Interfacial reaction and mechanical properties for Cu/Sn/Ag system low temperature transient liquid phase bonding

After bonding at 300 °C for various time, cross-sections of the joints were evaluated with SEM and EPMA. Figure 2 showed the evolution process of the interfacial microstructure between Cu and Ag substrate. For 15 min (Fig. 2a), molten Sn reacted with the components that diffused from substrates but still remained as a layered structure after isothermal solidification, and thin layers of scalloped Cu₆Sn₅ and Ag₃Sn were found on the surface of Cu and Ag substrate, respectively. Furthermore, the thickness of Ag₃Sn layer was larger than that of Cu₆Sn₅ layer, which indicated that Ag/Sn reaction was faster than Cu/Sn reaction during solid–liquid interdiffusion. The EPMA line-scanning results for the distribution of Cu, Sn and Ag atoms were consistent with the transverse microstructure of the joint, and there was no intersection occurred at the position of Sn layer for the Cu and Ag diffusion curves, which revealed that no ternary alloy phase was formed at this time.

In the joint bonded for 60 min (Fig. 2b), Sn could not be clearly observed in this SEM objective area; however, some tearing-like zones were noticed on the fracture surface of the joint that demonstrated an ultra-thin residual Sn layer still existed (Fig. 2d). This phenomenon also implied heterogeneous growth of the Ag₃Sn or Cu₆Sn₅ grains in different areas, owing to anisotropic mass flow. In addition, it was found that a thin Cu₃Sn layer appeared between Cu₆Sn₅ layer and Cu substrate at this moment, which had columnar crystal morphology that was different with the scalloped Cu₆Sn₅ grain [8, 13]. EPMA elemental distribution curves were also in agreement with the interfacial microstructure, and Cu distribution curve did not intersect with that of Ag as well.

As the bonding time prolonged to 150 min (Fig. 2c), residual Sn was completely consumed and the Cu₆Sn₅ layer contacted with Ag₃Sn layer, and the thickness of the Cu₃Sn layer increased with increasing dwell time. Moreover, EPMA measurements for elemental diffusion were also similar to the interfacial microstructure, and the Cu and Ag scanning curves did not intersect with each other.

Several island-like Cu₆Sn₅ IMCs were isolated into separate areas after bonding for 300 min (Fig. 2e), since unstable Cu₆Sn₅ phase continually converted into Cu₃Sn phase, and the scalloped Ag₃Sn also began to contact with Cu₃Sn layer. Furthermore, it could be found that the Cu₃Sn/Ag₃Sn interface was more planar in comparison to the Cu₆Sn₅/Ag₃Sn interface, and the scalloped Ag₃Sn grains seemed to evolve into plane shape to combine with columnar Cu₃Sn grains. This revealed that diffusion processes probably took place on the two sides of the Cu–Sn IMCs/Ag–Sn IMCs (i.e. Cu₆Sn₅/Ag₃Sn) interface, respectively, and changed the interfacial morphology.

When the bonding time reached to 420 min (Fig. 2f), it was finished that the dispersive Cu₆Sn₅ compounds transformed into Cu₃Sn compounds on the Cu side, meanwhile, ζ phase layer probably was formed between Ag substrate and Ag₃Sn layer according to the Ag and Sn EPMA distribution curves, which presented a long-distanced transition characterization where closed to the Ag/Ag₃Sn interface and differed from the previous measurements. It was because that the content of Sn in ζ phase have an extensive range of 12.8–25.4 at.% according to binary Ag–Sn phase diagram [22].

Interestingly, it could be noticed that there was no intersection for Cu and Ag EPMA scanning curves at the Cu₆Sn₅/Ag₃Sn or Cu₃Sn/Ag₃Sn interface all the way, so we deduced that no ternary alloy phase was formed in Cu/Sn/Ag system LTTLP joints. On the other hand, area distributions of Cu, Sn and Ag atoms for the joint bonded after 420 min were evaluated by EPMA method, as shown in Fig. 3, Cu atoms did not be detected on the Ag₃Sn side (Fig. 3b) and no Ag atoms were examined on the Cu₃Sn side (Fig. 3d), and their mapping images sketched a similar boundaries with the SEM images (Fig. 3a) at the Ag₃Sn/Cu₃Sn interface. These detection data also demonstrated the above deduction to some extent. In addition, it could be distinctly seen that about 7 μm-thick light transitional band appeared in the area distribution graph of Sn atoms (Fig. 3c), which exactly corresponded to the layer of ζ phase.

LTTLP bonding for Cu/Sn/Ag system was also performed at 260 and 340 °C to further explore the interfacial reaction, especially for the combination process of Cu–Sn IMCs and Ag–Sn IMCs. Before experimental observation, metallographic samples with residual Sn were mildly etched by corrosion solution, as shown in Figs. 4 and 5, and the IMCs morphology was obtained from SEM. As similar with the joints bonded at 300 °C, scalloped Cu₆Sn₅ and Ag₃Sn grains were formed on the surface of the Cu and Ag substrate, respectively, and size of Ag₃Sn grain was notably larger than that of Cu₆Sn₅ grain, which also indicated that growth rate of Ag₃Sn was more rapid than the Cu₆Sn₅. With increasing temperature and dwell time, the thickness of both Cu₆Sn₅ and Ag₃Sn layer increased until the residual Sn was completely consumed. As Cu₆Sn₅ layer initially contacted with Ag₃Sn layer, the IMCs on one side became a barrier for the growth of the IMCs on the other side (Figs. 4b, 5b), since Cu and Ag atoms could not inter-diffuse into another IMCs side. In addition, a few small-sized pores were observed at the interface (Fig. 4b), and the residual Sn was isolated into island-like areas depicted in Figs. 4b and 5b. Pores still existed when Sn was entirely disappeared, as shown in Fig. 4c. However, these pores were never found with continuing reaction and the Cu–Sn IMCs combined well with the Ag–Sn IMCs (Fig. 4d). According to reaction behaviors of the Cu–Sn–Ag LTTLP bonding as analyzed above, it was supposed that disappearance of the pores with longer bonding time was controlled by the self-diffusion processes on the two sides of the Cu–Sn IMCs/Ag–Sn IMCs interface. Interestingly, it could be noticed that Cu₃Sn layer grew very slowly at 260 °C in comparison of Fig. 4c, d, and no significant thickening was observed with the prolonging dwell time; furthermore, the interface of Cu–Sn IMCs and Ag–Sn IMCs still maintained remarkable waved-shape though bonding process performed for a long time. This phenomenon was very different from the conditions higher than 300 °C (Figs. 2, 5), which resulted from the lower diffusion rate of Cu and Ag atoms under solid state at lower temperature.

According to binary Cu–Sn and Ag–Sn phase diagrams [22], Cu₆Sn₅ and Ag₃Sn phase have high remelting temperature of 415 and 480 °C, respectively, and that of the Cu₃Sn and ζ phase even exceed 600 °C. Therefore, Cu/Sn/Ag LTTLP joints have excellent thermal reliability. Furthermore, the Cu–Sn IMCs and Ag–Sn IMCs combine with each other very well after bonding for a long time (i.e. 300 °C for 420 min) according to the interfacial microstructure. In conclusion, it can be considered that Cu/Sn/Ag LTTLP bonding is a potential system that is used in high temperature packaging applications.

With change of the microstructure of the LTTLP joints, mechanical behaviors varied accordingly during shear tests. As shown in Fig. 6, shear strength of the joint bonded at 300 °C increased with bonding time from 23.4 MPa for 15 min to 60.2 MPa for 420 min. It could be seen that the shear strength dramatically increased from 15 to 30 min with the reduction of the layered Sn, but slowly increased after bonding time reached to 60 min though island-like Sn existed in the joint. After the IMCs layers on two different sides contacted with each other, shear strength increased slowly with the bonding time, due to phase transformation and disappearance of pores under solid-state reaction.

Fractures of several typical LTTLP joints were observed using SEM to discuss their rupture behaviors. Figure 7 showed the fracture surface of the joint produced at 300 °C for 150 min, in which the residual Sn was totally consumed and the Cu₆Sn₅ layer contacted with Ag₃Sn layer as mentioned above. Both intergranular fracture planes and transgranular fracture facets were observed on the Cu and Ag side, but most are intercrystalline mode. Marked zones A and B were Cu₆Sn₅ and Ag₃Sn phase identified by the EDS, respectively, and this revealed that cracks majorly propagated along the Cu₆Sn₅/Ag₃Sn interface during shear test. However, Cu₃Sn IMCs was not observed on the fracture, which indicated that adhesive strength of Cu₆Sn₅/Cu₃Sn interface was stronger than that of the Cu₆Sn₅/Ag₃Sn interface, because some pores were formed at the Cu₆Sn₅/Ag₃Sn interface (Fig. 7b) and negatively affected their combination.

Figure 8 showed the fracture surface of the joint bonded at 300 °C for 300 min, in which the island-like Cu₆Sn₅ IMCs were dispersedly distributed nearby the Cu₃Sn/Ag₃Sn interface and the microstructure was mainly composed of Cu₃Sn and Ag₃Sn intermetallic phases. It could be seen that some rock candy shaped structures and shear bands exhibited on the rupture surface, which indicated that cleavage fracture was the fracture mode for this joint. As analyzed in Table 1, Cu₆Sn₅, Cu₃Sn and Ag₃Sn were all detected both on the Cu and Ag side, thus, cracks majorly extended in the IMCs layers but not along the Cu₃Sn/Ag₃Sn interface during the rupture process. It was deduced that the adhesive strength of Cu₃Sn/Ag₃Sn interface slightly improved in contrast to Cu₆Sn₅/Ag₃Sn interface based on the shear strength changing with bonding time (Fig. 6). Interestingly, pronounced Ag₃Sn shear band, as shown in Fig. 8c, was found all over the fracture surface on the Ag side, indicating the force direction. In comparison to the brittle Cu–Sn IMCs, Ag₃Sn phase has a better ductility according to their rupture features.

After the Cu₆Sn₅ IMCs was fully converted into Cu₃Sn, as shown in Fig. 9, fracture surface of the joint was observed using SEM. Only Cu₃Sn, no Ag₃Sn or ζ phase, was detected throughout the fracture surface, according to the EDS results. Thus, failure path of this joint propagated in the Cu₃Sn layer. And broken Cu₃Sn presented a rock candy shaped morphology, which revealed cleavage fracture characterization.

Fracture model of different joints was developed based on the above analysis, as shown in Fig. 10. After full consumption of the residual Sn (Fig. 10a), scalloped Ag₃Sn and Cu₆Sn₅ combined with each other but pores remained at the interface, therefore, cracks would initiate at these areas and not only extended along Cu₆Sn₅/Ag₃Sn interface, but also propagated in the Cu₆Sn₅ and Ag₃Sn crystals for their mutually embedded scallops. As Ag₃Sn combined with Cu₆Sn₅ badly at this time, intergranular fracture played a key role during shear test. When Cu₆Sn₅ existed as island-like form at the Cu₃Sn/Ag₃Sn interface (Fig. 10b), failure path dominantly passed through the Cu₃Sn and Ag₃Sn layer, and cut through inlaid Cu₆Sn₅ islands as well. As Cu₆Sn₅ totally converted into Cu₃Sn (Fig. 10c), two interfaces of ζ phase/Ag₃Sn and Ag₃Sn/Cu₃Sn remained in the IMCs and combined very well, so that cracks propagated throughout Cu₃Sn layer during failure process.

As mentioned above, several pores were found that distributed around the Cu₆Sn₅/Ag₃Sn interface when Cu₆Sn₅ layer initially contacted with the Ag₃Sn layer. Their top-view morphologies were observed by the fracture surface of the corresponding joint, as shown in Fig. 11, and it could be found that pores were situated between either Cu₆Sn₅ scallops or Ag₃Sn scallops due to grain boundary grooving. Bosco proposed that pores in Cu–Sn–Cu TLP-bonded joints formed before the Sn melted, which was caused by Cu/Sn solid-state reaction in heating stage, and Sn coating thickness and heating rate were regarded as the crucial factors [11]. However, this model was not suitable for our bonding configuration, since formation of the IMCs during heating process impossibly occurred in Sn foil bonding with a low pressure. We considered that such pores were left behind by insufficient filling of the Sn islands entrapped in Cu₆Sn₅/Ag₃Sn interface which would isothermally convert into IMCs but accompanied by volume shrinkage. As this type of defect was distributed in the space of several IMCs grains, we called it grain boundary pore.

As the top of Cu₆Sn₅ scallops approximately contacted with the opposite Ag₃Sn scallops, growth model in geometry was established to discuss combination mechanism of Ag–Sn and Cu–Sn IMCs. As shown in Fig. 12, two different circumstances, including top-to-top and top-to-root, were distinguished by the coming contact forms of the Cu₆Sn₅ with Ag₃Sn scallops. In detail, the form of top-to-top was the top of Cu₆Sn₅ scallops contacted with the top of Ag₃Sn scallops, while top-to-root was the top/root of Cu₆Sn₅ scallops contacted with the root/top of Ag₃Sn scallops. For the case of top-to-top, molten Sn was isolated into separate areas once the top of Cu₆Sn₅ scallop contacted with the top of Ag₃Sn scallop (Fig. 12a), and they hindered vertical growth of each other so that grew toward the liquid Sn islands. As continued reaction, residual Sn islands was totally consumed and filled with IMCs, and many pores were remained at the grain boundaries (Fig. 12b), besides, no coalescence took place between two opposite IMCs scallops for incapable of elements interdiffusion through the boundary. Then, Cu₆Sn₅/Ag₃Sn interface formed in the top-to-top form turned into relatively planar, and these areas would rupture along the interface and generate intergranular fracture during the shear test. In fact, Fig. 11 also described the top-view morphology of Cu₆Sn₅/Ag₃Sn interface for this case somehow. For the case of top-to-root, as depicted in Fig. 12c, d, only if molten Sn was almost consumed, Cu₆Sn₅ and Ag₃Sn layers could contact with each other in theory, and small pores were remained at the top/root of either Cu₆Sn₅ or Ag₃Sn scallops, which was in a good agreement with the experimental observation (Fig. 4c). In addition, either Cu₆Sn₅ or Ag₃Sn layer seemly embedded into the opposite IMCs and their contacted interface kept original wave-shaped morphology. This case of areas would be ruptured through the IMCs layer and formed transgranular fracture.

After Cu₆Sn₅ and Ag₃Sn layer were fully contacted with each other, no matter top-to-top or top-to-root combination, the IMCs would continue to transform into more stable phases, namely, Cu₃Sn and ζ; and pores distributed around the interface were gradually healed. According to above experimental observation, Cu₃Sn layer grew more rapidly than ζ phase layer under the same condition, and the Cu₆Sn₅/Ag₃Sn interface evolved into Cu₃Sn/Ag₃Sn interface firstly, accompanied by vanish of pores. This revealed that self-diffusion probably took place in the reaction process and leaded to the phenomenon that positions of pores were filled by the IMCs, as a result, the adhesive strength of Cu₃Sn/Ag₃Sn interface was stronger than that of Cu₆Sn₅/Ag₃Sn interface.