Materials Transactions, Vol. 45, No. 3 (2004) pp. 689 to 694
Special Issue on Lead-Free Soldering in Electronics
#2004 The Japan Institute of Metals
Low Cycle Fatigue Properties of Ni added Low Silver Content
Sn-Ag-Cu Flip Chip Interconnects
Yoshiharu Kariya1; *1 , Takuya Hosoi2; *3 , Takashi Kimura1; *2 , Shinichi Terashima3
and Masamoto Tanaka3
1
National Institute for Materials Science, Tsukuba 3050044, Japan
Graduate School of Shibaura Institute of Technology, Tokyo 1088548, Japan
3
Advanced Technology Research Laboratories, Nippon Steel Corporation, Futtsu 2938511, Japan
2
The straddle fatigue test has been performed to study the fatigue properties of Sn-1.2 mass%Ag-0.5 mass%Cu-0.05 mass%Ni for flip chip
interconnections. The low cycle fatigue resistance of the alloy is equivalent to that of Sn-3 mass%Ag-0.5 mass%Cu alloy, even though the
fatigue endurance of Sn-1 mass%Ag-0.5 mass%Cu alloy was poorer than that of the 3 mass%Ag alloy. The alloy has fine microstructure and
Ag3 Sn intermetallic compound makes a network structure together with fine (Cu,Ni)6 Sn5 compound. The microstructure resulted in high cyclic
strain hardening exponents, which leaded to good low cycle fatigue endurance of the alloy.
(Received September 19, 2003; Accepted December 17, 2003)
Keywords: flip chip, fatigue life, shear fatigue, nickel addition, ball grid Array (BGA) joints, low silver content, lead-free solder, tin-silvercopper
Increasing environmental and health concern over the
toxicity of lead has provided a driving force to ban the use of
Pb-Sn solders, and has stimulated to develop lead-free
solder.1,2) Environment friendly technology in electronics
involves Pb-free flip-chip solder development. Excellent
fatigue resistance is required for Pb-free flip-chip solder ball
as the CTE mismatch between Si chip and organic substrate
imposes a significant shear strain into the solder bumps.3) SnAg-Cu family will become candidates for flip-chip interconnects from the view point of fatigue resistance,4,5) while a
correlation between chemical compositions and fatigue
resistance is still not clarified to date. In our previous study,6)
we investigated the relationship between the low cycle
fatigue life of Sn-xAg-0.5 mass%Cu flip-chip interconnects
and silver content. The fatigue life of 3 mass%Ag solder alloy
was longer than that of other alloys tested. On the other hand,
low silver content alloy such as the 1 mass%Ag alloy
exhibited relatively good fatigue resistance in high strain
regime, whereas the fatigue life in low strain conditions was
the poorest in the alloys tested. From the result, we concluded
that 3 mass% is recommended for the silver content in the SnAg-Cu flip-chip bump for wide condition. Nevertheless, low
silver content Sn-Ag-Cu alloys remains to be further
investigated from the viewpoint of preventing brittle failure
at intermetallic layers because of both its ductile nature and
alloy cost saving. Fatigue properties of the low silver content
alloy need to be enhance to apply to flip-chip application.
Guo reported that small amount of nickel has a potential to
enhance creep strength of Sn-Ag system.7) Therefore, we
selected nickel as the fourth element to improve fatigue
properties of the low silver content Sn-Ag-Cu alloy, and the
mechanical shear fatigue life of flip-chip interconnections
made using the Ni added alloy has been investigated in the
similar manners in previous study.
2.
Experimental
2.1 Specimen
A 8 mm 8 mm Si chip was joined on 0.6 mm thick FR-4
substrate using 0.3 mm diameter solder ball with 0.5 mm
pitch as shown in Fig. 1. The ball arrangement is shown in
Fig. 2. 170 balls were mounted on the chip. Sn-1.2 mass%Ag0.5 mass%Cu-0.05 mass%Ni solder alloy was used in this
study. Sn-3.0 mass%Ag-0.5 mass%Cu and Sn-1.0 mass%Ag0.5 mass%Cu alloys were also used to compare fatigue
resistance. The surface was finished with Cr/Ni/Au on
aluminum electrodes for chip side and Ni/Au for copper
traces on the substrate. The FR-4 substrate was divided into
two parts before joining, and then the Si chip and the divided
substrates were joined with 1 mm gap as shown in Fig. 1. All
FR-4 Substrate
5
Introduction
3.
1.
φ
Solder ball
φ 2.5
30
1
Si Chip
12
14.5
5
10
15
Unit:mm
20
*1Eco-Materials
Center, National Institute for Materials Science
Analysis station, National Institute for Materials Science
*3Graduate Student, Shibaura Institute for Technology
*2Materials
Fig. 1 Schematic illustration of flip chip interconnects for split board shear
fatigue test.
690
Y. Kariya, T. Hosoi, T. Kimura, S. Terashima and M. Tanaka
17 x 5 balls
8
Unit:mm
0.4
8
Fig. 2
Schematic illustration of ball arrangement.
Fig. 5 Final assembly of specimen with fixing jigs.
220µm
Si Chip
Cr/Ni/Au
320µm
Ni/Au
Cu Pad
FR-4 Substrate
100µm
Fig. 3 Cross-sectional image of the flip-chip joint.
samples were joined by the conventional reflow process in air
at the peak temperature of 518 K for 60 s with a RMA flux.
Figure 3 shows cross sectional optical image of the flip-chip
bump. No significant voids were observed, and the bump
height and width were approximately 220 mm and 320 mm,
respectively for all joints.
2.2 Shear fatigue test
In this study, the mechanical shear fatigue test was
employed. In general, the thermal cycle testing is used for
reliability evaluations, while the test continues to be
conducted for a long period of time. Another approach is to
mechanically cycle the interconnects rather than to produce
the strains by changing the temperature, and thermomechanical fatigue life of the solder joints can be obtained
from mechanical fatigue test.8,9) In order to evaluate the
mechanical shear fatigue properties of the flip-chip interconnections, the straddle fatigue test was used in this
study.6,10,11) The methodology of the straddle fatigue test is
imposition of shear deformation into the solder alloy by
mechanical displacement of the substrates as shown in Fig. 4.
The fatigue test was performed using a precise displacement
controlled by the fatigue test machine. A linear DC motor
was employed for an actuator of the machine, and the
Si Chip
H
∆d/2
L
Solder ball
∆d/2
L+∆d
Fig. 4
Schematic illustration of shear deformation of split board specimen.
displacement resolution of the actuator and maximum load
capacity were 20 nm and 200 N, respectively. The actuator is
supported by air pressure without any mechanical contact to
avoid a friction load that influences measurement accuracy of
the load. The specimen was attached to the fixture using two
steel shims and eight screws as shown in Fig. 5. The fixture is
designed to position the bump center at the load line to
suppress bending deformation of the sample. The total
displacement range was measured using the eddy current
type gap sensor attached to the fixture at the room temperature (298 K) in air. The room temperature was maintained to
1 K throughout the test. The resolution of the gap sensor is
less than 0.1 mm at the room temperature. The total displacement range varied from 25 to 5 mm, and the loading
profile is a continuous cycling with the shear strain rate of
1 102 s1 . The fatigue life was defined as the number of
cycles at 20% load drop from initial load.
2.3 Microstructural observation
The specimens were sectioned by a tungsten wire saw to
prevent any damage. They were ground with two grades of
SiC papers (#500, and 1200) and then mechanically polished
with a diamond paste (15 mm). Finally, the specimens were
polished with colloidal silica suspension. After polishing,
they were observed using a scanning electron microscope in
backscatter electron mode (BSE). A FE-EPMA (electron
probe microanalyzer) developed by NIMS was used for
quantitative analysis and elemental mapping. The EPAMA
employs a schottky type field emission high brightness
electron gun which can extremely smaller probe diameter
than the conventional one with a W hairpin type electron gun,
and is able to analyze sub micron area of 200 nm or less.12)
3.
Results and Discussion
3.1 Microstructure
The as-fabricated microstructures of each solder bump are
shown in Fig. 6. The microstructure of the alloys consists of
sub micron size intermetallic compounds and a -Sn matrix
for all alloys. In the 1 mass%Ag bump, the intermetallics
coarsely dispersed within the matrix, and a volume of the
intermetallics is less than that in other bumps as can be seen
in the figure. The 3 mass%Ag bump has much more
intermetallics compared with the 1 mass%Ag bump, and
those appeared around the -Sn grains with a network
Low Cycle Fatigue Properties of Ni added Low Silver Content Sn-Ag-Cu Flip Chip Interconnects
(a)
(b)
10
10µm
691
(
(c)
10µm
10
10µm
Fig. 6 Backscatter electron images showing microstructure of each solder bump (a: Sn-1.0 mass%Ag-0.5 mass%Cu, b: Sn-3.0 mass%Ag0.5 mass%Cu, c: Sn-1.2 mass%Ag-0.5 mass%Cu-0.05 mass%Ni).
Cu Intensity ( arb. units )
structure. On the other hand, the Ni added alloy has relatively
small -Sn grains, and a number of the intermetallics can be
identified in the bump compared with the 1 mass%Ag bump,
even though the difference in the silver content is only
0.2 mass% between the 1 mass%Ag and the Ni added alloy.
In addition, the Sn grains are perfectly decorated with the
intermetallics, which make a network structure. The microstructure suggests that the small amount of Ni fine down the
microstructure of low silver content Sn-Ag-Cu alloy, which
may enhance the mechanical properties of the alloy.
Ag3 Sn and (Cu,Ni)6 Sn5 intermetallic compounds were
identified by the EPMA. Ag3 Sn was determined by quantitative point analysis, and (Cu,Ni)6 Sn5 compound was
determined by X-ray image data using X-ray intensity scatter
diagram method. The X-ray intensity scatter diagram for Sn1.2 mass%Ag-0.5 mass%Cu-0.05 mass%Ni is shown in
Fig. 7. Same results were observed for other alloys tested
here. Figure 8 shows EPMA elemental mapping image of Ag,
Cu and Ni for the 1 mass%Ag and the Ni added solder bumps.
As the electrodes were plated by nickel, nickel is dissolved in
Cu6 Sn5 for all bumps, even if nickel was not alloyed before
soldering. Both silver and copper identified coarsely within
the microstructure of the 1 mass%Ag alloy as can be seen in
(Cu8, Ni2)6Sn5
(Cu3, Ni2)6Sn5
the figure, which means Ag3 Sn and (Cu,Ni)6 Sn5 coarsely
disperse in the matrix. On the other hand, silver and copper
present very finely in the microstructure of the Ni added
alloy, and Ag3 Sn compound makes a perfect network
structure together with fine (Cu,Ni)6 Sn5 compound as shown
in the figure. The fine network structure of intermetallic
compounds in the Ni added alloy suggests that Ag3 Sn and
(Cu,Ni)6 Sn5 crystallize rapidly compared with the 1 mass%Ag alloy after crystallizing pro-eutectic Sn grains, while a
detail mechanism to form the microstructure is not able to
clarified at that time.
3.2 Relationship between displacement and fatigue life
Figure 9 shows the relationship between the total displacement range and fatigue life. The data on the total displacement range versus the fatigue life obey power law type
relationship as shown in Fig. 9. The fatigue life of the
1 mass%Ag alloy is evidently poorer in any displacement
conditions than the other alloys tested, and the life is almost
one third of the life of the 3 mass%Ag alloy. On the other
hand, the Ni added low silver content alloy has superior
fatigue resistance in any displacement regimes, and the life is
almost equivalent to that of 3 mass%Ag alloy. Since an
increase in silver content up to 2 mass% did not improve low
cycle fatigue resistance of Sn-xAg-0.5 mass%Cu alloy in our
previous study,6) the result suggests that small amount of Ni
has an effect to enhance low cycle fatigue resistance of low
silver content Sn-Ag-Cu alloy.
3.3 Cyclic stress-strain response for each joint
In the straddle fatigue test employed in this study, the
elastic deformation occurs in not only solder bump but also in
the other materials i.e. the Si chip and the FR-4 substrate.
Hence, the total displacement range observed during the test
may be described as follows.
dtotal ¼ dsolder(el) þ dsolder(pl)
þ dchip(el) þ dsub(el)
Ni Intensity ( arb. units )
Fig. 7 X-ray intensity scatter diagram of Sn-1.2 mass%Ag-0.5 mass%Cu0.05 mass%Cu.
ð1Þ
where dtotal is the total displacement applied which includes
the deformation for both side of the divided specimen,
dsolder(el) is the elastic deformation of solder, dsolder(pl) is
plastic deformation of solder, dchip(el) is the elastic
deformation of chip, and dsub(el) is elastic deformation of
substrate. The data plotted against the total displacement
range as shown in Fig. 9 do not indicate fatigue properties of
692
Y. Kariya, T. Hosoi, T. Kimura, S. Terashima and M. Tanaka
(a)
Cu
Ni
Ag
Cu, Ni, Ag
(b)
Cu
Ni
Ag
Cu, Ni, Ag
Fig. 8 EPMA elemental mapping image of Ag, Cu and Ni for solder bumps (a: Sn-1.0 mass%Ag-0.5 mass%Cu, b: Sn-1.2 mass%Ag0.5 mass%Cu-0.05 mass%Ni).
3
Load (N)
Total Displacement, ∆d t / µm
10
80
Continuous Cycling
T=298K
Sn-1.2mass%Ag-0.5mass%Cu-0.05mass%Ni
60
Sn-3mass%Ag-0.5mass%Cu
2
10
40
Sn-1mass%Ag-0.5mass%Cu
20
1
-20
10
-15
-10
-5
5
10
15
20
Displacement (µm)
0
10
-20
Sn-1mass%Ag-0.5mass%Cu
Sn-3mass%Ag-0.5mass%Cu
Sn-1.2mass%Ag-0.5mass%Cu-0.05mass%Ni
0
10
1
2
3
4
10
10
10
10
Number of Cycles to Failure, N f
5
-40
10
Fig. 9 Relationship between fatigue life and applied displacement range
for each flip chip joint.
-60
-80
Fig. 10 Hysteresis loops for each flip chip joint at dt ¼ 30 mm.
solder alloy itself. Figure 10 shows the hysteresis loops at the
total displacement range of dt ¼ 15 mm at 3cycles for each
flip-chip joint. The width of the loops at zero loads is the total
plastic deformation. The possibility that nonlinear load
displacement behavior could occur due to the FR-4 board
was checked by straining an undivided FR-4 board which did
not contain a Si chip and slit. The board was loaded to 720 N
with no nonlinear strain being produced. As the maximum
load throughout the fatigue test was less than 80 N, the
deformation of the board is still in the linear elastic region.
Therefore, the plastic deformation is allowed only in the
solder bumps in this study, so that the width of the loop at
zero loads reflects the amount of shear deformation of solder
bump.
The load range of the 1 mass%Ag alloy is much lower than
that of the 3 mass%Ag alloy, and the hysteresis loop width of
the alloy is lager than that of the 3 mass%Ag alloy as can be
seen Fig. 10. On the other hand, the hysteresis loop shape of
the Ni added low silver content alloy and 3 mass%Ag alloy
look exactly alike. An addition of Ni to low silver content
seems to increase the load range and to decrease the width of
the hysteresis loop. The result can be interpreted that the Ni
added low silver content alloy has superior strength compared with 1 mass%Ag alloy and the plastic deformation of
solder bump for the alloy is less than that of 1 mass%Ag
alloy, even though the same levels of the total displacement
range was applied.
Figure 11 shows the cyclic stress-strain curves for each
joint using the data at three cycles. If the strain and stress are
assumed to be uniform in the bump for convenience’ sake,
the shear stress and the shear strain are defined by following
equations.
¼ Pmax =Abump
¼
dp
4 Hbump
ð2Þ
ð3Þ
where is the shear stress, Pmax is the maximum load at three
cycles, Abump is the total cross area of solder bumps, is the
Low Cycle Fatigue Properties of Ni added Low Silver Content Sn-Ag-Cu Flip Chip Interconnects
693
-1
25
10
20
15
10
5
0
Sn-1mass%Ag-0.5mass5Cu
Sn-3mass%Ag-0.5mass%Cu
Sn-1.2mass%Ag-0.5mass%Cu-0.05mass%Ni
Shear Strain Range, ∆γ p
Shear Stress, τ / MPa
Continuous Cycling
T=298K
-2
10
-3
10
-4
10
0
1
2
3
Shear Strain, γ (%)
4
Sn-1mass%Ag-0.5mass%Cu
Sn-3mass%Ag-0.5mass%Cu
Sn-1.2mass%Ag-0.5mass%Cu-0.05mass%Ni
0
10
1
2
3
4
10
10
10
10
Number of Cycles to Failure, N f
5
10
Fig. 11 Cyclic stress-strain curves for each flip chip joint.
Fig. 12 Relationship between fatigue life and shear plastic strain range for
each flip chip joint.
shear strain, dp is the width of hysteresis loop, and Hbump is
the average height of solder bump. As shown in Fig. 11, the
cyclic flow stress of the 1 mass%Ag solder is much lower
than that of the 3 mass%Ag alloy. The cyclic stress-strain
response of the Ni added low silver content alloy is very
similar to the properties of the 3 mass%Ag alloy, even though
silver content of the alloy is almost equivalent to the content
of the 1 mass%Ag alloy. In general, fine microstructure
results in good mechanical strength and good ductility for an
alloy, and such mechanical properties are preferred to
enhance the fatigue resistance. The microstructure of the Ni
added alloy is obviously finer compared with the 1 mass%Ag
alloy, which leads to superior mechanical strength as shown
in Fig. 11. High strength solder bump induces relatively high
elastic deformation to the organic substrate that has low
elastic modulus as can be seen in Fig. 10. High elastic
deformation results in less plastic deformation of the system,
if the same level of the total displacement is applied to the
specimen. Therefore, the plastic deformation of the Ni added
solder bump is less compared with the 1 mass%Ag bump. As
low-cycle fatigue life depends on a level of applied plastic
deformation, the Ni added alloy exhibited long fatigue
endurance as shown in Fig. 9.
observed during the fatigue test in this study, and the plastic
deformation range consists of only the shear deformation of
solder bumps. Figure 12 shows the relationship between the
plastic strain range determined using the plastic deformation
range and fatigue life plotted on the double logarithm scale of
the Coffin-Manson plot. The plastic strain range is defined by
following equation.
3.4 Low cycle fatigue properties of each solder alloys
As shown in Fig. 10, the plastic deformation range of the
solder bumps can be obtained from the hysteresis loop
p ¼
ð4Þ
As can be seen in Fig. 12, all data obey the Coffin-Manson’s
relationship as follows.
p Nf ¼ C
ð5Þ
where p is the shear plastic strain range, Nf is the fatigue
life, is the exponent and C is the fatigue ductility. Fatigue
crack was initiated at upper corner of the bump and
propagated through the solder layer. The interfacial failure
was not observed for all alloys in this study. Therefore,
Fig. 12 indicates a fatigue characteristic of the solder bump
itself that does not include characteristics of any other
components such as the FR-4, and Si chip. Low cycle fatigue
properties of the Ni added alloy is equivalent to that of the
3%Ag alloy. Especially, the Ni added alloy exhibits superior
fatigue resistance compared with the 1%Ag alloy in low
strain regime. The Coffin-Manson relation for each alloy are:
For Sn-1.0 mass%Ag-0.5 mass%Cu:
For Sn-3.0 mass%Ag-0.5 mass%Cu:
For Sn-1.2 mass%Ag-0.5 mass%Cu-0.05 mass%Ni:
The fatigue ductility coefficient of the 1 mass%Ag alloy is
much higher than that of other alloys, which means ductility
of the 1 mass%Ag alloy is superior compared with other
alloys. The fatigue ductility exponent is about 0.5 for the
3 mass%Ag alloy and the Ni added alloy, and is about 0.7 for
the 1 mass%Ag alloy. In general, the fatigue ductility
exponent correlates with the cyclic strain hardening expo-
dp
2 Hbump
p Nf 0:68 ¼ 0:41
p Nf 0:55 ¼ 0:26
p Nf 0:53 ¼ 0:22
ð6Þ
ð8Þ
ð9Þ
nent, and high cyclic strain hardening exponent tends to lead
to low fatigue ductility exponent for metals.12) Figure 13
shows fatigue ductility exponent as a function of cyclic strain
hardening exponent. The cyclic strain hardening exponent is
defined as follows:
¼ K 0 "n
0
ð10Þ
694
Y. Kariya, T. Hosoi, T. Kimura, S. Terashima and M. Tanaka
of mechanical reliability.
Fatigue Ductility Exponent
1
4.
Conclusions
0.8
0.6
0.4
Sn-1mass%Ag-0.5mas%Cu
Sn-3mass%Ag-0.5mas%Cu
Sn-1.2mass%Ag-0.5mas%Cu-0.05mass%Ni
0.2
0
0.1
0.2
0.3
0.4
Cyclic Strain Hardening Exponent
0.5
Fig. 13 Fatigue ductility exponent as a function of cyclic strain hardening
exponent for each joint.
n0 is the cyclic strain hardening exponent and K 0 is constant.
As can be seen in Fig. 13, higher cyclic strain hardening
exponent exhibits lower strain fatigue ductility exponent. The
cyclic strain hardening exponent of the Ni added alloy is
higher than that of the 1 mass%Ag, which leads to low fatigue
ductility exponent. Low fatigue ductility exponent and high
fatigue ductility coefficient are necessary to have excellent
fatigue endurance. In other words, good strength and good
ductility are required for solder alloys which will undergo
low cycle fatigue. The 3 mass%Ag alloy has good low cycle
fatigue resistance, since the strength and the ductility of the
alloy are well balanced. The 1 mass%Ag alloy has high
ductility which results in good fatigue performance in high
strain regime, whereas the cyclic strain hardening exponent
of the alloy is much lower than that of the 3 mass%Ag alloy
that leads somewhat poor fatigue resistance in low strain
regime. From the results, low silver content alloy seems
unsuitable for a candidate of lead-free solder bump. However, the Ni added alloy has good strength and good ductility
similar to the 3 mass%Ag alloy, even though silver content is
much lower than that of near eutectic Sn-Ag-Cu alloy.
Therefore, the Ni added low silver content alloy will become
a candidate alloy for flip-chip application form the view point
The low cycle shear fatigue life of Sn-1.2 mass%Ag0.5 mass%Cu-0.05 mass%Ni was investigated to develop low
silver content Sn-Ag-Cu alloy for flip-chip application. The
low cycle fatigue properties of the Ni added alloy was
equivalent to that of the 3 mass%Ag alloy, even though silver
content is much lower than that of near eutectic Sn-Ag-Cu
alloy. The microstructure of the Ni added alloy was
obviously finer compared with the 1 mass%Ag alloy, which
generated superior mechanical strength with good ductility
and high cyclic strain hardening exponents similar to the
3 mass%Ag alloy. The high cyclic strain hardening exponents with good ductility of the alloy leads to good low cycle
fatigue endurance of the Ni added alloy similar to the
3 mass%Ag. The Ni added low silver content alloy will
become a candidate alloy for flip-chip application form the
view point of mechanical reliability.
REFERENCES
1) P. T. Vianco and F. R. Frear: JOM 7 (1993) 14–19.
2) S. K. Kang and A. K. Sarkhel: J. Electron. Mater. 23 (1994) 701–708.
3) D. R. Frear, J. W. Jang, J. K. Lin and C. Zhang: JOM, June (2001) 28–
32.
4) C. M. Miller, I. E. Anderso and J. F. Smith: J. Electron. Mater. 30
(1994) 595–601.
5) I. E. Anderson J. C. Foley, B. A. Cook, J. Harringa, R. L. Terpstra and
O. Unal: J. Electron. Mater. 30 (2001) 1050–1059.
6) Q. Guo, E. C. Cutiongco, L. M. Keer and M. E. Fine: ASME J.
Electronic Packaging 114 (1991) 145–151.
7) Q. Guo, S. Choi, K. N. Subramanian, T. R. Bieler, J. P. Lucase, A.
Achari and M. Paruchuri: Mater. Sci. Eng. A351 (2003) 190–199.
8) Q. Yu, M. Shiratori, S. Kaneko, T. Ishihara and S. Wang: J. JSME 64619A (1998) 558–563.
9) Y. Kariya, T. Hosoi, S. Terashima, M. Tanaka and M. Ostuka: J. of
Electron. Mater., in print.
10) H. D. Solomon: IEEE Trans. Comp. Hybrids, Manuf. Technol. 12
(1989) 473–479.
11) H. D. Solomon: ASME Journal of Electronic Packaging 144 (1992)
161–167.
12) T. Kimura, K. Nishida, S. Tanuma and H. Yamada: Jounal of Surface
Analysis 10 (2003) 203–211.
13) J. D. Morrow: ASTM, STP-378 (1965) 45–52.