Formation of nearly void-free Cu3Sn intermetallic joints using nanotwinned Cu
metallization
Wei-Lan Chiu, Chien-Min Liu, Yi-Sa Haung, and Chih Chen
Citation: Applied Physics Letters 104, 171902 (2014); doi: 10.1063/1.4874608 View online: http://dx.doi.org/10.1063/1.4874608
View Table of Contents: http://scitation.aip.org/content/aip/journal/apl/104/17?ver=pdfcov Published by the AIP Publishing
Articles you may be interested in
Effects of Ag on the Kirkendall void formation of Sn–xAg/Cu solder joints J. Appl. Phys. 108, 083532 (2010); 10.1063/1.3488629
Electromigration-induced Kirkendall voids at the Cu ∕ Cu 3 Sn interface in flip-chip Cu ∕ Sn ∕ Cu joints Appl. Phys. Lett. 90, 112114 (2007); 10.1063/1.2714100
Electromigration enhanced intermetallic growth and void formation in Pb-free solder joints J. Appl. Phys. 100, 084909 (2006); 10.1063/1.2359135
Effect of contact metallization on electromigration reliability of Pb-free solder joints J. Appl. Phys. 99, 094906 (2006); 10.1063/1.2193037
Electron microscopy study of interfacial reaction between eutectic SnPb and Cu/Ni(V)/Al thin film metallization J. Appl. Phys. 87, 750 (2000); 10.1063/1.371967
Formation of nearly void-free Cu
3Sn intermetallic joints using nanotwinned
Cu metallization
Wei-Lan Chiu, Chien-Min Liu, Yi-Sa Haung, and Chih Chena)
Department of Materials Science and Engineering, National Chiao Tung University, Hsinchu 30010, Taiwan
(Received 6 April 2014; accepted 21 April 2014; published online 30 April 2014)
Cu3Sn intermetallic compounds (IMCs) are more resistant to fracture than solders. In addition, the
Cu3Sn IMCs are more conductive than the solders. In this study, we manufactured Cu3Sn IMCs to
serve as a joint using electroplated nanotwinned Cu as a metallization layer to react with pure Sn at 260C and 340C. The results show that there were almost no Kirkendall voids generated inside the Cu3Sn layer. In addition, the kinetics of the Cu3Sn growth was analyzed to predict the time needed
to form the Cu3Sn joint.VC 2014 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4874608]
A 3D integrated circuit (3D IC) is a stack of different chips that performs vertical integration in a 3D space. Microbumps made of copper-tin are typically used as joints between chips because tin is better than copper-nickel in terms of wetting.1,2The copper-tin system metallur-gical reaction has been well studied over the years. It involves the formation of two types of intermetallic com-pounds (IMCs): Cu6Sn5and Cu3Sn.
3–6
The Cu-Sn interme-tallic compound has become very important for electronics packaging.7–11 Other studies have focused on the properties of the Cu-Sn intermetallic compounds. Cu6Sn5 and Cu3Sn
have relatively good mechanical properties. They are better than Sn in terms of melting temperature, Young’s modulus and hardness.12–16In addition, the Cu3Sn fracture toughness
is 5.72 MPa/m1/2, which is double the value for Cu6Sn5
(2.80 MPa/m1/2). By comparing the two types of IMCs, it has been shown that Cu3Sn is better at resisting fracture.
14
The electrical resistivity of Cu3Sn of 8.9 lXcm is lower than that
of the Sn of 11.5 lXcm and Cu6Sn5of 17.5 lXcm. In
addi-tion, the elastic modulus of Sn, Cu6Sn5, and Cu3Sn are 26.2,
85.56, and 108.3 MPa, respectively.15 Because of its me-chanical properties, Cu3Sn is also more suitable than Cu6Sn5
and Sn as a joint material for microbumps. Because there is a need for more I/O components in electronic components, more I/O components must be manufactured within the same area. As the switching component density increases inside the chip, the flip-chip joint diameter must shrink. In a 100 lm diameter flip-chip joint, the solder volume is signifi-cantly larger than the under bump metallization (UBM) vol-ume. Nevertheless, as the solder diameter is reduced from 100 lm to 20lm, the solder volume is reduced approxi-mately by 1/125. Under that condition, the volume of the UBM in the microbump becomes larger than the solder vol-ume. As a result, the solder may be converted to intermetal-lic compounds. Therefore, the properties of the intermetalintermetal-lic compounds become critical for reliability issues. Using inter-metallic compounds with good properties, Liet al. have tried to use 25 lm-thick Sn foil and two pieces of 10 lm-thick Cu foils to form sandwich structures by reflow to form a Cu3Sn
layer thickness of under 10 lm without generating any
Kirkendall voids.17Cu3Sn layers can also be generated using
aging with a solder layer (less than 1 lm) and a Cu UBM. Furthermore, Liet al. used ultrasonic bonding process at am-bient temperatures to form full Cu/Cu3Sn/Cu joint in 14 s;
however, numerous microvoids formed in the Cu3Sn layer.
18
If the temperature is too high or the duration is too long for forming the Cu3Sn microbumps, the use of Cu and Sn
foils for manufacturing will not be suitable for electronics packaging. For electronics packaging, electroplated copper is used for the UBM and wires. Cu3Sn formed by
metallurgi-cal reactions using electroplated copper and tin can generate Kirkendall voids, which endanger the reliabilities of the microelectronics devices.19–22
Hsiaoet al. adopted densely packed [111] nanotwinned Cu (nt-Cu) for UBM, with a thickness of 20 lm.23 When aged at 150C, the Sn3.5Ag reacted with the nt-Cu to form intermetallic joints with the Cu6Sn5 and Cu3Sn IMCs. No
voids are generated during this metallurgical reaction. In the present work, we use nt-Cu for UBM to react with pure Sn in molten state. By selecting 260C and 340C as reflowing temperatures and changing the solder height for reflowing, we investigate the time required to form complete Cu3Sn joints and the status of the generation of voids in the
joints. Microbumps made of Cu3Sn IMCs are formed after
Sn is completely consumed. By controlling the Sn height, Cu3Sn joints are formed in short periods of time at a
reflow-ing temperature of 260C. This process does not generate voids in the Cu3Sn layer, and the nt-Cu columnar grains are
maintained.
We used electroplating of nt-Cu columnar grain struc-tures to manufacture nt-Cu/Sn/nt-Cu and regular copper electroplating to manufacture the Cu/Sn/Cu structure. The manufactured nt-Cu specimen can be divided into the pad and the film. A 20 nm layer of Cu was sputtered onto a Si wafer with a 100-nm-thick Ti layer. The Cu was used as a seed layer. The pad manufacturing process included an expo-sure and a development process. After the pattern was defined, the nt-Cu was electroplated onto it. The procedure for the nt-Cu electroplating has been reported previously.23 The electroplated nt-Cu pad was circular, with a diameter of 100 lm and a thickness of 20 lm. The nt-Cu film was 10 lm thick, whereas the regular electroplated Cu film was 60 lm thick. When the tin was electroplated on the copper a)Author to whom correspondence should be addressed. Electronic mail:
chih@mail.nctu.edu.tw
substrates at room temperature, the thickness of the electroplated-tin layer on the nt-Cu pad was 0.44 lm. The electroplating-tin-layer thickness on the nt-Cu film was 1 lm. The tin-layer thickness on the regular Cu film was 60 lm. For joining the tin electroplated substrates and the copper electroplated substrate, we applied flux on the Sn layer and reflowed the samples at 260C for 1 min. We then applied pressures of 9.6 MPa and 0.78 MPa to the 0.44 lm bumps and 1 lm films, respectively. After a 1-min reflow, the pad-to-pad and film-to-film microbumps were formed. We manufactured 0.44, 1, and 10 lm-thick Sn layer between nt-Cu UBMs and a 60 lm-thick Sn layer between regular Cu UBMs. For the flip-chip microbumps associated with the 60 lm -thick layers, the reflow lasted for 24 h at 340C. For the flip chip microbumps associated with the 10 lm-thick layers, the reflow lasted for 20, 60 min, or 24 h at 260C or for 5, 40, or 60 min at 340C. For the flip chip microbumps associated with the 1 lm-thick layers, the reflow lasted for 10 min at 260C with 0.78 MPa added pressure. For the flip chip microbumps associated with the 0.44 lm -thick layers, the reflow lasted for 3 min at 260C with a pressure of 9.60 MPa. All of the samples went through the reflow pro-cess under normal atmospheric conditions. TableI summa-rizes the structures and experimental conditions for the samples in this study. After the reflow, the samples under-went grinding and polishing using alumina powder to reveal their microstructures. Scanning electron microscopy (SEM) was used to observe microstructural changes. A focused ion beam (FIB) was used to grind and polish the cross sections to observe the Kirkenall voids.
When using the regular Cu film without nanotwinned columnar grains and the liquid-state Sn film for the reflow reaction, Kirkendall voids were observed in the Cu3Sn layer
close to the original interface of Cu3Sn and Cu. Figure 1
shows these microstructures in an SEM cross-section after the metallurgical reaction at 340C for 24 h. Th original Sn thickness was 60 lm. Regular electroplated Cu was used as the UBM materials. In the metallurgical reaction, the tin layer was almost consumed such that a 100-lm-thick Cu3Sn
layer was formed. Many Kirkendall voids were generated in the Cu3Sn layer, which is consistent with previous
reports.19–22The Kirkendall voids may weaken the mechani-cal properties of the Cu3Sn IMC joint. We tried to eliminate
the Kirkendall voids by adopting the nt-Cu as the UBM materials for the reflow reaction. By using nt-Cu, almost void-free Cu3Sn joints can be manufactured. Figure 2(a)
shows the initial state of the nt-Cu/Sn/nt-Cu with a solder thickness of 10 lm at 260C after 1 min of soldering. The
tin was the main part of the joint. The thickness of the scallop-type Cu6Sn5 was approximately 2 lm. Figure 2(b)
shows the state after continuous reflow at 260C for 20 min. The tin was gradually consumed to produce Cu6Sn5 and
Cu3Sn IMCs. The Cu6Sn5 IMCs on both sides were
con-nected to each other. In addition, a 1-lm-thick l Cu3Sn was
formed after continuous reflowing for 60 min between the Cu6Sn5 and the Cu. As shown in Figure 2(c), the tin was
completely consumed, and the Cu6Sn5 was connected to a
single layer. Layers of 2.30-lm-thick and 2.87-lm-thick Cu3Sn were formed at the top and the bottom side,
respec-tively. Finally, after reflow for 24 h, the Cu6Sn5 was
con-verted to Cu3Sn to form a pure Cu3Sn joint. At the interface
of the Cu3Sn layer and the Cu, almost no Kirkendall voids
were generated, as shown in Figure2(d).
When the reflowing temperature increased, the metallur-gical reaction speed increased such that the time needed to
TABLE I. Structures and experimental conditions for the test samples of Cu/Sn/Cu and nt-Cu/Sn/nt-Cu.
Sample Sn thickness Applied pressure Reflow temperature Reflow time Cu/Sn/Cu 60 lm No 340C 24 h
nt-Cu/Sn/nt-Cu 10 lm None 340C 1, 5, 40, 60 min
10 lm None 260C 1, 20, 60 min, 24 h
1 lm 0.78 MPa 260C 10 min
0.44 lm 9.60 MPa 260C 3 min
FIG. 1. SEM image taken from the polished cross-section of Cu/Cu3Sn/Cu
without nanotwinned Cu. Many voids were found at the Cu/Cu3Sn interface.
FIG. 2. SEM images taken from the polished cross-section of nt-Cu/Sn/nt-Cu sample reflowed at 260C for: (a) as-reflowed; (b) 20 min; (c) 60 min; and (d) 24 h.
form the Cu3Sn IMC was reduced. Figure3(a)shows the
ini-tial state of the nt-Cu/Sn/nt-Cu at 260C after 1 min of reflow for a 10 lm -thick solder. After 5 min of reflow at 340C, the tin was completely consumed, with a layer of Cu6Sn5 formed at the both sides of the joint. The Cu6Sn5
IMCs became the main part of the solder, producing 2.34 and 2.90 lm-thick layers of Cu3Sn at the top and the bottom
sides, respectively, as shown in Figure3(b). After 40 min of reflow, the metallurgical reaction consumed the Cu6Sn5
IMCs, while the Cu3Sn continued to grow. Cu3Sn layers of
7.76 lm and 6.92 lm were generated at the top and the bot-tom sides of the joint, respectively, as shown in Figure3(c). Finally, after 1 h of reflow, the Cu3Sn layers at the both sides
started to make contact. The Cu3Sn then became the main
part of the solder as the connection was formed. Almost no Kirkendall voids were observed at the interfaces between the Cu3Sn layer and Cu, as shown in Figure3(d). For a tin
thick-ness of 10 lm, a 23 lm-thick Cu3Sn layer was formed by
reflowing at 260C. Increasing the temperature reduced the thickness of the Cu3Sn layer formed during the reflow.
By reducing the Sn thickness, we were able to lower the reflow temperature to 260C to form a complete Cu3Sn
joint. We were also able to reduce the time needed for the process. As shown in Figure4(a), using a solder thickness of 1 lm at 260C and a reflow time of 10 min, we were able to form a 2.7-lm-thick Cu3Sn joint with almost all of the
Cu6Sn5 consumed. No voids were produced in the Cu3Sn
layer. To affirm the void distribution, we used a FIB to polish the surface of the Cu3Sn layer, as shown in Figure 4(b).
Almost no Kirkendall voids were produced in the Cu3Sn
layer. Therefore, lowering the Sn thickness can effectively reduce the reflow time needed to form a Cu3Sn joint.
We also continued to lower the solder amount. As shown in Figure4(c), for a 0.44-lm-thick Sn layer formed by a 3-min reflow at 260C, a 1 lm-thick Cu3Sn joint was formed
with-out any voids. Therefore, we can use our electroplating method to electroplate nt-Cu to control the solder thickness and the reflow time needed for the solder to form nearly void-free Cu3Sn IMC joints.
The nt-Cu is able to eliminate Kirkendall voids because of the following two reasons. First, the nt-Cu has less resid-ual sulfur impurities after electroplating. It is reported that sulfur atoms in Cu may cause Kirkendall voids in Cu/Sn reactions.24–26 When electroplating Cu films, the sulfur atoms in the electroplating solution can precipitate out and become incorporated into the grain boundaries.25In contrast, the nt-Cu columnar grains can effectively reduce the area of grain boundary per unit volume and further reduce the sulfur impurities left in the Cu films. Therefore, nt-Cu can effec-tively reduce the generation of Kirkendall voids. Second, the densely packed nanotwins in Cu may serve as vacancy sinks. Several researchers reported that there are many defects, such as steps and kinks in Cu nanotwins.27–29These defects are all vacancy sinks; therefore, the vacancy concentration is effectively reduced, such that the concentration does not exceed the saturation concentration required to nucleate the voids.
The kinetics of the Cu3Sn growth was also analyzed
below. For the microbumps with solder thicknesses of 10 lm, it took 24 h at 260C to form Cu3Sn joints, whereas
it took 1 h at 340C to grow Cu3Sn joints. Higher
tempera-tures can increase the IMC growth velocity. It is known that
FIG. 3. SEM images taken from the polished cross-section of the nt-Cu/Sn/nt-Cu sample reflowed at 340C for (a) as-reflowed; (b) 5 min; (c) 40 min; and (d) 1 h.
FIG. 4. (a) SEM and (b) FIB images of the nt-Cu/Cu3Sn/nt-Cu sample after
the reflow at 260C for 10 min. (c) SEM image of the nt-Cu/ Cu
3Sn/nt-Cu
the scallop-shaped Cu6Sn5 grows under ripening control,
while the layered Cu3Sn grows under diffusion control.
17
We used the Arrhenius equation30 and the kinetic growth equation to estimate the time for converting the Sn to Cu3Sn
IMCs, as follows: k2ðTÞ ¼ k2 0exp Ea RT ; (1) xðtÞ ¼ kt0:5: (2)
The term k denotes the growth rate constant, while k0is the
frequency factor, Ea is the activation energy, T is the
abso-lute temperature, R (¼8.314 J/molK) is the gas constant and x and t are the Sn thickness and time needed to form Cu3Sn,
respectively. We combined Eqs. (1) and (2) to obtain the following equation: x t; Tð Þ ¼ k0:5 0 exp Ea 2RT t0:5: (3)
Equation (3) shows that a temperature increase can exponentially reduce the reflow time. At a fixed temperature, the thickness is proportional to the square root of the reflow time. In our experiment, the temperature increase was used to reduce the reflow time. There was still no generation of Kirkendall voids in the Cu3Sn layer. However, high reflow
temperatures can lead to the damage of electronic compo-nents. Therefore, we chose to change the solder height such that the reflow temperature could be reduced to 260C, as shown in Figure4. After lowering the solder heights to either 1 or 0.44 lm, Cu3Sn joints were formed after either 10 or
3 min. Based on the time needed to form the pure Cu3Sn
layers for these three solder heights at 260C, we obtain
xðtÞ ¼ 0:0338 t0:5þ 0:0078; (4)
where x(t) is in microns and t is in seconds. The derived n¼ 0.5 value shows that the Sn reaction to form Cu3Sn is
occurs under diffusion control. We can use Eq.(4) to esti-mate the reflow time or solder thickness. For example, we need a 1 min reflow at 260C to form a Cu3Sn joint. After
using the formula, we can derive the thickness needed for electroplating, which is 0.27 lm. The melting point of the Cu3Sn IMC is as high as 676C.15Its Young’s modulus is
also higher than that of Pb-free solders. Thus, the Cu3Sn
joints are expected to have better electromigration resist-ance.31,32 In addition, the fracture toughness of Cu3Sn is
higher than those of Cu6Sn5 and Pb-free solders. The
me-chanical properties of Cu3Sn are also better than those of
Cu6Sn5.33Therefore, the Cu3Sn joints may have great
poten-tial as interconnects in 3D ICs.
In summary, using nt-Cu as the metallization films, we can fabricate Cu3Sn IMC joint almost without Kirkendall
voids. At 260C, the reflow converted the entire molten-state of Sn and Cu6Sn5 into a Cu3Sn joint while
maintaining the nt-Cu columnar grain structure. We demon-strated that we can lower the temperature for the formation
of the Cu3Sn joint to 260C with duration of several
minutes. In addition, using kinetics analysis, we can forecast the time needed for Sn to form Cu3Sn joints at 260C. These
Cu3Sn microbumps are expected to possess better
electromi-gration resistance and mechanical properties. The fabrication of void-free Cu3Sn joints has great potential to be applied to
3D integrated circuit microbumps and other joints.
We are grateful for financial support from the National Science Council, Taiwan, under the Contract No. NSC 99-2221-E-009-040-MY3.
1H. K. Kim, H. K. Liou, and K. N. Tu,J. Mater. Res.
10, 497 (1995).
2
P. G. Kim, J. W. Jang, T. Y. Lee, and K. N. Tu,J. Appl. Phys.86, 6746 (1999).
3D. Ma, W. D. Wang, and S. K. Lahiri,J. Appl. Phys.91, 3312 (2002). 4J. W. Yoon, Y. H. Lee, D. G. Kim, H. B. Kang, S. J. Suh, C. W. Yang, C.
B. Lee, J. M. Jung, C. S. Yoo, and S. B. Jung,J Alloys Compd.381, 151 (2004).
5M. L. Huang, T. Loeher, A. Ostmann, and H. Reichl,Appl. Phys. Lett.86,
181908 (2005).
6
W. C. Kuan, S. W. Liang, and C. Chen,Microelectron. Reliab.49, 544 (2009).
7B. Lee, J. Park, J. Song, K. W. Kwon, and H. J. Lee,J. Electron. Mater.
40, 324 (2011).
8
K. N. Tu,Acta Metall.21, 347 (1973).
9
K. N. Tu and R. D. Thompson,Acta Metall.30, 947 (1982).
10H. K. Kim, H. K. Liou, and K. N. Tu,Appl. Phys. Lett.66, 2337 (1995). 11C. Y. Liu, C. Chen, and K. N. Tu,J. Appl. Phys.
85, 3882 (1999).
12
N. S. Bosco and F. W. Zok,Acta Metall.52, 2965 (2004).
13
H. P. R. Frederise, R. J. Field, and A. Feldman,J. Appl. Phys.72, 2879 (1992).
14X. Deng, N. Chawla, K. K. Chawla, and M. Koopman,Acta Metall.
52, 4291 (2004).
15
D. R. Frear, S. N. Burchett, H. S. Morgan, and J. H. Lau,The Mechanics of Solder Alloy Interconnections (Van Nostrand Reinhold, New York, 1994), p. 60.
16
C. C. Lee, P. J. Wang, and J. S. Kim, “Are intermetallics in solder joints really brittle?,” inProceedings of the IEEE Electronic Components and Technology Conference, IEEE, New York, 2007, p. 648.
17J. F. Li, P. A. Agyakwa, and C. M. Johnson,Acta Metall.
59, 1198 (2010).
18
M. Li, Z. Li, Y. Xiao, and C. Wang,Appl. Phys. Lett.102, 094104 (2013).
19
K. Zeng, R. Stierman, T. C. Chiu, D. Edwards, K. Ano, and K. N. Tu,
J. Appl. Phys.97, 024508 (2005).
20R. Labie, W. Ruythooren, and J. V. Humbeeck, Intermetallics
15, 396 (2007).
21
C. E. Ho, S. C. Yang, and C. R. Kao,J. Mater. Sci. Mater. Electron.18, 155 (2007).
22J. Y. Kim, Jin Yu, and S. H. Kim,Acta Metall.
57, 5001 (2009).
23
H. Y. Hsiao, C. M. Liu, H. W. Lin, T. C. Liu, C. L. Lu, Y. S. Huang, and C. Chen,Science336, 1007 (2012).
24Jin Yu and J. Y. Kim,Acta Metall.56, 5514 (2008). 25J. Y. Kim and J. Yu,Appl. Phys. Lett.
92, 092109 (2008).
26
L. Yin and P. Borgesen,J. Mater. Res.26, 455 (2011).
27
T. C. Liu, C. M. Liu, Y. S. Haung, C. Chen, and K. N. Tu,Scr. Metall.68, 241 (2013).
28L. Lu, X. Chen, X. Huang, and K. Lu,Science
323, 607 (2009).
29
Y. M. Wang, F. Sansoz, T. LaGrange, R. T. Ott, J. Marian, T. W. Barbee Jr, and A. V. Hamza,Nature Materials12, 697 (2013).
30G. T. Lim, B. J. Kim, K. Lee, J. Kim, Y. C. Joo, and Y. B. Park,
J. Electron. Mater.38, 2228 (2009).
31
H. Y. Chen and C. Chen,J. Mater. Res.26, 983 (2011).
32
S. Y. Huang, C. J. Zhan, Y. W. Huang, Y. M. Lin, C. W. Fan, S. C. Chung, K. S. Kao, J. Y. Chang, M. L. Wu, T. F. Yang, J. H. Lau, and T. H. Chen, “Effects of UBM structure/material on the reliability performance of 3D chip stacking with 30 m-pitch solder micro bump interconnections,” in Proceedings of the IEEE Electronic Components and Technology Conference, IEEE, San Diego, 2012, p. 1287.
33G. Ghosh,J. Mater. Res.
19, 1439 (2004).
