This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/copyright

Author's personal copy

Kirkendall voids formation in the reaction between Ni-doped SnAglead-free solders and different Cu substrates

Y.W. Wang, Y.W. Lin, C.R. Kao *

Department of Materials Science and Engineering, National Taiwan University, No. 1, Sec. 4, Roosevelt Road, Taipei City, Taipei 10617, Taiwan

a r t i c l e i n f o

Article history:Received 25 July 2008Received in revised form 24 September2008Available online 4 November 2008

a b s t r a c t

The reactions between Sn2.5Ag solder doped with different levels of Ni (0–0.1 wt.%) and two differenttypes of Cu substrates, electroplated Cu and high-purity Cu substrates, were studied. The main objectivewas to investigate the effect of Cu substrate and the effect of Ni additions on the formation of Kirkendallvoids within the Cu3Sn phase. Reaction conditions included one reflow and subsequent aging at 160 �Cfor up to 2000 h. After reflow, Cu6Sn5 was the only reaction product observed for all the different soldersand substrates used. During aging, both Cu6Sn5 and Cu3Sn formed. Nevertheless, Kirkendall voids wereobserved only when the electroplated Cu was used, and was not observed when high-purity Cu was used.It was proposed that impurities in electroplated Cu helped the nucleation of these voids. The Ni additionsmade the Cu3Sn layer thinner. For the case of the electroplated Cu substrates, the amount of Kirkendallvoids decreased correspondingly with the Ni additions.

� 2008 Elsevier Ltd. All rights reserved.

1. Introduction

At temperatures greater than 50–60 �C, the solid-state reactionsbetween Cu and many Sn-based solders produce two reactionproducts, Cu6Sn5 and Cu3Sn [1]. The growth of Cu3Sn tends to in-duce the formation of Kirkendall voids, while the growth of Cu6Sn5

does not [1–4]. Zeng et al. [4] reported that a large number ofKirkendall voids formed when electroplated Cu was reacted withthe eutectic PbSn solder at 100–150 �C. These voids located notonly at the Cu/Cu3Sn interface but also within the Cu3Sn layer.The formation of Kirkendall voids is not limited to the eutecticPbSn solder. Kirkendall voids were also reported to form whenelectroplated Cu was reacted with SnAgCu solders [5], or pure Sn[1]. The relatively rapid diffusion of Cu out of the Cu3Sn layer couldbe the major contributing factor for the formation of these voids. Itwas indeed shown that although both Cu and Sn were mobilewithin Cu3Sn, the Cu flux was somewhat greater than that of theSn flux (three times greater at 200 �C) [6–8].

The formation of Kirkendall voids accompanying the Cu3Sngrowth raises serious reliability concerns because excessive voidformation increases the potential for brittle interfacial fracture[2,4,5,9,10]. Since the formation of Kirkendall voids is associatedwith the growth of Cu3Sn, it is reasonable to assume that theamount of Kirkendall voids can be reduced by reducing the Cu3Snthickness. Recently, It was shown that Ni addition to Sn3.5Ag(3.5 wt.% Ag, balance Sn) in amounts as minute as 0.1 wt.% could

substantially retarding the Cu3Sn growth [11–13], raising the hopethat Ni addition can alleviate the problems caused by Kirkendallvoids.

In addition to the effect of Ni addition, it was further pointedout that the type of Cu substrate used also had a marked effecton the Kirkendall voids formation [1]. It was reported thatKirkendall voids formed when electroplated Cu was used, whileno such void was detected when OFHC (oxygen-free high conduc-tivity) Cu substrate was used [1].

The objective of this study was to investigate the effect of Cusubstrate and the effect of Ni additions on the formation ofKirkendall voids within the Cu3Sn phase. Specifically, the reactionsbetween Sn2.5Ag solder doped with different levels of Ni (0–0.1 wt.%) and two different types of Cu substrate, the electroplatedCu and the OFHC Cu substrates, were studied.

2. Experimental

Solder balls with four different compositions, Sn2.5Ag,Sn2.5Ag0.01Ni, Sn2.5Ag0.03Ni, and Sn2.5Ag0.1Ni were preparedfrom 99.999% purity elements. The solder ball compositions wereverified by using ICP-AES (inductively coupled plasma atomicemission spectroscopy), and it was estimated that the reportedcompositions had a maximum 0.005 wt.% uncertainty. For a givensolder ball composition, there was no measurable ball-to-ball var-iation in composition. Each solder ball had a mass of 10 mg.

The solder balls were placed on Cu substrates with 600 lmdiameter wettable openings. The solder could not wet the Cu sub-strate outside this opening, and the solder as a result formed a

0026-2714/$ - see front matter � 2008 Elsevier Ltd. All rights reserved.doi:10.1016/j.microrel.2008.09.010

* Corresponding author. Tel./fax: +886 2 33663745.E-mail addresses: crkao@ntu.edu.tw, kaocr@hotmail.com (C.R. Kao).

Microelectronics Reliability 49 (2009) 248–252

Contents lists available at ScienceDirect

Microelectronics Reliability

journal homepage: www.elsevier .com/locate /microrel

Author's personal copy

sphere only over this wettable opening. Two types of Cu substrateswere used, electroplated Cu and OFHC Cu. The Cu plating solutionwas CuSO4-based and did not contain the C6H12O6S4Na2 additive.The reflow temperature profile had a peak temperature of 235 �Cand 90 s duration during which the solder was molten. The nomi-nal ramp rate and cooling rate were both 1.5 �C/s. After reflow, thesamples were subjected to solid-state aging at 160 �C for 500, 1000or 2000 h.

The samples were then mounted in epoxy, sectioned by using alow-speed diamond saw, and metallurgically polished in prepara-tion for characterization. The reaction zone for each sample wasexamined using an optical microscope and a scanning electronmicroscope (SEM). The compositions of the reaction products weredetermined using an electron microprobe, operated at 15 keV. Inmicroprobe analysis, the concentration of each element was mea-sured independently, and the total weight percentage of all ele-ments was within 100 ± 1% in each case. For every data point, atleast four measurements were made and the average value wasreported.

3. Results

Fig. 1a1–a4 shows the backscattered electron micrographs forthe samples using the electroplated Cu substrate with differentNi additions that were reflowed for one time. The correspondingmicrographs for the OFHC substrates are shown in Fig. 2a1–a4.For both types of Cu substrates, the Cu6Sn5 phase is clearly visible.At this stage, the Cu6Sn5 phase was the only phase observed at theinterface, and Cu3Sn was not observed. Comparing Fig. 1a1–a4 totheir counterparts in Fig. 2a1–a4, one concludes that these two dif-ferent substrates produced essentially the same microstructure. Inother words, the substrate effect was not important in the as-re-flow condition.

Figs. 1a1–a4 and 2a1–a4 also shows that the additions of Ni toSn2.5Ag made the amount of Cu6Sn5 at the interface increase sub-stantially. For Sn2.5Ag without any Ni addition, the Cu6Sn5 layerwas thin, continuous, and void-free, consistent with the classicalscallop microstructure reported in the literature [14–16]. Withthe addition of 0.01 wt.% Ni, Cu6Sn5 still retained the scallop micro-structure. Nevertheless, with the addition of 0.03 wt.% Ni, Cu6Sn5

became thicker, and had many voids between these Cu6Sn5 grains.As had been pointed out in our previous study [11,12], these voidswere originally occupied by trapped solder. During sample prepa-ration, the trapped solder in these voids was etched away. Whenthe Ni addition increased even more, the amount of Cu6Sn5 nearthe interface increased correspondingly. The above observationshowed that Cu6Sn5 formed at the interface without regard tothe Ni additions. The alloy additions, however, did change theamount of Cu6Sn5 and its microstructures at the interface.

Fig. 1b1–d4 shows the micrographs for the samples using theelectroplated Cu substrate with different Ni additions that wereaged for 500, 1000 and 2000 h. The corresponding micrographsfor the OFHC substrates are shown in Fig. 2b1–d4. In all cases,two intermetallic compounds, the Cu6Sn5-based phase and theCu3Sn-based phase, formed. These two compounds both exhibiteda dense layered structure. The voids within the Cu6Sn5 phaseformed during the reflow had disappeared after 500 h of aging.

Figs. 1 and 2 show that, for all combinations of Ni additions andCu substrates, both Cu6Sn5 and Cu3Sn thickened with aging. Duringaging, the most striking feature is that the additions of Ni substan-tially reduced the Cu3Sn to Cu6Sn5 thickness ratio. For sampleswith the same aging, the Ni additions had different effects on thegrowth of Cu6Sn5 and Cu3Sn. Ni additions to Sn2.5Ag increasedthe Cu6Sn5 thickness substantially, as shown in Fig. 3. The Ni addi-tions, however, decreased the Cu3Sn thickness substantially, asshown in Fig. 4. In short, the types of the Cu substrate did not seemto have a clear effect on the growths kinetics of Cu6Sn5 and Cu3Sn.

Fig. 1. Backscattered electron micrographs for Sn2.5Ag alloys with different Ni additions that were reflowed once over the electroplated Cu substrates and then aged at160 �C for 0, 500, 1000, or 2000 h. The symbol SA denotes the Sn2.5Ag alloy, and SA0.01Ni denotes Sn2.5Ag0.01Ni, etc.

Y.W. Wang et al. / Microelectronics Reliability 49 (2009) 248–252 249

Author's personal copy

In addition, both types of Cu substrates produced Cu6Sn5 andCu3Sn with very similar microstructures.

Although different Cu substrates produced the same intermetal-lic compounds at the interface with very similar microstructure andgrowth kinetics. One key difference is the existence of theKirkendall voids within the Cu3Sn phase when the electroplatedCu was used, as shown in Fig. 1. Zoom-in micrographs forFig. 1d1–d4 showing the details of Kirkendall voids are presentedin Fig. 5. For all the solder compositions used, the Kirkendall voidsstarted to appear at an aging time between 500–1000 h. In contrast,there was no such void observed when the OFHC Cu was used evenafter aging for 2000 h, as shown in Fig. 2. With the electroplated Cusubstrate, the amount of the Kirkendall voids increased with the

aging time. For Sn2.5Ag without any Ni addition, 2000 h of agingat 160 �C produced a nearly continuous series of Kirkendall voidsnear the Cu/Cu3Sn interface, shown in Fig. 5a. Such a pronounced de-gree of void formation without question posted a serious reliabilityconcern for solder joints. With Ni additions, the Cu3Sn layer becamethinner, and the amount of Kirkendall voids decreased correspond-ingly, as shown in Fig. 5. In other words, Ni addition did have itseffectiveness in reducing the amount of the Kirkendall voids.

4. Discussion

This study showed that the reaction products were alwaysCu6Sn5 during reflow, and Cu6Sn5+Cu3Sn during aging at 160 �C.

Fig. 2. Backscattered electron micrographs for Sn2.5Ag alloys with different Ni additions that were reflowed once over the OFHC Cu substrates and then aged at 160 �C for 0,500, 1000, or 2000 h. The symbol SA denotes the Sn2.5Ag alloy, and SA0.01Ni denotes Sn2.5Ag0.01Ni, etc.

Fig. 3. Total thickness of Cu6Sn5 formed near the interface versus the aging time at160 �C. Fig. 4. Thickness of Cu3Sn formed near the interface versus the aging time at 160 �C.

250 Y.W. Wang et al. / Microelectronics Reliability 49 (2009) 248–252

Author's personal copy

Although the compound formation did not change with the Niaddition, the amount of compound at the interface did increasesubstantially with Ni additions, as shown in Figs. 1 and 2. Our re-cent preliminary data show that for a given Ni addition, differentcooling rates produced different amounts Cu6Sn5 near the inter-face, with slower cooling rate producing thicker Cu6Sn5. This obser-vation seems to suggest that a major part of the Cu6Sn5 actuallyformed during the solidification of the solder. A faster cooling ratecorresponded to less time for Cu diffusion, and as a result the dis-solved Cu atoms precipitated out as Cu6Sn5 evenly throughout theentire joint. Consequently, less Cu6Sn5 located near the interface.Conversely, a slower cooling rate corresponded to more time forCu diffusion, and as a result the dissolved Cu atoms could diffuseback to the interface to precipitate out as Cu6Sn5, and more Cu6Sn5

located near the interface. Similar cooling rate effect had also beenreported in a very recent study on the reaction between Sn andCuNi alloy substrates with various compositions [17].

An interesting observation was that additions Ni was able toreduce the Cu3Sn thickness substantially. It was effective evenafter aging at 160 �C for 2000 h, which surpassed the typicalindustrial high temperature storage requirement of 150 �C for1000 h. The reason why Ni addition was effective in reducingthe Cu3Sn thickness is unclear at this moment. Several theorieshave been proposed, including thermodynamic arguments [18]and kinetic arguments [12,19]. More studies are needed to eluci-date this point. The reduced Cu3Sn thickness with the Ni addition

was helpful in reducing the amount of voids within Cu3Sn, be-cause the formation of these voids was directly related to thegrowth of the Cu3Sn layer.

This study shows that the types of Cu substrate used have aprominent effect on the observation of Kirkendall voids withinthe Cu3Sn layer. The reason why the Cu substrate had such a strongeffect deserves further discussion. Other than the Kirkendall voids,the two different substrates, electroplated and OFHC, behaved verysimilarly, in term of the intermetallic growth kinetics and micro-structure. Since the growth kinetics and microstructure were con-trolled to a large degree by the interdiffusion of the elements, itseems reasonable to assume that as far as diffusion-related phe-nomena were concerned the two types of substrate behaved thesame. In short, the Cu substrates simply served as the end-mem-bers of the diffusion couples, and maintained a constant Cu ther-modynamic activity at the Cu/Cu3Sn interface.

If the diffusion-related phenomena were the same for bothtypes of substrate, then what was the reason why the Kirkendallvoids were observed for one type of substrate, but not for theother? It was very likely that the impurities in electroplated Cuhelped the nucleation of the voids. Assuming all the atoms inCu6Sn5 and Cu3Sn diffused by vacancy mechanism, one notes thatthere were two necessary conditions for the formation of theKirkendall voids. The first was that the total out-going atomic fluxwas greater than the total in-coming atomic flux for a certaincontrolled volume. There was consequently a net in-coming va-cancy flux into the controlled volume due to the un-balancedatomic flux. These in-coming vacancies were at the same time de-stroyed at vacancy sinks that existed in any materials, such asdislocations. The vacancy concentration was determined by thein-coming vacancy flux minus those vacancies destroyed by var-ious vacancy sinks. The vacancy concentration had to be higherthan a certain critical value for the nucleation of the voids to oc-cur. A vacancy concentration higher than this critical value wasthe second necessary condition for the formation of the voids.In electroplated Cu, impurities and crystal imperfections areimbedded during the electroplating process. These impuritiesand imperfections might play key roles in assisting the voidnucleation. These impurities and imperfections might lower thecritical vacancy concentration value by serving as the heteroge-neous nucleation sites for void nucleation. This was probablythe reason why Kirkendall voids were only observed for electro-plated Cu even though the diffusional characteristics, expressedin term of growth kinetics and phase microstructure, were quitesimilar for both substrates. Very recently, it was pointed out thatthe electroplating additive indeed had a very strong effect on thevoid formation and its distribution within the Cu3Sn layer. It wasindicated that the electroplating additive assisted the formationof the voids by accelerating the void nucleation [20].

The identity of Cu6Sn5 is in fact a rather complicated issue. Attemperatures lower than 186 �C, Cu6Sn5 exhibits a monoclinic g’structure [21,22]. Recent studies, nevertheless, showed that Niwas able to stabilize the high temperature hexagonal g structure.Since Ni was observed in Cu6Sn5 in this study, these Cu6Sn5 mightin fact have the high temperature hexagonal g structure. A recentphase diagram study, that has detailed information on this issue,has been published very recently [23].

5. Summary

(1) Kirkendall voids formed only when electroplated Cu wasused. No such void was observed when OFHC Cu was used.

(2) Apart from the Kirkendall voids, the two different Cu sub-strates behaved very similarly, in term of intermetallic thick-ness and microstructure.

Fig. 5. Zoom-in micrographs for Fig. 1d1–d4 showing enlarged views of Kirkendallvoids.

Y.W. Wang et al. / Microelectronics Reliability 49 (2009) 248–252 251

Author's personal copy

(3) In the case of electroplate Cu substrate, Kirkendall voidsformed after aging at 160 �C for more than 500 h. Theamount of voids increased with the aging time.

(4) The additions Ni was helpful in retarding the Cu3Sn growth,thus reducing the amount of Kirkendall voids when the elec-troplated Cu was used.

Acknowledgements

This work was supported by the National Science Council ofR.O.C. through Grant NSC-95-2221-E-002-443-MY3. The EPMAanalysis performed by Ms. S.Y. Tsai of National Tsing Hua Univer-sity is also acknowledged.

References

[1] Laurila T, Vuorinen V, Kivilahti JK. Interfacial reactions between lead-freesolders and common base materials. Mater Sci Eng 2005;R49:1–60.

[2] Ahat S, Sheng M, Luo L. Microstructure and shear strength evolution of SnAg/Cu surface mount solder joint during aging. J Electron Mater2001;30(10):1317–22.

[3] Vianco PT, Rejent JA, Hlava PF. Solid-state intermetallic compound layergrowth between copper and 95.5Sn–3.9Ag–0.6Cu solder. J Electron Mater2004;33(9):991–1004.

[4] Zeng K, Stierman R, Chiu TC, Edwards D, Ano K, Tu KN. Kirkendall voidformation in eutectic SnPb solder joints on bare Cu and its effect on jointreliability. J Appl Phys 2005;97:024508.

[5] Chiu TC, Zeng K, Stierman R, Edwards D, Ano K. Effect of thermal aging onboard level drop reliability for Pb-free BGA packages. In: Proceedings of 2004electronic components and technology conference, IEEE; 2004. p. 1256–62.

[6] Oh M. Doctoral thesis, Lehigh University; 1994.[7] Oberndorff P. Doctoral thesis, Technical University of Eindhoven; 2001.[8] Paul A. Doctoral thesis, Technical University of Eindhoven; 2004.[9] Mei Z, Ahmad M, Hu M, Ramakrishna G. Kirkendall voids at Cu/solder interface

and their effects on solder joint reliability. In: Proceedings of 2005 electroniccomponents and technology conference, IEEE; 2005. p. 415–20.

[10] Date M, Shoji T, Fujiyoshi M, Sato K, Tu KN. Impact reliability of solder joints.In: Proceedings of 2004 electronic components and technology conference,IEEE; 2004. p. 668–74.

[11] Tsai JY, Hu YC, Tsai CM, Kao CR. A study on the reaction between Cu andSn3.5Ag solder doped with small amounts of Ni. J Electron Mater2003;32(11):1203–8.

[12] Ho CE, Yang SC, Kao CR. Interfacial reaction issues for lead-free electronicsolders. J Mater Sci Mater Electron 2007;18:155–74.

[13] Wang YW, Lin YW, Tu CT, Kao CR. Effects of minor Fe, Co, and Ni additions onthe reaction between SnAgCu solder and Cu. J Alloy Compd, accepted forpublication.

[14] Yang W, Felton LE, Messler Jr RW. The effect of soldering process variables onthe microstructure and mechanical properties of eutectic Sn–Ag/Cu solderjoints. J Electron Mater 1995;24(10):1465–72.

[15] Choi S, Bieler TR, Lucas JP, Subramanian KN. Characterization of the growth ofintermetallic interfacial layers of Sn–Ag and Sn–Pb eutectic solders and theircomposite solders on Cu substrate during isothermal long-term aging. JElectron Mater 1999;28(11):1209–15.

[16] Choi WK, Lee HM. Effect of soldering and aging time on interfacialmicrostructure and growth of intermetallic compounds between Sn–3.5Agsolder alloy and Cu substrate. J Electron Mater 2000;29(10):1207–13.

[17] Vuorinen V, Yu H, Laurila T, Kivilahti JK. Formation of intermetallic compoundsbetween liquid Sn and Various CuNix metallizations. J Electron Mater2008;37(6):792–805.

[18] Gao F, Takemoto T, Nishikawa H. Effects of Co and Ni addition on reactivediffusion between Sn–3.5Ag solder and Cu during soldering and annealing.Mater Sci Eng 2006;A420:39–46.

[19] Yu H, Vuorinen V, Kivilahti JK. Effect of Ni on the formation of Cu6 Sn5 and Cu3

Sn intermetallics. In: Proceedings of the 2006 electronic components andtechnology conference, IEEE; 2006. p. 1204–9.

[20] Kim JY, Yu J. Effects of residual impurities in electroplated Cu on the Kirkendallvoid formation during soldering. Appl Phys Lett 2008;92:092109.

[21] Vuorinen V, Laurila T, Mattila T, Heikinheimo E, Kivilahti JK. Solid-statereactions between Cu(Ni) alloys and Sn. J Electron Mater 2007;36(10):1355–62.

[22] Nogita K, Nishimura T. Nickel-stabilized hexagonal (Cu, Ni)6 Sn5 in Sn–Cu–Nilead-free solder alloys. Scripta Mater 2008;59:191–4.

[23] Schmetterer C, Flandorfer H, Luef CH, Kodentsov A, Ipser H. Cu–Ni–Sn: a keysystem for lead-free soldering. J Electron Mater 2008. doi:10.1007/s11664-008-0522-4 [available on-line].

252 Y.W. Wang et al. / Microelectronics Reliability 49 (2009) 248–252