous protrusion in comparison with the scallop shaped IMC on OSP substrate. In addition, the overall morphology of IMC tends to be flat during the progress of interface reaction, with the thickness gradually increasing.
In the experiment, the thickness of the Ni(P) layer of ENEPIG 1 and ENEPIG 2 was very thin, being only about 0.1 μm. During the reflow process, the Ni(P) layer reacts rapidly with the liquid solder, making it difficult to observe the existence of the Ni layer after reflow. However, without the diffusion of the P atoms in the Ni(P) alloy layer, the Sn diffused from the solder to the Ni(P) layer will react with the remaining Ni(P) layer to form Ni–Sn–P, as shown in
Figure 4c and
Figure 5c. After 120 h of aging, a thin layer of discontinuous flake IMC was formed on the interface between ENEPIG 1 and ENEPIG 2, and the IMC was (Cu, Ni)
3Sn through EDS composition analysis, as shown in
Figure 6 and
Figure 7. Furthermore, (Cu, Ni)
3Sn gradually tended to be continuous and thicker as the aging experiment progressed.
The energy spectrum analysis results of the interface IMC of the 120 h ENEPIG 2 are summarized in
Table 2. The IMC of points A and B was approximately (Cu, Ni)
6Sn
5, and the IMC of point C was (Cu, Ni)
3Sn. In
Table 2, it can be found that a small amount of P and Pd was detected at points A and B, separately. The blocking Ni–Sn–P layer may further react with the solder and be consumed, leading to the detection of the small amount of P. Moreover, Pd rapidly diffused into Sn in the reflow process to form a dispersed distribution, which explains the reason for the existence of Pd. During the progress of the interface reaction, Pd can diffuse into the interface IMC.
When the aging time reached 256 h, after calculation, the IMC of point A and point B remained as (Cu, Ni)6Sn5, however, the content of Ni in the newly generated IMC of point C was only 0.1%. Due to the massive consumption of the Ni–P layer, the Ni that participated in the reaction stably exists in the previously generated IMC with little Ni diffusing to the interface for the reaction.
As the Ni(P) layer has an inhibitory effect on Cu diffusion, the thinner Ni(P) layer deposited on ENEPIG 1 and ENEPIG 2 was quickly consumed during reflow and aging, which weakened the inhibitory effect on Cu diffusion. As a result, there exists a brittle (Cu, Ni)
3Sn IMC phase below (Cu, Ni)
6Sn
5 in ultrathin ENEPIG, which is similar to that observed for the OSP substrate. In contrast, ENEPIG 3 had a strong inhibitory effect on Cu diffusion due to the deposition of a thick Ni(P) layer, and there was still no discontinuous brittle (Cu, Ni)
3Sn at 500 h, as shown in
Figure 8.
The aging time was extended to 1000 h for further observation of the IMC growth of the three ENEPIGs and the OSP under long-term reaction conditions, and the results are shown in
Figure 9.
After long-term aging, the surface morphology of IMC with different surface finishes tended to be flat, and the IMC at the ENEPIG interface gradually changed from thin rod to layered. From the perspective of IMC growth thickness, the total thickness of OSP and ultrathin ENEPIG IMC was similar, while the brittleness and (Cu, Ni)3Sn IMC thickness of the ultrathin ENEPIG interface were much smaller than those of OSP. Therefore, although the inhibitory effect on Cu diffusion was gradually weakened due to the depletion of the Ni(P) layer, the Ni–Sn–P layer formed by the reaction still inhibited Cu diffusion.
For ENEPIG with the standard Ni layer thickness, the IMC layer with the smallest total thickness was formed after 1000 h, in which the continuously layered, brittle (Cu, Ni)3Sn was still not found, and there was still no completely consumed Ni(P) layer at the interface. Therefore, ENEPIG with standard Ni layer thickness has the best inhibitory effect on Cu diffusion.
Sn solder balls were corroded by using an alcohol solution of 4% HNO
3 + 1% HCl, and the upper surfaces of IMC at different surface finish interfaces after aging for 64h were observed. The SEM results are shown in
Figure 10. The observation results of the upper surface of the interface IMC layer are consistent with that of the cross-section of the interface IMC during the previous aging process. Cu
6Sn
5 IMC that is formed at the OSP interface presents a massive bulge, whose growth is relatively uneven. For ultra-thin ENEPIG, fibrous (Cu, Ni)
6Sn
5 IMC with a random extension direction appears; for ENEPIG with a high Ni layer thickness, the fibrous result is not entirely obvious due to the slow growth of IMC.
The interface reaction processes of several surface finishes were sorted out, and IMC growth models of OSP, ultrathin ENEPIG and standard ENEPIG were proposed, as shown in
Figure 11. The element diffusion rate at Sn solder/pad interface was the main factor affecting the IMC evolution of different surface finishes.
As there was no Ni barrier layer in OSP, the scallop shaped Cu6Sn5 IMC was formed by a rapid reaction between Sn and Ni after refluxing. With the extension of aging time, the morphology of Cu6Sn5 IMC changed from scallop to plane layer, and a new IMC Cu3Sn was formed below Cu6Sn5. Fiber-like (Cu, Ni)6Sn5 IMC was formed in ENEPIG due to the existence of a Ni barrier layer, and a discontinuous brittle (Cu, Ni)3Sn IMC layer gradually appeared with the aging reaction. For ultrathin ENEPIG, the thickness of the electroless Ni(P) layer was less than 1 μm. Ni was rapidly consumed during reflow and short-term aging, however, the thin Ni–Sn–P layer formed by interfacial reaction still possessed a partial blocking effect on Cu diffusion.
3.2. Kinetic Study of IMC Growth
The relationship between IMC layer thickness and aging time can generally be expressed by the following equation [
29]:
where T is the thickness of IMC, D is the diffusion coefficient, t is the aging time, and n is the power-law index.
For the surface treated samples with different Ni layer thicknesses, the IMC thickness of the interface after different aging stages over 0–1000 h was measured, and the relationship between the change in IMC thickness and time was analyzed. The results are shown in
Figure 12.
During the aging process, the thickness growths of all surface finishes of IMC were parabolic, which conforms to the law that the interface reaction is controlled by substance diffusion. Due to the addition of the Ni layer for blocking, it can be seen that the total thickness of the IMC layer and the brittle IMC layer of the ultrathin ENEPIG and ENEPIG decreased, and the brittle IMC layer never appeared in ENEPIG, which has the best effect on suppressing the diffusion of Cu at the interface.
Through the IMC growth thickness and the square root of aging time curves, two distinct and obvious stages of IMC growth for OSP and ENEPIG can be observed, which both meet the linear relationship between thickness and square root of aging time. From the previous discussion, we know that these two stages are the independent growth stage of Cu6Sn5 and the competitive stage of Cu6Sn5 consumption and Cu3Sn growth, separately. However, in the initial aging stage, a small step with a very low IMC growth rate appears in ENEPIG, which is also due to the blocking effect of Ni layer.
3.3. Shear Test Results
After the shear test of solder joints with different surface finishes after reflow and aging, there were four different failure modes, as shown in
Figure 13:
It is generally believed that the fracture position that appears at the IMC interface is a brittle fracture with a lower shear strength, while the fracture position that appears inside the solder ball is a ductile fracture with a higher shear strength. The solder joints with a ductile fracture have better interface bonding properties. The failure modes of different surface finishes under different aging times were counted, and the results are shown in
Figure 14.
Although the Ni layer of ultra-thin ENEPIG reacted quickly, its brittle fracture mode appeared later. This is due to the secondary barrier effect of the Ni–Sn–P layer, which makes it better than OSP but worse than ENEPIG in terms of fracture mode. Compared with electroplated NiAu samples, the brittle fracture ratio of ultrathin ENEPIG after 1000 h of aging is slightly lower than that of electroplated Ni/Au, leading to a better fracture mode than that of electroplated NiAu. By comparing the two ultrathin ENEPIG fracture modes, no obvious difference was found in the experiment.
During aging, not only will the failure mode of the shear test gradually change, but the shear strength between the welding ball and surface finish will also change under different aging times. The summary is shown in
Figure 15.
With the progress of the aging process, the shear strength of all surface finishes showed a trend of first increasing and then decreasing. In the initial stage of reflow, the interface reaction time is very short, and the IMC growth is insufficient. At this time, the bonding is weak, and the shear strength is low. As the aging process progresses, the IMC layer continues to grow and the bonding capacity between the interfaces gradually increases from the initial weak bonding to the highest at about 16 h. However, IMC itself is quite brittle compared with solder. The overall brittleness at the interface increases with the growth of IMC, resulting in a decrease in shear strength. This phenomenon is consistent with the results of fracture mode.
Comparing the shear strength curves of each surface finish under different aging times, it can be found that ENEPIG and Ni/Au with a Ni layer have higher shear strength than OSP. On the one hand, the Ni layer inhibits the growth of IMC and weakens the influence of brittle IMC. On the other hand, Cu-Ni-Sn ternary IMC has higher mechanical strength than Cu-Sn. In addition, the fibrous tubular IMC will increase the roughness of the interface and the surface area of the solder and IMC contact, forming more anchor points to hinder the relative movement between the solder and the IMC interface, thereby generating greater resistance during the shearing process. Finally, the interface bonding ability increases.
3.4. Drop Test Results
This paper used drop experiments to evaluate the impact resistance of different surface finishes. The experimental results are shown in
Table 3. The total number of drop experiments was designed to be performed 30 times according to the actual application scenarios of the device.
The results showed that ENEPIG 2, ENEPIG 3 and OSP all passed the drop test, but ENEPIG 1 and Ni/Au failed in this test.
We can observe the interface of the drop test sample, and the result is shown in
Figure 16. During the experiment, the IMC layer generated by the interface reaction of all the samples was relatively thin, being only 2~3 μm without the brittle IMC layer observed. The morphology and composition of the IMC were similar to the results of the aging experiment.
The failed samples were sectioned to observe the location of cracks at the solder joints, and the results are shown in
Figure 17. The cracks in the solder balls of the failed samples in either electroplated Ni/Au or ENEPIG 1 appeared at the junction of the IMC and Cu pads, and the cracks continued to extend to the entire interface, indicating that interface IMC strength is the key to affecting the drop performance of different surface finishes.
In the drop test, when the inside of the solder ball is subjected to impact from the pad, the Sn crystal grains will form dislocations to absorb part of the impact energy, and the generation of dislocations will increase the tensile strength and yield strength of the solder itself. At the same time, the strain-strengthening mechanism [
30] enables the solder to increase its own strength to be higher than the fracture strength of the IMC at a very high strain rate (1% s
−1 to 10% s
−1) in the drop test, resulting in failures that mostly occur in the brittle IMC layer, as shown in
Figure 18.
In this experiment, the sample was connected with PCB by one-time reflow soldering, where the growth time of IMC was short. As OSP had no Ni barrier layer, the solder reacted rapidly with the Cu pad to form IMC of a certain strength, which had good drop resistance. For the NiAu samples, Cu diffusion was inhibited by the Ni barrier layer; therefore, IMC was not fully formed in a short time, and the strength was low, leading to a poor drop performance. Due to the addition of a Pd layer in ENEPIG, there will be a small amount of PD in the IMC layer; this improves the strength of the IMC layer, resulting in a better drop performance than NiAu.
For ultrathin ENEPIG, the Ni layer was exhausted after reflow due to its thinner thickness. Only the secondary barrier effect of the Ni–Sn–P layer inhibited the diffusion of Cu. The generated IMC strength was better than that of NiAu and ENEPIG, and the IMC strength containing Pd was higher than the conventional (Cu, Ni)6Sn5 IMC. Therefore, it can be considered that the ultrathin ENEPIG with a Ni(P) layer thickness of 0.183 μm had a certain degree of drop resistance, which can meet the reliability that electronic devices require. The reasons of ENEPIG 1 failing the drop test may be as follows: the uneven surface of the thin coating, the formation of holes or the uneven growth of IMC during the interface reaction process, which may easily cause stress concentration at the interface and lead to failure.
3.5. Temperature Cycle Test Results
During the TCT experiment, the temperature cycle range was from −40 °C to 125 °C. According to the actual device application requirements, the design cycle was repeated 125 times. The experimental results are shown in
Table 4.
In this experiment, all the samples passed the requirements of 125 cycles, which shows that ultra-thin ENEPIG can meet the requirements of practical application. In order to further compare the difference in temperature cycling resistance of different surface finishes, the number of cycles of the samples was increased. It was found that all samples had an open circuit after 1602 cycles for ENEPIG 1 and after 1124 cycles for OSP. Due to the time cost required for the cycle, no more cycle experiments were carried out.
The IMC growth of the sample after TCT is observed in this section. As shown in
Figure 19, due to the impact of thermal load for a long time period, the IMC growth was sufficient, and its thickness was significantly higher than that of the drop test.
Combining the results of the temperature cycling experiment with IMC growth, it can be observed that there is no Ni barrier layer in OSP. When IMC grows rapidly, the volume mismatch caused by the CTE difference will produce a large amount of stress to promote the generation and extension of microcracks, meaning that the performance of TCT is the worst. In ENEPIG, due to the existence of a phosphorus-rich layer, CTE mismatch is stronger than NiAu; therefore, the thermal fatigue resistance is weaker than NiAu. Due to the blocking effect of a thin nickel layer and the Ni–Sn–P layer, ultrathin ENEPIG can inhibit the growth of IMC and reduce the stress generation to a certain extent, but the failure risk under high cycles is higher than that of ENEPIG.
To observe the failed samples of ENEPIG 1 and OSP, the results are shown in
Figure 20. The effect of thermal stress and strain on small-size solder balls was more serious, and the interface reaction was rapid, resulting in microcracks and voids. Therefore, the failure location was mostly concentrated on the side of the chip and substrate with a small bump size.
Due to the long-term temperature cycling process, a very obvious IMC layer was formed at the interface. Unlike the drop test, where cracks mostly occurred at the IMC interface, the failure in the high- and low-temperature cycle test occurred inside the solder near the IMC interface.
In the process of temperature cycling, the different CTE makes the volume change in each substance in the solder joint mismatched, which leads to stress and microcracks in the solder and promotes β- Sn grains recrystallize to form equiaxed grains [
30], producing a continuous grain boundary network that provides a channel for the diffusion of microcracks. The crack extends with the volume expansion of microstructure, and finally leads to failure. It can be clearly seen from
Figure 20 that the Sn grain boundary of the failure crack continued to extend, wherein the crack of OSP surface finish almost runs through the whole solder ball/IMC interface, while that of ENEPIG 1 diffuses upward along the grain boundary.