Short communication
Asymmetrical growth of Cu
6
Sn
5
intermetallic compounds due to rapid
thermomigration of Cu in molten SnAg solder joints
Ming-Yung Guo
a, C.K. Lin
a, Chih Chen
a,*, K.N. Tu
baDepartment of Materials Science and Engineering, National Chiao Tung University, Hsinchu 30010, Taiwan, ROC
bDepartment of Materials Science and Engineering, University of California at Los Angeles, Los Angeles, CA 90095-1595, USA
a r t i c l e i n f o
Article history:Received 27 February 2012 Received in revised form 18 May 2012
Accepted 5 June 2012 Available online 27 June 2012
Keywords: A. Intermetallics, miscellaneous B. Diffusion B. Thermal properties C. Joining
a b s t r a c t
We observed asymmetrical growth of Cu6Sn5intermetallic compounds (IMCs) on the two interfaces of
Cu/SnAg/Cu solder joints during reflow at 260C on a hot plate. The IMCs grew to 12.3mm on the cold
end and 3.5mm on the hot end after reflow for 40 min. However, the consumption of Cu on the cold end is less than that on the hot end. We propose that rapid thermomigration of Cu is responsible for the asymmetrical growth of the IMCs. With the simulated thermal gradient of 51C/cm across the liquid
solder, the heat of transport of Cu is calculated as 20 kJ/mol.
Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction
As the microelectronic industry approaching the end of Moore’s law of Very-Large-Scale Integration (VLSI) in silicon chip tech-nology, three dimensional integration circuit (3D IC) emerges to be a promising solution to scaling limit in VLSI circuits[1,2]. In 3DIC, a Si chip is stacked on another Si chip, and through-silicon-vias (TSV) of Cu and microbumps of solder are employed to provide interconnection between the chips. The thickness or height of the solder layer ranges from few microns to 20 microns. Interfacial reaction between TSV of Cu and solder takes place in both the upper and lower chips. Because of the low microbump height, the cross interaction across the two ends of a microbump becomes very serious.
On thermomigration, many studies reported that it occurs during accelerated electromigration tests inflip chip solder joints and it may cause damage in the neighboring joints which carried no current[2e10]. Huang et al. calculated that a thermal gradient of
1000C/cm is needed to observe thermomigration of Sn and Pb
atoms during electromigration tests[4]. For thermomigration of Cu atoms, Chen et al. reported a thermal gradient over 400C/cm is required[10]. The above thermomigration takes place in a solid state. In liquid state reactions, diffusion rate is much faster than that
in solid state, then the thermal gradient needed may be much lower. Whether Cu thermomigration occurs in liquid-state reaction during reflow is of interest, however, up to now no studies report it.
In addition, asymmetrical growth of CueNieSn intermetallic
compound (IMC) has been reported in Cu/solder/Ni structures during reflow[11e13]. However, no asymmetrical growth of Cu6Sn5 IMCs is reported on the two interfaces of Cu/SnAg/Cu solder joints during reflow.
Here, we investigated the Cu thermomigration in Cu/30
m
mSnAg/Cu joints at 260 C on a hot plate as well as in an oven. Asymmetrical reactions due to thermomigration of Cu atoms was observed in the samples reflowed on a hot plate after 5 min, but symmetrical reaction, without thermomigration of Cu, occurred in
samples reflowed in the oven. Finite element simulation and
theoretic calculation were performed to verify the experimental results.
2. Experimental
Sandwich structures of Cu/SnAg/Cu were fabricated. First, a 20 nm Ti and 200 nm Cu seed layer were sputtered on a Si wafer, followed by electroplating of an array of patterned Cu under-bump-metallization (UBM), 100
m
m in diameter and 20m
m in thickness. Second, 19m
m thick Sn2.5Ag solder were electroplated on all the patterned Cu UBMs. The wafer was reflowed at 260C for 1 min to form solder cap on the Cu UBM. Then, the wafer was cut into 1 2 cm2pieces. To fabricate the test samples, a Si die wasflipped* Corresponding author. Tel.: þ886 3573 1814; fax: þ886 3572 4727. E-mail address:chih@cc.nctu.edu.tw(C. Chen).
Contents lists available atSciVerse ScienceDirect
Intermetallics
j o u rn a l h o m e p a g e : w w w . e l s e v ie r . c o m / l o c a t e / i n t e r m e t
0966-9795/$e see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.intermet.2012.06.003
over to align with another die and reflowed at 260C for 3 min. In order to investigate the Cu thermomigration in liquid state, the flip-chip samples underwent additional reflow of 5, 10, 20, 40 min on a hot plate or in an oven maintained at 260C. Then the samples were cooled in air and the cooling rate was about 5C/cm.
After the reflow, the samples were cross-sectioned and polished for interfacial microstructure examination by scanning electron microscopy (SEM). Finite element analysis was carried out to simulate the temperature gradient across the solder joints. The die size was 2308
m
m 2308m
m, which is the same size with the real sample. The thickness of the Si chip was 700m
m. For the boundary conditions, the temperature on the surface of the bottom die was set to be 260C. All the free surfaces of the samples contact with the air and the convention coefficient was set to be 15 w/m2k. 3. Results and discussionFig. 1(a) shows the cross-sectional SEM image for the
as-fabricated sample. The bump height was approximately 30
m
m.The sample inFig. 1(a) experienced 3 min reflow on a hot plate, where the bottom die contacted the hot plate and the top die was exposed to the air. Therefore, the bottom die was the hot end and the top die was the cold end, as labeled in thefigure. IMC of Cu6Sn5 formed at both Cu/solder interfaces. The measured thickness for the interfacial IMCs was 2.3
m
m and 2.9m
m on the hot end and cold end, respectively.Fig. 1(b) presents the cross-sectional SEM image for the sample after additional 10 min reflow at 260C. The Cu6Sn5IMC on the cold end is measured to be 5.2
m
m, whereas it is only 3.5m
m on the hot end. As reflow time increased to 20 min,Fig. 1(c) shows that the IMCs on cold end continue to grow thicker, about 6.7m
m. Yet the IMCs on the hot end did not grow at all, remaining about 3.4m
m. When the reflow time increased to 40 min,Fig. 1(d) shows that the asymmetrical growth appears more significantly. The IMCs on the cold end was 12.3m
m, yet it is still 3.5m
m on the hot end.However, the consumption of Cu UBM is in the opposite direc-tion on the hot end and on the cold end. The Cu UBM was approximately 20 1.0
m
m in the sample before jointing. After the reflow at 260C for 40 min, as shown inFig. 1(d), the Cu UBM decreased to 17.9 0.2m
m on the cold end, but it reduced to 15.1 0.1m
m on the hot end, indicating that the consumption of Cu UBM was faster on the hot end. We recall that the IMC on the hot end grew much slower than that on the cold end.As a controlled experiment for comparison, samples were reflowed in an oven of uniform temperature for various periods. No obvious difference in IMC thickness on both ends was found.Fig. 2
shows the cross-sectional SEM image for the flip-chip sample
reflowed at 260C for 40 min. The IMC thickness on bottom and top
interface was measured to be 5.7 0.2
m
m and 6.3 0.3m
m,respectively. There was no obvious difference in IMC thickness for all the reflow conditions in oven.
Fig. 3summarizes the thickness of Cu6Sn5IMC as a function of reflow time on the hot end and cold end. In addition, the average IMC thickness for the sample reflowed in the oven was also plotted in the figure. The results indicate that the IMC on the cold end grows the fastest and the IMC on the hot end grows slower than that in the oven.
Since no electrical current was in the tests and since in molten state, stress may not be significant, only thermal gradient may be responsible for the asymmetric IMC growth. A thermal gradient can exist in the sample during reflow on a hot plate, because heat was dissipated through the free surface of the top die. However, it is hard to measure the thermal gradient in the solder joint because the temperature difference may be very small across the molten solder. To find out the thermal gradient across the solder joint,
instead, we usedfinite element analysis to simulate the thermal gradient by using a commercial software analysis.Fig. 4shows the temperature distribution in the molten solder when convection coefficient was set to 15 W/m2K. The temperature difference was 0.15C across the solder layer, resulting in a thermal gradient of 51C/cm in the molten solder.
Fig. 1. Cross-sectional SEM image of theflip-chip joints for (a) as-fabricated, (b) reflowed for 10 min, (c) reflowed for 20 min, (d) reflowed for 40 min at 260C on a hot
plate. Asymmetrical growth of the interfacial became clearer as reflow time increased. M.-Y. Guo et al. / Intermetallics 29 (2012) 155e158
To verify if the simulation results are reasonable, calculation was performed on the basis of the equation below[14].
J ¼ CD kT Q* T vT vx (1)
where J is the thermomigration flux, C is concentration, D is
diffusivity, Q*is heat of transport, k is Boltzman constant, T is temperature, andvT=vx is thermal gradient. The solubility of Cu in liquid SnAg solder is 1.54 wt % at 260C[15]. The diffusivity of Cu in molten SnAg solder is taken to be 3.2 105cm2/s[16]. In our
study, we can obtain thermomigration flux from the Cu
consumption data inFig. 1. The thermomigrationflux in units of #atoms/cm2-sec can be expressed as
J ¼ atoms At ¼
A
rD
xNAAMt (2)
where A is cross-sectional area of the solder joint, t is reflow time,
D
x is the Cu consumption thickness due to thermomigration,r
is Cu density (7.3 g/cm3), NAis Avogadro number, M is molecular weight of Cu (63.5 g/mol). InFig. 1(d), the calculated Cu thermomigration flux is 1.49 1016atom/s cm2. ThereforeQ
vTvx
¼ 1 103ðkJ$k=mol$cmÞ (3)
With the simulated thermal gradient of 51C/cm, we obtain the
value of Q* as 20 kJ/mol. This value seems to be reasonable.
Meechan and Lehman studied Cu thermomigration using a pure Cu disc maintaining one end at 1249C and the other end at 530C. The temperature gradient was 1194C/cm. The measured heat of transport is 5 3.5 kcal/mol[17]. Furthermore, Stracke and Herzig investigated Cu thermomigration in Pb at the temperature range of 181Ce303C[18]. They reported that Cu migrated to the cold end and obtained the heat of transport to be 5.1 kcal/mol. In the present study, Cu migrates in molten solder and Cu with a very high diffusivity[19]. Therefore we obtained a higher value of Q*, 20 kJ/ mol.
Although the thermal gradient is approximately 51C/cm across the solder layer, thermomigration of Cu affects significantly the growth of the interfacial IMC. This is because the thickness of the solder layer was only 30
m
m. The diffusion length was approxi-mately two orders of magnitude shorter than the specimen adop-ted by Meechan and Lehman[17], therefore a low thermal gradient is sufficient to cause thermomigration. In the microbumps for 3D IC application, the method of hot pressing may be used to join the microbumps[20], which has a thermal gradient across the solder layer. Hence, Cu thermomigration should play an important role on the interfacial IMC growth in 3D IC.4. Conclusion
In summary, we observed a significant asymmetrical growth of IMC in molten SnAg solder joints during reflowing on a hot plate at 260C. For example, the Cu6Sn5IMC grew to 12.3
m
m on the cold end, yet it was only 3.5m
m on the hot end after the reflow for 40 min. We propose that it is due to thermomigration of Cu.Thermomigration flux can be measured from the asymmetrical
consumption of Cu on the two ends; the consumption is more on the hot end. With a simulated thermal gradient of 51C/cm across the molten solder, the heat of transport of Cu is calculated to be 20 kJ/mol.
Acknowledgments
The authors would like to thank the National Science Council of the Republic of China, Taiwan, for the financial support in this research under Contract No. NSC 99-2221-E-009-040-MY3. References
[1] Tu K-N. Solder joint technology. New York, NY: Springer; 2007.
[2] Lin J-C, Chiou W-C, Yang K-F, Chang H-B, Lin Y-C, Liao E-B, et al. High density 3D integration using CMOS foundry technologies for 28 nm node and beyond. Electronic Components Technol Conf IEDM 2010;10:22e5.
[3] Chen C, Tong H-M, Tu K-N. Electromigration and thermomigration in Pb-free flip-chip solder joints. Annu Rev Mater Res 2010;40:531e55.
[4] Huang A-T, Gusak A-M, Du K-N, Lai Y-S. Thermomigration in SnPb composite flip chip solder joints. Appl Phys Lett 2006;88:141911e3.
[5] Hsiao H-Y, Chen C. Thermomigration in Pb-free SnAg solder joint under alternating current stressing. Appl Phys Lett 2009;94:092107e9.
Fig. 3. The thickness of Cu6Sn5IMCs as a function of reflow time on the hot end, cold
end and in an oven.
Fig. 4. Simulated temperature distributions in the 30-mm liquid solder during reflow on a hot plate with a convection coefficient 15 W/m2K.
Fig. 2. Cross-sectional SEM image of theflip-chip joints underwent additional reflow for 40 min at 260C in an oven.
[6] Chen H-Y, Lin H-W, Liu C-M, Chang Y-W, Huang A-T, Chen C. Thermomigra-tion of Ti inflip-chip solder joints. Scri Mater 2012;66:694e7.
[7] Gu X, Yung K-C, Chan Y-C, Yang D. Thermomigration and electromigration in Sn8Zn3Bi solder joints. J Mater Sci Mater Electron 2011;22:217e22. [8] Alam M-O, Wu B-Y, Chan Y-C, Tu K-N. High electric current density-induced
interfacial reactions in micro ball grid array (mBGA) solder joints. Acta Mater 2006;54:613e21.
[9] Chan Y-C, Yang D. Failure mechanisms of solder interconnects under current stressing in advanced electronic packages. Prog Mater Sci 2010;55: 428e75.
[10] Chen H-Y, Chen C, Tu K-N. Failure induced by thermomigration of interstitial Cu in Pb-freeflip chip solder joints. Appl Phys Lett 2008;93:122103e5. [11] Wang S-J, Liu C-Y. Asymmetrical solder microstructure in Ni/Sn/Cu solder
joint. Scr Mater 2006;55:347e50.
[12] Huang Y-S, Hsiao H-Y, Chen C, Tu K-N. Effect of concentration gradient on interfacial reactions in microbumps of Ni/SnAg/Cu during liquid-state soldering. Scr Mater 2012;66:741e4.
[13] Tseng H-W, Liu C-Y. Evolution of Ag3Sn compound formation in Ni/Sn5Ag/Cu
solder joint. Mater Lett 2008;62:3887e9.
[14] Shewmon P-G. Thermo- and electrotransport in solids. Warrendale, PA: TMS; 1989 [chapter 7].
[15] Zeng K, Tu K-N. Six cases of reliability study of Pb-free solder joints in elec-tronic packaging technology. Mater Sci Eng R 2002;38:55e105.
[16] Ma C-H, Swalin R-A. A study of solute diffusion in liquid tin. Acta Metall 1960; 8:388e95.
[17] Meechan C-J, Lehman G-W. Diffusion of Au and Cu in a temperature gradient. J Appl Phys 1962;33:634e41.
[18] Stracke E, Herzig C-H. Electromigration, thermomigration, and solubility of copper in lead. Phys Stat Sol (a) 1978;47:513e21.
[19] Yeh D-C, Huntington H-B. Extreme fast-diffusion system: nickel in single-crystal tin. Phys Rev Lett 1984;53:1469e72.
[20] Zhan C-J, Chuang C-C, Juang J-Y, Lu S-T, Chang T-C. Assembly and reliability characterization of 3D chip stacking with 30mm pitch lead-free solder micro bump interconnection. Electronic Components Technol Conf 2010:1043e9. M.-Y. Guo et al. / Intermetallics 29 (2012) 155e158