Materials Transactions, Vol. 46, No. 6 (2005) pp. 1295 to 1300
#2005 The Japan Institute of Metals
Intermetallic Compound Formation and Growth Kinetics in Flip Chip Joints
Using Sn–3.0Ag–0.5Cu Solder and Ni–P under Bump Metallurgy
Dae-Gon Kim*1 and Seung-Boo Jung*2
Department of Advanced Materials Engineering, Sungkyunkwan University,
300 Cheoncheon-dong, Jangan-gu, Suwon 440-746, South Korea
The interfacial microstructure of Sn–3.0Ag–0.5Cu solder with ENIG (electroless Ni/immersion Au) UBM was studied using scanning
electron microscopy and transmission electron microscopy. (Cu,Ni)6 Sn5 intermetallic compound layer was formed at the interface between the
solder and Ni–P under bump metallization upon reflow. However, after isothermal aging, some AuSn4 but with a certain amount of Ni dissolved
in it, i.e., (Au,Ni)Sn4 , appeared above the (Cu,Ni)6 Sn5 layer. Two distinctive layers, P-rich and Ni–Sn–P, were additionally found from the TEM
observation. The analytical studies using EDS equipped in TEM revealed that the averaged composition of the P-rich layer was close to that of a
mixture of Ni3 P and Ni, while that of the Ni–Sn–P layer was analogous to the P-rich layer containing a small amount of Sn in it. The thickness of
the (Cu,Ni)6 Sn5 layer increased with increasing aging time and temperature. The layer growth of the intermetallic compound satisfied a
parabolic law in the given temperature range. The increase of the intermetallic compound layer thickness was mainly controlled by a diffusion
mechanism in the temperature range studied, because the values of time exponent (n) were approximately equal to 0.5. The apparent activation
energy for the growth of (Cu,Ni)6 Sn5 intermetallic compound layer was 55 kJ/mol.
(Received November 30, 2004; Accepted April 28, 2005; Published June 15, 2005)
Keywords: flip chip, Sn–3.0Ag–0.5Cu, electroless nickel, interfacial reaction, reaction diffusion
1.
Introduction
With the push for further miniaturization and improved
performance of integrated circuit devices, the advancement
in electronic packaging technology has led to the growing
application of direct chip attach (DCA) techniques on organic
substrates such as flip chip (FC) assemblies.1,2) The FC
interconnection is the connection of an IC chip to a carrier or
substrate with the active face of the chip facing toward the
substrate. The FC technology is generally considered as the
ultimate first level connection because the highest density can
be achieved and the path length is shortest so that optimal
electrical characteristics are achieved.3–5)
However, as the flip chip technology is beginning to
evolve from the researching area to commercial production,
cost becomes an important issue for its successful implementation. In the early 1990s, a mask-less chemical wafer
bumping technology, which combines electroless Ni plating
and immersion Au process with stencil solder printing, was
developed.1,6) The reason for its low cost is that the process
bypasses the high cost process of UBM sputtering and
photolithography using thick photoresist. Therefore, this
technology is fast becoming the trend among various
electronic packaging companies as it potentially drives cost
down below those expected in the past.
Pb-bearing solders have long been used as an interconnection material due to the advantages of the mechanical
properties, low melting points, and excellent wetting properties.7,8) However, the effort to develop and to promote Pb-free
soldering has intensively occured owing to environmental
and health concerns. Among various Pb-free solders, Sn–Ag–
Cu alloys are regarded as the most promising candidates for
replacement of Sn–Pb solders. They have superior mechanical properties of strength, elongation, creep, and fatigue
*1Graduate
student, Sungkyunkwan University
author, [email protected]
*2Corresponding
resistance than those of Sn–Pb solders.9,10) For this reason,
various associations related with electronics manufacturing
in the world, e.g., JEIDA (Japan Electronic Industry Development Association) and NEMI (National Electronics Manufacturing Initiative), recommend the Sn–Ag–Cu solders for
the replacement of Sn–Pb solders.11) However, the intermetallic compound (IMC) growth in Sn–Ag–Cu solder joints is
faster than that in eutectic Sn–37Pb solder joints due to their
higher reflow temperatures and Sn contents. Too thick IMC
layer is sensitive to stress and provides sites of initiation and
paths of propagation for cracks, because the layer is brittle
and a microstructural mismatch exists between solder and
metallization.12) Therefore, from the electronic package
reliability point of view, it is very important to study the
IMC layer formation and growth kinetics when high temperature storage, burn-in, or extended bake-out creates thick
IMC layers.
Recent years have witnessed a large amount of researches
concerning the interactions between Sn-rich solders and
electroless Ni-based UBMs.13–20) The information published
so far shows that it is still a concern, when using the Ni-based
UBM with popular Sn-based solders. The studies of
interfacial reactions between different solder and UBM
combinations should be conducted because differences are
expected in the types of IMC and mechanism of reaction
product growth. He et al. reported the solid state interfacial
reaction of Sn-based solders with Ni–P UBM.15,16) They
reported that three distinctive layers, Ni3 Sn4 , NiSnP and
Ni3 P, were found between Sn-containing solders and Ni–P
UBM. In addition, they found Kirkendall voids in the Ni3 P
layer. Zeng et al. reflowed Sn–Ag–Cu solder with Ni–P UBM
for five times and found a Ni–Sn–P ternary phase between
Ni3 Sn4 and Ni3 P layers.18) On the other hand, Hwang and
Suganuma and some other researchers reported that the Ni3 P
layer was not composed of pure Ni3 P compounds but
consisted of Ni3 P and Ni from their selected area diffraction
(SAD) pattern analysis and analytical method using trans-
1296
D.-G. Kim and S.-B. Jung
mission electron microscopy (TEM).20) Hwang and Suganuma also found that the NiSnP compound layer could be
indexed to be Ni3 SnP, which has NiAs or InNi2 -type
structure. Various findings, some of which may not be in
agreement with each other, have been reported on the
interfacial reaction between Sn–Ag–Cu solder with Ni–P
UBM, so there should be more needs to identify the
interfacial reaction with various analysis methods.
We focused the formation and growth of IMCs at the
interface between Sn–3.0Ag–0.5Cu solder and electroless
Ni–P UBM, and analytical studies of the Ni3 P layer.
Scanning electron microscopy (SEM) and TEM were
employed to investigate the microstructure and phase analysis of the solder joints. The kinetics of IMC growth and the
effect of aging temperature and time on morphologies of
IMCs were examined.
2.
Experimental Procedures
The solder composition used in this study was a Pb-free
Sn–3.0Ag–0.5Cu (in mass%). Silicon wafers of 4 inches in
diameter were sputtered with Ti of 0.2 mm-thickness and Cu
of 0.8 mm-thickness. The Ti and Cu layers are an adhesion
layer and interconnection pad layer, respectively. The metallized wafers had area array pad arrangements at 500 mm pitch
with rectangular pad opening of 130 mm in width. The
bumping for the flip chip devices was performed using an
immersion Au/electroless Ni–P in thickness of 0.15 and
6 mm, respectively. The Ni–P serves as a diffusion barrier
between the Cu pad and solder, while Au layer is required to
enhance the wettability of the solder to the metallization and
protect the Ni–P layer from oxidation before solder application. In the Ni–P layer, the concentration of P is about 15 at%.
The lateral overlappment of the Au/Ni–P UBM on the chip
passivation layer, SiO2 , was about 5 mm. Solder paste was
applied by stencil printing method and subsequent reflow
employing an IR four zone reflow machine (RF-430-N2,
Japan Pulse Laboratory Ltd. Co.) with a maximum temperature of 523 K for 60 s. After reflow, the average solder bump
height and diameter were about 145 and 190 mm, respectively. Then the wafers were cleaned to remove flux residues.
Figure 1 shows the area arrays of flip chip solder bumps used
in this study and the cross-sectional SEM image of the solder
joint. The reflowed specimens were isothermally aged at 353,
(a)
373, 393 and 423 K for 72, 144, 360, 720, 1200, 1440 and
2400 h in air furnace to observe the interfacial reactions of a
solid state solder. After aging, the specimens were mounted
in epoxy, and then ground with SiC papers followed by
subsequent polishing with 1 and 0.3 mm alumina powders.
Nital etching solution was employed for the purpose of
metallographic observation. The microstructural observation
was conducted with SEM, and the resulting IMCs were
measured by energy dispersive spectrometer (EDS). Especially, for more detailed analysis of the interfacial reaction,
cross-sectional TEM observation was carried out with
analytical studies using EDS (Noran) equipped in TEM
(HF-2000 operated at 200 kV, Hitach Co.). TEM samples
were prepared by ultra-microtomy with a Leica Untracut T
Ultramicrotome by (i) cutting a cross-sectioned cubic sample
followed by molding with commercial epoxy, (ii) setting the
molded sample in ultramicrotomy arm and trimming the
cross-section surface with diamond knife, and (iii) slicing the
trimmed sample by 15 nm in sectioning thickness using a
boat-type diamond knife.
3.
Results and Discussion
For the phase analysis of the reaction layer between the
Sn–3.0Ag–0.5Cu solder and electroless Ni–P layer, crosssectional TEM micrographs were taken. Figure 2 shows a
typical TEM micrograph of the as-reflowed specimen. There
are three phases present on the electroless Ni–P layer in
Fig. 2. The lower two phases are planar and the upper one
phase has a relatively rough morphology. The interfaces
between any two phases were well defined and no precipitate
or pore was observed. Figure 3 summarizes the chemical
compositions of the electroless Ni–P and reaction products
by analytical studies employing EDS equipped in TEM.
From the chemical composition of the uppermost layer, the
reaction product could be indexed to be Cu6 Sn5 containing a
certain amount of Ni in it, i.e., (Cu,Ni)6 Sn5 . Interesting
phenomenon is shown in Fig. 3(c) which the composition of
the P-rich layer does not have the stoichiometry of the pure
Ni3 P compound that was reported elsewhere. Many previous
researches reported that the layer was composed of crystallized Ni3 P compounds for the reaction of Sn-based solders
and electroless Ni–P layer,13–15) while Hwang and Suganuma
reported that the P-rich layer consisted of Ni3 P and Ni in the
(b)
Fig. 1 Area arrays (a) and cross-sectional view (b) of the flip chip solder bumps.
Intermetallic Compound Formation and Growth Kinetics in Flip Chip Joints
(Cu,Ni)6Sn5
Ni3(Sn,P)
Ni3P + Ni
Ni-P
100 nm
Fig. 2
TEM cross-sectional micrograph of the as-reflowed solder joint.
Sn–3.5Ag and Ni–P UBM system with their selected area
diffraction (SAD) pattern studies.20) Our analytical studies
using TEM-EDS revealed that the averaged composition of
the layer was more close to that of a mixture of Ni3 P and Ni
than to that of pure Ni3 P compound.
The crystallization of electroless Ni–P layer into Ni3 P and
Ni was induced by the solder reaction with Ni.9) When
molten solder comes in contact with the original electroless
Ni–P layer, (Cu,Ni)6 Sn5 IMC is formed which induces
crystallization of the adjacent layer of the original amorphous
electroless Ni–P deposit. Due to the consumption of the Ni to
form the Cu–Ni–Sn IMCs, P is expelled to the remaining Ni–
P layer and forms a P-rich layer. Because the high P content
of the P-rich layer, the Ni3 P is initially crystallized, and then
Ni is crystallized beside the nanocrystalline Ni3 P where the P
(a)
Sn
Ni
Sn
1297
content is lower. Therefore, we consider the P-rich layer as a
mixture of the Ni3 P and Ni. Another layer of about 50 nm in
thickness was also found between the (Cu,Ni)6 Sn5 reaction
product and the P-rich layer in Fig. 2. This layer was
identified as a ternary compound which consisted of Ni, Sn,
and P, as shown in Fig. 3(b). The nanocrystalline Ni–Sn–P
phase was also reported elsewhere.15,16,20) This layer must
result from the interaction between the P-rich layer and liquid
Sn atoms. In our analytical studies, the composition of the
layer is similar to that of the P-rich layer containing a small
amount of Sn, and this is supporting the above assumption.
Many studies discussed this Ni–Sn–P layer with their SEM or
electron probe micro-analysis (EPMA), while a very limited
number of researches was carried out using TEM analyses.
Hwang and Suganuma reported that the Ni–Sn–P layer could
be indexed to be Ni3 SnP, which has NiAs or InNi2 -type
structure.20) This is well consistent with our analytical results.
Considering all these data, this ternary phase is believed to be
Ni3 (Sn,P).
Figure 4 shows the back scattered electron (BSE) images
of the interfaces between the solder and electroless Ni–P
layer before and after isothermal aging at 423 K for 72, 144,
360, 1200, and 2400 h. At the as-reflowed interface between
the solder and electroless Ni–P, a layer of (Cu,Ni)6 Sn5 was
formed in thickness of about 0.9 mm, while the pre-deposited
Au layer appears to have dissolved or reacted with the solder
leaving no observable Au at the interface. As shown in
Figs. 4(c) and (f), the dissolved Au formed a binary AuSn4
compounds within the solder region. Some binary AuSn4
compounds within the solder region re-settled at the interface
as a ternary compound, (Au,Ni)Sn4 , as can be seen in
Fig. 4(e). In this study, it could not be observed that the size
of the ternary compounds increased with aging period, while
the increase of the (Cu,Ni)6 Sn5 IMC thickness could be
Cu20Ni35Sn45 (b)
Ni74Sn6P20
Ni
Ni
Ni
P
Cu
(c)
Ni
Ni
Sn
Cu
Ni79P21 (d)
Ni
Ni
Ni86P14
Ni
P
P
Ni
Ni
Fig. 3 Analytical analysis of each layer using EDS equipped in TEM; (a) (Cu,Ni)6 Sn5 , (b) Ni3 (Sn,P), (c) Ni3 P þ Ni, and (d) electroless
Ni–P.
1298
D.-G. Kim and S.-B. Jung
(a)
(b)
(c)
AuSn4
(Cu,Ni)6Sn5
(d)
(e)
(f)
Ag3Sn
Ag3Sn
(Au,Ni)Sn 4
AuSn4
(Cu,Ni)6Sn5
P-rich layer
Ni-P
Fig. 4 SEM interfacial micrographs between the Sn–3.0Ag–0.5Cu and Ni–P layer aged at 423 K; (a) as-reflowed, (b) 72, (c) 144, (d) 360,
(e) 1200, and (f) 2400 h.
(a)
(b)
Cu29.02Ni24.14Sn46.84
Cu12.14Ni38.55Sn49.31
(c)
(d)
Cu37.28Ni19.34Sn43.38
Ag74.02Sn25.98
Fig. 5 Top views of the interfacial IMCs formed between the Sn–3.0Ag–0.5Cu and Ni–P UBM; (a), (b) as-reflowed and (c), (d) aged at
423 K for 1200 h.
observed. After 2400 h of aging time, the thickness of the
(Cu,Ni)6 Sn5 remarkably increased as clearly depicted in
Fig. 4(f). Thickness of the IMCs at the interface between
solder and substrate is very important to the reliability of the
whole package, because too thick IMC layer is sensitive to
stress and sometimes provides sites of initiation and paths of
propagation for cracks. Therefore, the growth of IMC layer
degrades the solder ball joint strength, and thus it is
indispensable to study the IMC layer growth kinetics when
high temperature storage, burn-in, or extended bake-out
creates thick IMC layers.21,22)
Figure 5 illustrates the top views of the interfacial IMCs
resulted by the solder being etched away. Figures 5(a) and
(b) are those of as-reflowed specimens, while Figs. 5(c) and
(d) are those of aged specimens at 423 K for 1200 h. From the
Figs. 5(b) and (d), hexagonal and needle type (Cu,Ni)6 Sn5
IMC grains can be observed in both before and after
isothermal aging. However, the compositions of the two
8
6
1299
Table 1 Growth rate constants and time exponents as a function of aging
temperature.
353 K
373 K
393 K
423 K
2
(Cu, Ni)6Sn5 thickness, d / µ m
2
Intermetallic Compound Formation and Growth Kinetics in Flip Chip Joints
Temp.
(K)
4
Intermetallic
compound
k2
(1019 m2 /s)
R2
n
R2
353
(Cu, Ni)6 Sn5
0.428
0.989
0.601
0.996
373
(Cu, Ni)6 Sn5
0.936
0.993
0.578
0.992
393
(Cu, Ni)6 Sn5
2.476
0.994
0.568
0.996
423
(Cu, Ni)6 Sn5
9.218
0.997
0.592
0.995
2
-41
0
0
500
1000
1500
2000
Q = 55 kJ/mol
2500
-42
-1
Aging time, t / h
-43
2
2
ln k / m s
Fig. 6 Thickness of the (Cu,Ni)6 Sn5 IMC layer formed between the Sn–
3.0Ag–0.5Cu and Ni–P UBM.
types of the IMCs are slightly different, i.e., the hexagonal
type IMCs have more Cu content than that of the needle type
IMCs. The morphology differences between the specimens
before and after aging are also obvious: with the specimens
after aging, the amount of needle type of IMCs decreases and
the size of the hexagonal type of IMCs increases rather than
that of as-reflowed specimens. In addition, it can be also seen
that the Ni content in the hexagonal type IMCs of the aged
specimens is lower than that of the as-reflowed specimens.
This is mainly caused by the depletion of the Ni diffusion
from the Ni–P layer, because the P-rich layer which is formed
between the Ni3 (Sn,P) and Ni–P layer acts as a diffusion
barrier layer of Ni.
The thickness of the reaction layer in the diffusion couples
can be generally expressed by
ðd d0 Þ ¼ ktn
ð1Þ
where d is the thickness of the IMC layer at time t, d0 is the
initial thickness of the IMC layer, k is the growth rate
constant and n is the time exponent.
Figure 6 shows the thickness of the IMC layer ðd d0 Þ2 as
a function of the aging time (t) for various aging temperatures. The IMC layer thicknesses were measured at every
interval of aging time. The squared thickness of the
(Cu,Ni)6 Sn5 IMC layer increased linearly with the aging
time and the growth was faster for the higher aging
temperatures. Generally, the solid-state growth for IMC
layers can follow linear or parabolic growth kinetics. Linear
growth implies that the growth rate is limited by the reaction
rate at the growth site. In contrast, parabolic growth implies
that the growth is limited by volume diffusion. If the growth
processes are controlled by the volume diffusion, the growth
of the IMC layer after aging will follow eq. (1) with
n ¼ 0:5.23)
In Fig. 6, it can be seen that the growth of IMC layer
followed a parabolic law, implying that the growth of the
IMC layer is diffusion-controlled. The growth rate constant
was calculated from a linear regression analysis of ðd d0 Þ2
versus t, where the slope ¼ k2 . Table 1 lists the growth rate
constants calculated for the (Cu,Ni)6 Sn5 IMC at different
aging temperatures. Most of the linear correlation coefficient
-44
-45
-46
2.3
2.4
2.5
2.6
-1
2.7
-3
1/T, T / 10 K
2.8
2.9
-1
Fig. 7 Arrhenius plots of the (Cu,Ni)6 Sn5 IMC layer growth.
values (R2 ) for these plots were greater than 0.98. This
confirms that the growth of the IMC layers is diffusioncontrolled over the temperature range studied, although the
time exponents are not exactly 0.5.
The following Arrhenius relationship was used to determine the activation energy for the (Cu,Ni)6 Sn5 IMC:
Q
2
2
k ¼ k0 exp ð2Þ
RT
where k0 2 is the constant, Q is the activation energy, R is the
gas constant (8.314 J/molK) and T is the aging temperature
(in absolute unit). Figure 7 shows the Arrhenius plot for the
growth of the (Cu,Ni)6 Sn5 IMC layer, and the activation
energy for the (Cu,Ni)6 Sn5 layer growth is obtained to be
55 kJ/mol.
4.
Conclusion
The present study detailed the microstructural evolution of
the IMC layer formed between the Sn–3.0Ag–0.5Cu solder
and electroless Ni–P UBM. The results are summarized as:
(1) In the initial reaction, the reaction product was
(Cu,Ni)6 Sn5 , while the Au layer appeared to have
dissolved or reacted with the solder leaving no
observable Au at the interface. The thickness of the
(Cu,Ni)6 Sn5 increased with aging time and temperatures. In a more detailed analysis of the reaction layer
using TEM, a layer of Ni–Sn–P compound was
observed. The chemical composition of the Ni–Sn–P
compound is close to that of Ni3 (Sn,P) reported by
others.
1300
D.-G. Kim and S.-B. Jung
(2) Some particles of AuSn4 but with a certain amount of
Ni dissolved in it, i.e., (Au,Ni)Sn4 , re-settled above the
(Cu,Ni)6 Sn5 layer after a certain period of aging. The
increase of the (Au,Ni)Sn4 IMC thickness could not be
observed in this study.
(3) A P-rich layer was observed between the Ni–Sn–P
compound layer and the electroless Ni–P layer in the
TEM micrographs. We found that the composition of
the P-rich layer did not have the stoichiometry of the
pure Ni3 P compound which was reported elsewhere.
But the analytical results from the TEM-EDS revealed
that the averaged composition of the layer was more
close to that of a mixture of Ni3 P and Ni which was
reported by Hwang and Suganuma and so on.
(4) The thickness of the (Cu,Ni)6 Sn5 IMC layer increased
with increasing aging time and temperature. There was
a linear relationship between the squared layer thickness and the aging time. The good linear correlation
indicates that the IMC growth is mainly diffusioncontrolled. The apparent activation energy for the
(Cu,Ni)6 Sn5 IMC growth is 55 kJ/mol.
Acknowledgements
This work was supported by grant No. RTI04-03-04 from
the Regional Technology Innovation Program of the Ministry
of Commerce, Industry and Energy (MOCIE).
REFERENCES
1) B. Pahl, C. Kallmayer, R. Aschenbrenner and H. Reichl: Microelectron.
Reliab. 44 (2004) 309–314.
2) J. W. Nah, J. H. Kim, H. M. Lee and K. W. Paik: Acta Mater. 52 (2004)
129–136.
3) J. W. Jang, P. G. Kim, K. N. Tu, D. R. Frear and P. Thompson: J. Appl.
Phys. 85 (1999) 8456–8463.
4) T. Y. Lee, K. N. Tu and D. R. Frear: J. Appl. Phys. 90 (2001) 4502–
4508.
5) N. Duan, J. Scheer, J. Bielen and M. Van Kleef: Microelectron. Reliab.
43 (2003) 1317–1327.
6) J. Kloeser, P. Coskina, R. Aschenbrenner and H. Reichl: Microelectron.
Reliab. 42 (2002) 391–398.
7) J. W. Nah, K. W. Paik, J. O. Suh and K. N. Tu: J. Appl. Phys. 94 (2003)
7560–7566.
8) K. N. Tu, T. Y. Lee, J. W. Jang, L. Li, D. R. Frear, K. Zeng and J. K.
Kivilahti: J. Appl. Phys. 89 (2001) 4843–4849.
9) K. Zeng, V. Vuorinen and J. K. Kivilahti: IEEE T. Electron. Pack. 25
(2002) 162–167.
10) A. Sharif, M. N. Islam and Y. C. Chan: Mat. Sci. Eng. B 113 (2004)
184–189.
11) J. Lau, W. Dauksher, J. Smetana, R. Horsley, D. Shangguan, T.
Castello, I. Menis, D. Love and B. Sullivan: Solder. Surf. Mt. Tech. 16
(2004) 12–26.
12) H. T. Lee, M. H. Chen, H. M. Jao and T. L. Liao: Mat. Sci. Eng. A 358
(2003) 134–141.
13) J. W. Yoon and S. B. Jung: J. Alloy. Compd. 376 (2004) 105–110.
14) J. W. Yoon, C. B. Lee and S. B. Jung: J. Electrn. Mater. 32 (2003)
1195–1200.
15) M. He, Z. Chen and G. Qi: Acta Mater. 52 (2004) 2047–2056.
16) M. He, W. H. Lau, G. Qi and Z. Chen: Thin Solid Films 462–463
(2004) 376–383.
17) K. C. Hung and Y. C. Chan: J. Mater. Sci. Lett. 19 (2000) 1755–1760.
18) K. Zeng and K. N. Tu: Mat. Sci. Eng. R 38 (2002) 55–105.
19) K. N. Tu and K. Zeng: Mat. Sci. Eng. R 34 (2001) 1–58.
20) C. W. Hwang and K. Suganuma: J. Mater. Res. 18 (2003) 2540–2543.
21) C. B. Lee, I. Y. Lee, S. B. Jung and C. C. Shur: Mater. Trans. 43 (2002)
751–756.
22) J. W. Yoon, S. W. Kim and S. B. Jung: Mater. Trans. 45 (2004) 727–
733.
23) K. H. Prakash and T. Sritharan: J. Electron. Mater. 32 (2003) 939–947.