- Email: [email protected]
Effect of Ag, Fe, Au and Ni on the growth kinetics of Sn–Cu intermetallic compound layers
Microelectronics Reliability 49 (2009) 242–247
Contents lists available at ScienceDirect
Microelectronics Reliability journal homepage: www.elsevier.com/locate/microrel
Effect of Ag, Fe, Au and Ni on the growth kinetics of Sn–Cu intermetallic compound layers T. Laurila a,*, J. Hurtig b, V. Vuorinen a, J.K. Kivilahti a a b
Department of Electronics, Faculty of Electronics, Communications and Automation, Helsinki University of Technology, P.O. Box 3000, FIN-02015 TKK, Finland SMT Electronics Reliability, Devices, R&D, Nokia Corporation, P.O. Box 86, FI-24101 SALO, Finland
a r t i c l e
i n f o
Article history: Received 2 July 2008 Received in revised form 19 August 2008 Available online 22 October 2008
a b s t r a c t The effect of Ag, Fe, Au and Ni on the interfacial reactions between Sn-based solder and Cu substrate has been investigated in this paper. Based on the solubility of the alloying elements in the Sn–Cu intermetallic compound (IMC) layers these elements can be divided into two categories: (i) alloying elements that do not dissolve significantly in either Cu6Sn5 or Cu3Sn and (ii) elements that exhibit significant solubility in Cu6Sn5 and also to Cu3Sn. It is shown that the latter group of elements have stronger effect on the growth behaviour of IMC’s in the Sn–Cu system than those belonging to the first group. Of the investigated elements Ni had the most prominent effect on the growth kinetics. It reduced greatly the thickness of Cu3Sn and consequently also the total IMC layer thickness. Au had similar but markedly weaker effect. On the contrary, Fe and Ag only slightly decreased the total IMC layer thickness, and more importantly did not change the thickness ratio of Cu6Sn5 to Cu3Sn in comparison to the pure Sn–Cu system. Ó 2008 Elsevier Ltd. All rights reserved.
1. Introduction Owing to the recent technological as well as legislational development, more fundamental reliability challenges are encountered at all interconnection levels in electronic applications. This is due to two reasons: (i) firstly, the need to integrate new functions (i.e. more powerful components) into smaller and thinner products, which leads to increasing power densities and, (ii) secondly, the need to employ new environment friendly and functional materials that increases greatly the complexity of the interconnection metallurgies at all product levels. This can have a marked effect on the reliability of especially portable electronics, which experience accidental shock loadings, strong local heating of power components as well as meet varying operational environments. Therefore, to ensure the best possible quality of electronic equipment, fundamental understanding of the reliability controlling interfacial chemical reactions inside the component package as well as at the interconnections at the board level is imperative [1–3]. Especially under the mechanical shock loading, where the strain-rate hardening of the solder material forces cracks to propagate in the intermetallic compound (IMC) layers, the role of interfacial intermetallic reactions as well as microstructural evolution in solder interconnections becomes more prominent [4,5]. Thus, there is a continuous interest to better understand and influence the properties of IMC layers in order to increase the reliability of solder interconnections. One way to influence the interfa* Corresponding author. Tel.: +358 9 451 5781; fax: +358 9 451 4982. E-mail address: tomi.laurila@tkk.fi (T. Laurila). 0026-2714/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.microrel.2008.08.007
cial reactions and the resulting product layers in a given system is to alloy either metallization (conductor) or solders with small amounts of additional elements. It is to be noted that also the presence of impurities in the interconnection system may have marked effects on the properties and growth of IMC layers. Additional elements can basically have three major effects on the reaction between the solder and the conductor metal: (i) firstly, they can increase or decrease the reaction/growth rate, (ii) secondly, additives can change the physical properties of the phases formed (in the case of Cu and Sn, e and g), and (iii) thirdly they can form additional reaction layers at the interface or they can displace the binary phases that would normally appear and form other reaction products instead. Further, these elements can be divided roughly into two major categories: (i) elements that only change the activities of species taking part in the interfacial reaction and do not participate themselves (not extensively soluble in IMC layer) and (ii) elements that take part in the interfacial reaction in question (generally show marked solubility in IMC layer). The elements belonging to the latter category usually have more pronounced effect on the IMC formation. There have been number of investigations about the effect of different alloying elements as well as impurities on the growth of IMC layers in Sn–Cu system [6–9]. Unfortunately, the results and their interpretation are not unambiguous. What has been stated above, points out the clear need for further systematic investigation of the effect of different alloying elements as well as impurities on the IMC formation and growth in Sn–Cu system. In this paper we have chosen Ag, Fe, Au and Ni as the elements that will be investigated. Ag is a known major alloying element in Sn-based lead free solders and it is also used
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247
243
as a surface finish material similarly than Ni and Au. On the other hand, Fe can be present in the solder system mainly as an impurity. Of the chosen elements, Ag and Fe represent species belonging to the first category, thus they are not markedly soluble in either Cu6Sn5 or Cu3Sn. On the contrary both Ni and Au exhibit extensive solubilities in one or both of the aforementioned IMC’s. Especially Ni has multiple effects on the properties of the Cu–Sn intermetallic compound layers as shown previously both experimentally and by simulations [10–13]. Thus, it is of great interest to compare the effects of elements belonging to these two categories on the Sn–Cu interfacial reaction. In this paper, effect of the aforementioned elements on the interfacial reactions is investigated both experimentally with the help of reaction couples and by utilizing the available thermodynamic-kinetic data on the systems. Fig. 1. Reaction layers formed in the reaction between pure 100Sn and Cu after annealing at 150 °C for 2560 h.
2. Materials and methods The Sn solder alloys were manufactured from commercially pure powders (99.9+, Ventron corporation and Merck & Co.) by melting the premixed powders in a Pyrex crucible under vacuum for several hours at 350 °C while frequently mixing the alloys. Different alloys are listed in Table 1. The amount of alloying element was determined by the following procedure. The equilibrium solubility of the alloying element in liquid Sn at 250 °C was checked from the corresponding binary phase diagram and then this amount was doubled in order to ensure that the solders were saturated with respect to the alloying element. The only exception was Au, as this procedure would have given the alloy content of 22 at.% for Au, which is clearly an unrealistic from the practical point of view, as Au is generally used as very thin protective coatings in electronics. The substrate used for the soldering was a square shaped Cu sheet (about 20 20 mm). These substrates were grinded with SiC emery paper and cleaned with HNO3 solution before soldering. Solders were melted in a crucible (on top of a heat plate) made of inert material and Cu sheets coated with RMA flux (Rosin Mildly Activated) were dipped into the molten solder for 40 s. Temperature during soldering was 260 °C. Samples were left to solidify into the crucible, which was removed from the heat plate and cooling thus took place in the air. After that the samples were removed from the crucible by using slow speed diamond saw and emery paper. About 2 mm of solder was left on top of the Cu sheets. After this the samples were annealed in a forced convection oven (Heraeus Instruments UT6) for different periods of time up to 2560 h at 150 °C. Cross-sections of the samples were prepared with the standard metallographic methods and they were analyzed with a field emission scanning electron microscope (JEOL 6335F FESEM) equipped with an energy dispersive spectrometer (EDS, Oxford ISIS). The reported IMC thickness (measured form the crosssections of the samples with SEM) are average values from at least 20 measurements.
tion of square root of time for the 100Sn–Cu couple is shown in Fig. 2. In Fig. 2 both the experimental data points as well as the lines obtained as a result of fitting of the data with linear regression are presented. As can be seen from these figures the growth of the total IMC thickness (Cu6Sn5 + Cu3Sn) is parabolic and that the thickness of Cu3Sn, which is initially much smaller than that of Cu6Sn5, exceeds its rival after about 300 h. This behaviour has been previously seen to occur in many experimental investigations and thus provides the base line for our present inquiries [1,2,10,11]. Fig. 3 shows the interface between Sn11Ag and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a function of square root of time for this couple is shown in Fig. 4. As can be seen the growth behaviour is very similar to that of 100Sn–Cu system as shown in previous publications [14,15] and also indicated by the Sn–Ag–Cu phase diagram and thermodynamic data [16,17]. The thickness of Cu3Sn surpasses that of Cu6Sn5 somewhat earlier in this system than in 100Sn–Cu. The large Ag3Sn IMC’s are also clearly seen in the micrograph. The reasons for the observed behaviour can be interpreted with the help of thermodynamic-kinetic information. As Ag does not dissolve in Cu6Sn5 or Cu3Sn the only way for it to influence the interfacial reactions is by influencing the activities of the reacting species, in this case that of Sn [17]. The presence of Ag slightly decreases the activity of Sn in the solder and thus reduces the driving force for the diffusion of Sn through Cu6Sn5 layer. As Sn is the main diffusing species in Cu6Sn5 at these temperatures [1,10,11,18–20] the resulting decrease in the Sn flux through the layer favours the growth of Cu3Sn, where, instead of Sn, Cu is the main diffusing species [1,10,11,20]. It is to be noted,
3. Results and discussion Fig. 1 shows the interface between pure 100Sn and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a funcTable 1 Different alloy compositions used in this study Metal
Purity (%)
Alloy composition (at.-%)
Tin, Sn Silver, Ag Gold, Au Iron, Fe Nickel, Ni
99.95 99.99 99.99 99.99 99.99
100Sn (reference) Sn11Ag Sn2Au Sn1Fe Sn1Ni
Fig. 2. The plot of IMC thickness as a function of square root of time (h) for the 100Sn–Cu couple.
244
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247
Fig. 3. Reaction layers formed in the reaction between Sn11Ag and Cu after annealing at 150 °C for 2560 h.
thickness is somewhat reduced in comparison to Sn–Cu couple. This is because as the flux of Sn through Cu6Sn5 is decreased, also the growth of Cu3Sn is slowed down due to the reduced availability of Sn at the Cu3Sn/Cu6Sn5 interface. Finally, it is to be noted that as the solubility of Ag in solid Sn is very small it has only minor effect on the activity of Sn [17]. Therefore, the difference between the growth kinetics of IMCs in Sn11Ag–Cu and 100Sn–Cu systems do not display marked differences. One can also ask whether the formation of Ag3Sn platelets at the interface may play some role in the intermetallic compound formation by physically blocking the diffusion of Sn–Cu6Sn5 layer at specific locations. However, this effect is anticipated to be of much less of significance than the effect of Ag on the activity of Sn. This is because there still remains plenty of paths for Sn diffusion to Cu–Sn IMC layers despite the Ag3Sn precipitates and it can even be argued that the phase interface between Ag3Sn precipitates and Sn-based solder matrix offer fast diffusion paths for Sn atoms along the interface, thus acting against the above mentioned ‘‘blocking effect”. Fig. 5 shows the interface between Sn1Fe and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a function of square root of time for this couple is shown in Fig. 6. Again the behaviour of this system closely resembles those of Sn–Cu and Sn11Ag–Cu. This is a result of the fact that Fe does not markedly dissolve in Cu–Sn intermetallic compound either and thus can only influence the activities of the elements in the system [17]. There exists one paper that reports some ternary solubility of Fe in both Cu6Sn5 and Cu3Sn [27], but in this work our chemical analyses did not support this result. The growth kinetics of this systems falls to somewhere between those of Sn–Cu and Sn11Ag–Cu, thus indicating that the influence of Fe on the activity of Sn is not as noticeable as in the case of Ag. Based on the thermodynamic data, the solubility of Fe in solid Sn is even smaller than in the case of Ag and
Fig. 4. The plot of IMC thickness as a function of square root of time (h) for the Sn11Ag–Cu couple.
however, that the IMC growth behaviour in the Cu–Sn system is not only strongly temperature dependent but in addition relatively complex as briefly discussed next. From room temperature up to 50–60 °C only the g0 -phase (low temperature form of Cu6Sn5) grows with an observable rate and the reaction is controlled by the release of Cu atoms from the Cu lattice. The main diffusing species at room temperature is Cu and it diffuses interstitially in Sn and along the grain boundaries of Cu6Sn5 [21–23]. Above 60 °C the e-phase starts to grow with measurable rate and its fraction out of the total intermetallic layer increases as annealing continues. Between 60 °C and 200 °C diffusion of Sn starts to control the growth of the g-phase [18–20,24]. This is rational, since as the temperature is increased, volume diffusion starts to dominate and grain boundary and interstitial diffusion of Cu does not play such a big role anymore. During this temperature interval, the e-phase continues to grow at the expense of the g-phase and both Cu and Sn are mobile during its growth. Consistent with the Cu3Au rule Cu is the more mobile species in the ephase. It should also be noted that if solid-state aging follows soldering, the resulting morphology after aging is dependent on the initial morphology formed during the solid–liquid contact. In addition, it must be noted that the growth of the phases in a given multiphase reactive diffusion couple are not independent, but are affected by other growing layers [25]. Hence, Cu3Sn consumes some of the Cu6Sn5 as it grows, and this effect is enhanced when the flux of Sn through Cu6Sn5 is reduced, as shown in previous publications [3,26]. This effect can be seen in Fig. 4 as not only the thickness of Cu3Sn surpass that of Cu6Sn5 earlier that in the case of pure Sn–Cu couple, but also the total IMC
Fig. 5. Reaction layers formed in the reaction between Sn1Fe and Cu after annealing at 150 °C for 2560 h.
Fig. 6. The plot of IMC thickness as a function of square root of time (h) for the Sn1Fe–Cu couple.
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247
therefore solid Sn taking part in the three phase equilibria between Sn, FeSn2 and Cu6Sn5 can be considered to be almost pure (thus has activity very close to that of pure Sn) [17,27–29]. Hence, there is only a slight difference between driving forces for diffusion in 100Sn–Cu and Sn1Fe–Cu couples (as well as Sn11Ag–Cu). Fig. 7 shows the interface between Sn2Au and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a function of square root of time for this couple is shown in Fig. 8. In this case it can be seen that the growth behaviour of the IMC layers is quite different as compared to the previous systems. The most striking difference is that the thickness of Cu3Sn does not surpass that of Cu6Sn5 at any stage during the annealing. The reasons for this behaviour can again be interpreted with the help of thermodynamic-kinetic information. Au is known to exhibit extensive solubility in Cu6Sn5 where it substitutes Cu atoms in the Cu sublattice and can reach solubilities up to 19.5 at.% [29]. Thus, now the influence of Au to growth kinetics of Cu–Sn IMC’s is direct. It is also noticeable that Au does not show such an extensive solubility in Cu3Sn [17,30–32]. Based on the information above, the reasons for the observed growth kinetics can be rationalized. The experimental results show that the thickness of the total IMC layer is reduced from about 12 lm to 7–8 lm (after 2560 h) when Au is present in the system, but that the thickness of (Cu,Au)6Sn5 is only slightly decreased. Thus, it seems that the dissolved Au in (Cu,Au)6Sn5 (in this investigation Au content in (Cu,Au)6Sn5 was about 13 at.%) influences mainly the growth of Cu3Sn. How this can be, as Au does not even dissolve in Cu3Sn? Based on the thermodynamic data it is known that Au stabilizes Cu6Sn5 [17]. This
Fig. 7. Reaction layers formed in the reaction between Sn2Au and Cu after annealing at 150 °C for 2560 h.
Fig. 8. The plot of IMC thickness as a function of square root of time (h) for the Sn2Au–Cu couple.
245
means that driving force for diffusion through (Cu,Au)6Sn5 is increased and that of through Cu3Sn is reduced, similarly as in the case of Ni, as will be briefly discussed later on in this paper and shown in detail in Ref. [26]. As the diffusion flux is directly proportional to the driving force, according to the well-known Nernst– Einstein relation, the material flux can be expected to grow in Cu6Sn5 and to decrease in Cu3Sn when Au is brought into the Sn–Cu system. In addition, as some of the Cu3Sn is formed at the expense of Cu6Sn5 (by breaking down the compound and forming Cu3Sn instead) the increased stability of (Cu,Au)6Sn5 makes this reaction more difficult, thus further reducing the amount of Cu3Sn in the system. At the moment it is not unambiguously clear what microstructural effects Au induces in the (Cu,Au)6Sn5 compound layer and how these possible changes influence the diffusion kinetics. It is possible that Au has similar effects on the structural transformation of g–g0 than Ni [12,13], but this requires further investigations. Fig. 9 shows the interface between Sn1Ni and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a function of square root of time for this couple is shown in Fig. 10 but there are also noticeable differences between the two systems. The behaviour displayed by the Sn2Au system is further enhanced in the case of Ni. From Fig. 10 it can be seen that: (i) total IMC layer thickness is smaller than in the Sn–Cu case, (ii) its growth does not follow parabolic kinetics and (iii) especially the thickness of Cu3Sn is drastically reduced. As the growth kinetics related to this system have been described in detail (both in liquid and in solidstate) in [3,26] together with lengthy list of appropriate references, only a brief discussion is presented here. The deviation from the parabolic growth law of the total IMC comes mainly from the (Cu,Ni)6Sn5 layer, as it is seen that Cu3Sn follows nicely diffusion
Fig. 9. Reaction layers formed in the reaction between Sn1Ni and Cu after annealing at 150 °C for 2560 h.
Fig. 10. The plot of IMC thickness as a function of square root of time (h) for the Sn1Ni-Cu couple.
246
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247
controlled growth kinetics (Fig. 10). The reason for the abnormal growth of (Cu,Ni)6Sn5 most probably has its origin in the liquid stage reaction. Because of the effect of Ni on the interaction between Cu and Sn atoms in the liquid (and thus also on the solubilities), the formation of a relatively thick and highly porous twophase structure [(Cu,Ni)6Sn5 + Sn] takes place, as explained thermodynamically in [3]. Thus, during the solid-state annealing, sintering of this porous structure takes place first and only after it transforms to the dense (Cu,Ni)6Sn5 structure, can the ‘‘normal” parabolic growth continue. Based on the experimental results it seems that the sintering of the porous (Cu,Ni)6Sn5 layer has barely reached its end within the time frame of the annealing. On the other hand, the decreased growth of Cu3Sn can be explained with the help of thermodynamic-kinetic considerations about the Cu– Sn–Ni system. During soldering process, the solder is in molten state for a few minutes. During such a short period of time the ternary compounds that have been observed or proposed to be present in the system [10,33,34] cannot form. Therefore, the metastable phase diagram without these compounds can more realistically describe the equilibria between solder alloys and conductor metals [1,3,26]. Using the optimized dataset [17], the Gibbs free energy curves corresponding to this metastable phase diagram were calculated. When 1 at.% of Ni is added to Sn, the Nicontent of (Cu,Ni)6Sn5 is about 20 at.% (measured by EDS) and its stability is increased to gg*(Fig. 11). The Cu3Sn being in equilibrium with it is practically free of Ni, and so its stability will remain practically the same as that of pure Cu3Sn (ge*) (Fig. 11). As a result, the driving forces for the diffusion over the intermetallic compounds will change as shown in Fig. 11. More detailed information about the use of Gibbs energy diagrams can be found from Refs. [35,36]. The driving force for the diffusion of Sn through the (Cu,Ni)6Sn5 layer DGSn is now about 10 times the driving force (DGSn) for the 100Sn–Cu couple. On the other hand, the driving
Fig. 12. Thickness of the IMC layers after annealing at 150 °C for 2560 h.
force for the diffusion of Cu through Cu3Sn (DGCu ) reduces almost to zero (it is therefore not shown in Fig. 11). This means that through the Nernst–Einstein relation the fluxes through the Cu3Sn and the (Cu,Ni)6Sn5 layers are markedly altered, thus resulting in to the observed growth behaviour (i.e. the marked decrease in the layer thickness of Cu3Sn layer), as also previously shown [3,26]. To summarize the experimental findings, Fig. 12 shows the thickness of the interfacial IMC layers from all the studied systems after annealing for 2560 h at 150 °C. It is evident that the effect of Ag and Fe is relatively small. On the other hand, Au and especially Ni have marked effect on the growth kinetics (i.e. relative as well as absolute thickness) of both Sn–Cu intermetallic compounds. 4. Conclusions The effect of Ag, Fe, Au and Ni dissolved in liquid Sn on the interfacial reactions between Sn-based solder and Cu substrate has been investigated. Based on the solubility of the alloying elements in the IMC layers these elements can be divided into two categories: (i) alloying elements that do not dissolve in either Cu6Sn5 or Cu3Sn and (ii) elements that exhibit significant solubility in (usually) Cu6Sn5 and (possible) to Cu3Sn. It was shown that the latter group of elements have markedly stronger effect on the growth behaviour of IMC’s in the Sn–Cu system than those belonging to the first group, because these (that do not dissolve in IMC’s) can influence diffusion fluxes in the layers only indirectly through the activity of Sn. On the other hand, if the element dissolves in IMC layer it can alter its stability and probably also its microstructure, thus influencing the growth kinetics directly. Of the investigated elements Ni had the most prominent effect on the IMC growth behaviour. It reduced greatly the thickness of Cu3Sn and consequently also the total IMC layer thickness. Au had similar but weaker effect. It is important to notice that both elements exhibit significant solubilities to Cu6Sn5 layer. On the contrary, Fe and Ag that do not dissolve in IMC’s, also do not alter the ratio of Cu6Sn5–Cu3Sn layer in the system and only slightly decreased the total IMC layer thickness. Thus, it can be concluded that if one wants to influence the growth kinetics (and thus reliability of the solder interconnections) of Cu6Sn5 and Cu3Sn by alloying, one should use elements that exhibit marked solubility in one or both of the IMC’s. References
Fig. 11. Gibbs free energy diagram at 150 °C showing the driving forces for diffusion over the interfacial reaction zone.
[1] Laurila T, Vuorinen V, Kivilahti JK. Mater Sci Eng R 2005;R49(1–2):1–60. [2] Zeng K, Tu KN. Mater Sci Eng R 2002;R38(2):55–105.
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247 [3] Vuorinen V, Yu H, Laurila T, Kivilahti JK. J Electron Mater 2008;37(6): 792–805. [4] Marjamäki P. Doctoral dissertation, Helsinki University of Technology; 2007,
[19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36]
247
Onishi M, Fujibuchi H. Trans Jpn Inst Metals 1975;16:539. Oh M. Doctoral dissertation, Lehigh University; 1994. Tu KN. Acta Metall. 1973;21:347. Tu KN, Thompson R. Acta Metall. 1982;30:947. Tu KN. Mater Chem Phys 1996;46:217. Revay L. Surf Tech 1977;5:57. Laurila T, Molarius J. Critic Rev Solid State Mater Sci 2003;28:185–230. Vuorinen V, Laurila T, Mattila T, Heikinheimo E, Kivilahti JK. J Electron Mater 2007;36(10):1355–62. Zhou H, Zheng J. Acta Metallur Sin 1987;23(1):39–41. Hari Kumar KC, Wollants P, Delaey L. Calphad 1996;20(2):139–49. Marcotte V, Schroeder K. MRS Symposium Proceedings 19; 1983. Roeder J. Doctoral dissertation, Lehigh University; 1988. Zakel E. PhD thesis, Technical University Berlin; 1994. Karlsen O, Kjekshus A, Rost E. Acta Chem Scandinavica (Denmark) 1992;46(2):147. Lin C-H, Chen S-W, Wang C-H. J Electron Mater 2002;31(9):907. Schmetterer C, Flandorfer H, Luef Ch, Kodentsov A, Ipser H. J Electron Mater, in press. Hillert M. In: Aaronson H, editor. Lectures on the theory of phase transformations. AIME; 1975. Hillert M. Phase equilibria, phase diagrams and phase transformations. Cambridge Press; 1998.
Contents lists available at ScienceDirect
Microelectronics Reliability journal homepage: www.elsevier.com/locate/microrel
Effect of Ag, Fe, Au and Ni on the growth kinetics of Sn–Cu intermetallic compound layers T. Laurila a,*, J. Hurtig b, V. Vuorinen a, J.K. Kivilahti a a b
Department of Electronics, Faculty of Electronics, Communications and Automation, Helsinki University of Technology, P.O. Box 3000, FIN-02015 TKK, Finland SMT Electronics Reliability, Devices, R&D, Nokia Corporation, P.O. Box 86, FI-24101 SALO, Finland
a r t i c l e
i n f o
Article history: Received 2 July 2008 Received in revised form 19 August 2008 Available online 22 October 2008
a b s t r a c t The effect of Ag, Fe, Au and Ni on the interfacial reactions between Sn-based solder and Cu substrate has been investigated in this paper. Based on the solubility of the alloying elements in the Sn–Cu intermetallic compound (IMC) layers these elements can be divided into two categories: (i) alloying elements that do not dissolve significantly in either Cu6Sn5 or Cu3Sn and (ii) elements that exhibit significant solubility in Cu6Sn5 and also to Cu3Sn. It is shown that the latter group of elements have stronger effect on the growth behaviour of IMC’s in the Sn–Cu system than those belonging to the first group. Of the investigated elements Ni had the most prominent effect on the growth kinetics. It reduced greatly the thickness of Cu3Sn and consequently also the total IMC layer thickness. Au had similar but markedly weaker effect. On the contrary, Fe and Ag only slightly decreased the total IMC layer thickness, and more importantly did not change the thickness ratio of Cu6Sn5 to Cu3Sn in comparison to the pure Sn–Cu system. Ó 2008 Elsevier Ltd. All rights reserved.
1. Introduction Owing to the recent technological as well as legislational development, more fundamental reliability challenges are encountered at all interconnection levels in electronic applications. This is due to two reasons: (i) firstly, the need to integrate new functions (i.e. more powerful components) into smaller and thinner products, which leads to increasing power densities and, (ii) secondly, the need to employ new environment friendly and functional materials that increases greatly the complexity of the interconnection metallurgies at all product levels. This can have a marked effect on the reliability of especially portable electronics, which experience accidental shock loadings, strong local heating of power components as well as meet varying operational environments. Therefore, to ensure the best possible quality of electronic equipment, fundamental understanding of the reliability controlling interfacial chemical reactions inside the component package as well as at the interconnections at the board level is imperative [1–3]. Especially under the mechanical shock loading, where the strain-rate hardening of the solder material forces cracks to propagate in the intermetallic compound (IMC) layers, the role of interfacial intermetallic reactions as well as microstructural evolution in solder interconnections becomes more prominent [4,5]. Thus, there is a continuous interest to better understand and influence the properties of IMC layers in order to increase the reliability of solder interconnections. One way to influence the interfa* Corresponding author. Tel.: +358 9 451 5781; fax: +358 9 451 4982. E-mail address: tomi.laurila@tkk.fi (T. Laurila). 0026-2714/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.microrel.2008.08.007
cial reactions and the resulting product layers in a given system is to alloy either metallization (conductor) or solders with small amounts of additional elements. It is to be noted that also the presence of impurities in the interconnection system may have marked effects on the properties and growth of IMC layers. Additional elements can basically have three major effects on the reaction between the solder and the conductor metal: (i) firstly, they can increase or decrease the reaction/growth rate, (ii) secondly, additives can change the physical properties of the phases formed (in the case of Cu and Sn, e and g), and (iii) thirdly they can form additional reaction layers at the interface or they can displace the binary phases that would normally appear and form other reaction products instead. Further, these elements can be divided roughly into two major categories: (i) elements that only change the activities of species taking part in the interfacial reaction and do not participate themselves (not extensively soluble in IMC layer) and (ii) elements that take part in the interfacial reaction in question (generally show marked solubility in IMC layer). The elements belonging to the latter category usually have more pronounced effect on the IMC formation. There have been number of investigations about the effect of different alloying elements as well as impurities on the growth of IMC layers in Sn–Cu system [6–9]. Unfortunately, the results and their interpretation are not unambiguous. What has been stated above, points out the clear need for further systematic investigation of the effect of different alloying elements as well as impurities on the IMC formation and growth in Sn–Cu system. In this paper we have chosen Ag, Fe, Au and Ni as the elements that will be investigated. Ag is a known major alloying element in Sn-based lead free solders and it is also used
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247
243
as a surface finish material similarly than Ni and Au. On the other hand, Fe can be present in the solder system mainly as an impurity. Of the chosen elements, Ag and Fe represent species belonging to the first category, thus they are not markedly soluble in either Cu6Sn5 or Cu3Sn. On the contrary both Ni and Au exhibit extensive solubilities in one or both of the aforementioned IMC’s. Especially Ni has multiple effects on the properties of the Cu–Sn intermetallic compound layers as shown previously both experimentally and by simulations [10–13]. Thus, it is of great interest to compare the effects of elements belonging to these two categories on the Sn–Cu interfacial reaction. In this paper, effect of the aforementioned elements on the interfacial reactions is investigated both experimentally with the help of reaction couples and by utilizing the available thermodynamic-kinetic data on the systems. Fig. 1. Reaction layers formed in the reaction between pure 100Sn and Cu after annealing at 150 °C for 2560 h.
2. Materials and methods The Sn solder alloys were manufactured from commercially pure powders (99.9+, Ventron corporation and Merck & Co.) by melting the premixed powders in a Pyrex crucible under vacuum for several hours at 350 °C while frequently mixing the alloys. Different alloys are listed in Table 1. The amount of alloying element was determined by the following procedure. The equilibrium solubility of the alloying element in liquid Sn at 250 °C was checked from the corresponding binary phase diagram and then this amount was doubled in order to ensure that the solders were saturated with respect to the alloying element. The only exception was Au, as this procedure would have given the alloy content of 22 at.% for Au, which is clearly an unrealistic from the practical point of view, as Au is generally used as very thin protective coatings in electronics. The substrate used for the soldering was a square shaped Cu sheet (about 20 20 mm). These substrates were grinded with SiC emery paper and cleaned with HNO3 solution before soldering. Solders were melted in a crucible (on top of a heat plate) made of inert material and Cu sheets coated with RMA flux (Rosin Mildly Activated) were dipped into the molten solder for 40 s. Temperature during soldering was 260 °C. Samples were left to solidify into the crucible, which was removed from the heat plate and cooling thus took place in the air. After that the samples were removed from the crucible by using slow speed diamond saw and emery paper. About 2 mm of solder was left on top of the Cu sheets. After this the samples were annealed in a forced convection oven (Heraeus Instruments UT6) for different periods of time up to 2560 h at 150 °C. Cross-sections of the samples were prepared with the standard metallographic methods and they were analyzed with a field emission scanning electron microscope (JEOL 6335F FESEM) equipped with an energy dispersive spectrometer (EDS, Oxford ISIS). The reported IMC thickness (measured form the crosssections of the samples with SEM) are average values from at least 20 measurements.
tion of square root of time for the 100Sn–Cu couple is shown in Fig. 2. In Fig. 2 both the experimental data points as well as the lines obtained as a result of fitting of the data with linear regression are presented. As can be seen from these figures the growth of the total IMC thickness (Cu6Sn5 + Cu3Sn) is parabolic and that the thickness of Cu3Sn, which is initially much smaller than that of Cu6Sn5, exceeds its rival after about 300 h. This behaviour has been previously seen to occur in many experimental investigations and thus provides the base line for our present inquiries [1,2,10,11]. Fig. 3 shows the interface between Sn11Ag and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a function of square root of time for this couple is shown in Fig. 4. As can be seen the growth behaviour is very similar to that of 100Sn–Cu system as shown in previous publications [14,15] and also indicated by the Sn–Ag–Cu phase diagram and thermodynamic data [16,17]. The thickness of Cu3Sn surpasses that of Cu6Sn5 somewhat earlier in this system than in 100Sn–Cu. The large Ag3Sn IMC’s are also clearly seen in the micrograph. The reasons for the observed behaviour can be interpreted with the help of thermodynamic-kinetic information. As Ag does not dissolve in Cu6Sn5 or Cu3Sn the only way for it to influence the interfacial reactions is by influencing the activities of the reacting species, in this case that of Sn [17]. The presence of Ag slightly decreases the activity of Sn in the solder and thus reduces the driving force for the diffusion of Sn through Cu6Sn5 layer. As Sn is the main diffusing species in Cu6Sn5 at these temperatures [1,10,11,18–20] the resulting decrease in the Sn flux through the layer favours the growth of Cu3Sn, where, instead of Sn, Cu is the main diffusing species [1,10,11,20]. It is to be noted,
3. Results and discussion Fig. 1 shows the interface between pure 100Sn and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a funcTable 1 Different alloy compositions used in this study Metal
Purity (%)
Alloy composition (at.-%)
Tin, Sn Silver, Ag Gold, Au Iron, Fe Nickel, Ni
99.95 99.99 99.99 99.99 99.99
100Sn (reference) Sn11Ag Sn2Au Sn1Fe Sn1Ni
Fig. 2. The plot of IMC thickness as a function of square root of time (h) for the 100Sn–Cu couple.
244
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247
Fig. 3. Reaction layers formed in the reaction between Sn11Ag and Cu after annealing at 150 °C for 2560 h.
thickness is somewhat reduced in comparison to Sn–Cu couple. This is because as the flux of Sn through Cu6Sn5 is decreased, also the growth of Cu3Sn is slowed down due to the reduced availability of Sn at the Cu3Sn/Cu6Sn5 interface. Finally, it is to be noted that as the solubility of Ag in solid Sn is very small it has only minor effect on the activity of Sn [17]. Therefore, the difference between the growth kinetics of IMCs in Sn11Ag–Cu and 100Sn–Cu systems do not display marked differences. One can also ask whether the formation of Ag3Sn platelets at the interface may play some role in the intermetallic compound formation by physically blocking the diffusion of Sn–Cu6Sn5 layer at specific locations. However, this effect is anticipated to be of much less of significance than the effect of Ag on the activity of Sn. This is because there still remains plenty of paths for Sn diffusion to Cu–Sn IMC layers despite the Ag3Sn precipitates and it can even be argued that the phase interface between Ag3Sn precipitates and Sn-based solder matrix offer fast diffusion paths for Sn atoms along the interface, thus acting against the above mentioned ‘‘blocking effect”. Fig. 5 shows the interface between Sn1Fe and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a function of square root of time for this couple is shown in Fig. 6. Again the behaviour of this system closely resembles those of Sn–Cu and Sn11Ag–Cu. This is a result of the fact that Fe does not markedly dissolve in Cu–Sn intermetallic compound either and thus can only influence the activities of the elements in the system [17]. There exists one paper that reports some ternary solubility of Fe in both Cu6Sn5 and Cu3Sn [27], but in this work our chemical analyses did not support this result. The growth kinetics of this systems falls to somewhere between those of Sn–Cu and Sn11Ag–Cu, thus indicating that the influence of Fe on the activity of Sn is not as noticeable as in the case of Ag. Based on the thermodynamic data, the solubility of Fe in solid Sn is even smaller than in the case of Ag and
Fig. 4. The plot of IMC thickness as a function of square root of time (h) for the Sn11Ag–Cu couple.
however, that the IMC growth behaviour in the Cu–Sn system is not only strongly temperature dependent but in addition relatively complex as briefly discussed next. From room temperature up to 50–60 °C only the g0 -phase (low temperature form of Cu6Sn5) grows with an observable rate and the reaction is controlled by the release of Cu atoms from the Cu lattice. The main diffusing species at room temperature is Cu and it diffuses interstitially in Sn and along the grain boundaries of Cu6Sn5 [21–23]. Above 60 °C the e-phase starts to grow with measurable rate and its fraction out of the total intermetallic layer increases as annealing continues. Between 60 °C and 200 °C diffusion of Sn starts to control the growth of the g-phase [18–20,24]. This is rational, since as the temperature is increased, volume diffusion starts to dominate and grain boundary and interstitial diffusion of Cu does not play such a big role anymore. During this temperature interval, the e-phase continues to grow at the expense of the g-phase and both Cu and Sn are mobile during its growth. Consistent with the Cu3Au rule Cu is the more mobile species in the ephase. It should also be noted that if solid-state aging follows soldering, the resulting morphology after aging is dependent on the initial morphology formed during the solid–liquid contact. In addition, it must be noted that the growth of the phases in a given multiphase reactive diffusion couple are not independent, but are affected by other growing layers [25]. Hence, Cu3Sn consumes some of the Cu6Sn5 as it grows, and this effect is enhanced when the flux of Sn through Cu6Sn5 is reduced, as shown in previous publications [3,26]. This effect can be seen in Fig. 4 as not only the thickness of Cu3Sn surpass that of Cu6Sn5 earlier that in the case of pure Sn–Cu couple, but also the total IMC
Fig. 5. Reaction layers formed in the reaction between Sn1Fe and Cu after annealing at 150 °C for 2560 h.
Fig. 6. The plot of IMC thickness as a function of square root of time (h) for the Sn1Fe–Cu couple.
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247
therefore solid Sn taking part in the three phase equilibria between Sn, FeSn2 and Cu6Sn5 can be considered to be almost pure (thus has activity very close to that of pure Sn) [17,27–29]. Hence, there is only a slight difference between driving forces for diffusion in 100Sn–Cu and Sn1Fe–Cu couples (as well as Sn11Ag–Cu). Fig. 7 shows the interface between Sn2Au and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a function of square root of time for this couple is shown in Fig. 8. In this case it can be seen that the growth behaviour of the IMC layers is quite different as compared to the previous systems. The most striking difference is that the thickness of Cu3Sn does not surpass that of Cu6Sn5 at any stage during the annealing. The reasons for this behaviour can again be interpreted with the help of thermodynamic-kinetic information. Au is known to exhibit extensive solubility in Cu6Sn5 where it substitutes Cu atoms in the Cu sublattice and can reach solubilities up to 19.5 at.% [29]. Thus, now the influence of Au to growth kinetics of Cu–Sn IMC’s is direct. It is also noticeable that Au does not show such an extensive solubility in Cu3Sn [17,30–32]. Based on the information above, the reasons for the observed growth kinetics can be rationalized. The experimental results show that the thickness of the total IMC layer is reduced from about 12 lm to 7–8 lm (after 2560 h) when Au is present in the system, but that the thickness of (Cu,Au)6Sn5 is only slightly decreased. Thus, it seems that the dissolved Au in (Cu,Au)6Sn5 (in this investigation Au content in (Cu,Au)6Sn5 was about 13 at.%) influences mainly the growth of Cu3Sn. How this can be, as Au does not even dissolve in Cu3Sn? Based on the thermodynamic data it is known that Au stabilizes Cu6Sn5 [17]. This
Fig. 7. Reaction layers formed in the reaction between Sn2Au and Cu after annealing at 150 °C for 2560 h.
Fig. 8. The plot of IMC thickness as a function of square root of time (h) for the Sn2Au–Cu couple.
245
means that driving force for diffusion through (Cu,Au)6Sn5 is increased and that of through Cu3Sn is reduced, similarly as in the case of Ni, as will be briefly discussed later on in this paper and shown in detail in Ref. [26]. As the diffusion flux is directly proportional to the driving force, according to the well-known Nernst– Einstein relation, the material flux can be expected to grow in Cu6Sn5 and to decrease in Cu3Sn when Au is brought into the Sn–Cu system. In addition, as some of the Cu3Sn is formed at the expense of Cu6Sn5 (by breaking down the compound and forming Cu3Sn instead) the increased stability of (Cu,Au)6Sn5 makes this reaction more difficult, thus further reducing the amount of Cu3Sn in the system. At the moment it is not unambiguously clear what microstructural effects Au induces in the (Cu,Au)6Sn5 compound layer and how these possible changes influence the diffusion kinetics. It is possible that Au has similar effects on the structural transformation of g–g0 than Ni [12,13], but this requires further investigations. Fig. 9 shows the interface between Sn1Ni and Cu after annealing at 150 °C for 2560 h. The plot of IMC thickness as a function of square root of time for this couple is shown in Fig. 10 but there are also noticeable differences between the two systems. The behaviour displayed by the Sn2Au system is further enhanced in the case of Ni. From Fig. 10 it can be seen that: (i) total IMC layer thickness is smaller than in the Sn–Cu case, (ii) its growth does not follow parabolic kinetics and (iii) especially the thickness of Cu3Sn is drastically reduced. As the growth kinetics related to this system have been described in detail (both in liquid and in solidstate) in [3,26] together with lengthy list of appropriate references, only a brief discussion is presented here. The deviation from the parabolic growth law of the total IMC comes mainly from the (Cu,Ni)6Sn5 layer, as it is seen that Cu3Sn follows nicely diffusion
Fig. 9. Reaction layers formed in the reaction between Sn1Ni and Cu after annealing at 150 °C for 2560 h.
Fig. 10. The plot of IMC thickness as a function of square root of time (h) for the Sn1Ni-Cu couple.
246
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247
controlled growth kinetics (Fig. 10). The reason for the abnormal growth of (Cu,Ni)6Sn5 most probably has its origin in the liquid stage reaction. Because of the effect of Ni on the interaction between Cu and Sn atoms in the liquid (and thus also on the solubilities), the formation of a relatively thick and highly porous twophase structure [(Cu,Ni)6Sn5 + Sn] takes place, as explained thermodynamically in [3]. Thus, during the solid-state annealing, sintering of this porous structure takes place first and only after it transforms to the dense (Cu,Ni)6Sn5 structure, can the ‘‘normal” parabolic growth continue. Based on the experimental results it seems that the sintering of the porous (Cu,Ni)6Sn5 layer has barely reached its end within the time frame of the annealing. On the other hand, the decreased growth of Cu3Sn can be explained with the help of thermodynamic-kinetic considerations about the Cu– Sn–Ni system. During soldering process, the solder is in molten state for a few minutes. During such a short period of time the ternary compounds that have been observed or proposed to be present in the system [10,33,34] cannot form. Therefore, the metastable phase diagram without these compounds can more realistically describe the equilibria between solder alloys and conductor metals [1,3,26]. Using the optimized dataset [17], the Gibbs free energy curves corresponding to this metastable phase diagram were calculated. When 1 at.% of Ni is added to Sn, the Nicontent of (Cu,Ni)6Sn5 is about 20 at.% (measured by EDS) and its stability is increased to gg*(Fig. 11). The Cu3Sn being in equilibrium with it is practically free of Ni, and so its stability will remain practically the same as that of pure Cu3Sn (ge*) (Fig. 11). As a result, the driving forces for the diffusion over the intermetallic compounds will change as shown in Fig. 11. More detailed information about the use of Gibbs energy diagrams can be found from Refs. [35,36]. The driving force for the diffusion of Sn through the (Cu,Ni)6Sn5 layer DGSn is now about 10 times the driving force (DGSn) for the 100Sn–Cu couple. On the other hand, the driving
Fig. 12. Thickness of the IMC layers after annealing at 150 °C for 2560 h.
force for the diffusion of Cu through Cu3Sn (DGCu ) reduces almost to zero (it is therefore not shown in Fig. 11). This means that through the Nernst–Einstein relation the fluxes through the Cu3Sn and the (Cu,Ni)6Sn5 layers are markedly altered, thus resulting in to the observed growth behaviour (i.e. the marked decrease in the layer thickness of Cu3Sn layer), as also previously shown [3,26]. To summarize the experimental findings, Fig. 12 shows the thickness of the interfacial IMC layers from all the studied systems after annealing for 2560 h at 150 °C. It is evident that the effect of Ag and Fe is relatively small. On the other hand, Au and especially Ni have marked effect on the growth kinetics (i.e. relative as well as absolute thickness) of both Sn–Cu intermetallic compounds. 4. Conclusions The effect of Ag, Fe, Au and Ni dissolved in liquid Sn on the interfacial reactions between Sn-based solder and Cu substrate has been investigated. Based on the solubility of the alloying elements in the IMC layers these elements can be divided into two categories: (i) alloying elements that do not dissolve in either Cu6Sn5 or Cu3Sn and (ii) elements that exhibit significant solubility in (usually) Cu6Sn5 and (possible) to Cu3Sn. It was shown that the latter group of elements have markedly stronger effect on the growth behaviour of IMC’s in the Sn–Cu system than those belonging to the first group, because these (that do not dissolve in IMC’s) can influence diffusion fluxes in the layers only indirectly through the activity of Sn. On the other hand, if the element dissolves in IMC layer it can alter its stability and probably also its microstructure, thus influencing the growth kinetics directly. Of the investigated elements Ni had the most prominent effect on the IMC growth behaviour. It reduced greatly the thickness of Cu3Sn and consequently also the total IMC layer thickness. Au had similar but weaker effect. It is important to notice that both elements exhibit significant solubilities to Cu6Sn5 layer. On the contrary, Fe and Ag that do not dissolve in IMC’s, also do not alter the ratio of Cu6Sn5–Cu3Sn layer in the system and only slightly decreased the total IMC layer thickness. Thus, it can be concluded that if one wants to influence the growth kinetics (and thus reliability of the solder interconnections) of Cu6Sn5 and Cu3Sn by alloying, one should use elements that exhibit marked solubility in one or both of the IMC’s. References
Fig. 11. Gibbs free energy diagram at 150 °C showing the driving forces for diffusion over the interfacial reaction zone.
[1] Laurila T, Vuorinen V, Kivilahti JK. Mater Sci Eng R 2005;R49(1–2):1–60. [2] Zeng K, Tu KN. Mater Sci Eng R 2002;R38(2):55–105.
T. Laurila et al. / Microelectronics Reliability 49 (2009) 242–247 [3] Vuorinen V, Yu H, Laurila T, Kivilahti JK. J Electron Mater 2008;37(6): 792–805. [4] Marjamäki P. Doctoral dissertation, Helsinki University of Technology; 2007,
[19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36]
247
Onishi M, Fujibuchi H. Trans Jpn Inst Metals 1975;16:539. Oh M. Doctoral dissertation, Lehigh University; 1994. Tu KN. Acta Metall. 1973;21:347. Tu KN, Thompson R. Acta Metall. 1982;30:947. Tu KN. Mater Chem Phys 1996;46:217. Revay L. Surf Tech 1977;5:57. Laurila T, Molarius J. Critic Rev Solid State Mater Sci 2003;28:185–230. Vuorinen V, Laurila T, Mattila T, Heikinheimo E, Kivilahti JK. J Electron Mater 2007;36(10):1355–62. Zhou H, Zheng J. Acta Metallur Sin 1987;23(1):39–41. Hari Kumar KC, Wollants P, Delaey L. Calphad 1996;20(2):139–49. Marcotte V, Schroeder K. MRS Symposium Proceedings 19; 1983. Roeder J. Doctoral dissertation, Lehigh University; 1988. Zakel E. PhD thesis, Technical University Berlin; 1994. Karlsen O, Kjekshus A, Rost E. Acta Chem Scandinavica (Denmark) 1992;46(2):147. Lin C-H, Chen S-W, Wang C-H. J Electron Mater 2002;31(9):907. Schmetterer C, Flandorfer H, Luef Ch, Kodentsov A, Ipser H. J Electron Mater, in press. Hillert M. In: Aaronson H, editor. Lectures on the theory of phase transformations. AIME; 1975. Hillert M. Phase equilibria, phase diagrams and phase transformations. Cambridge Press; 1998.