Size Effects On IMC Growth of Cu

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

Available online at www.sciencedirect.com

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

Original Article

Size effects on IMC growth of Cu/Ni/Sn-3.5Ag


microbump joints during isothermal aging and
prediction of shear strength using ANN

Chu Tang, Zhuo Chen*, Mingang Fang, Xiaoyu Xiao, Gui Chen,
Wenhui Zhu**
State Key Laboratory of High Performance Complex Manufacturing, College of Mechanical and Electrical Engineering,
Central South University, Changsha, China

article info abstract

Article history: Significant downsizing trend is currently directing the advanced microelectronic packaging
Received 28 October 2021 industry, which poses new challenges to the manufacturing process and reliability of high-
Accepted 17 March 2022 density chip interconnections. In this paper, to investigate the size effect of microbump
Available online 22 March 2022 joints on the interfacial structure and bonding strength, the chip-to-chip bonding joints
fabricated by microbumps of same Cu/Ni/Sn-3.5Ag construction but of various sizes were
Keywords: experimentally examined. After thermocompression bonding process to fabricate the chip-
Interfacial reactions to-chip microbump interconnections, the intermetallic compound (IMC) growth on the Ni/
Shear strength Sn-3.5Ag interface during 150  C isothermal aging was compared for joints of 100, 40 and
Size effect 25 mm in diameter as well as 35, 25 and 15 mm in solder layer height, using field emission
Intermetallics electron microscopy and electron probe microanalysis. A clear size effect on the thickness
Artificial neural network and morphology of Ni3Sn4 IMC was observed, showing thicker IMC in joints of smaller
Microbump interconnection diameter and lower height. The Ni concentration distribution in the solder layer was
verified to be dependent on the joint size, with possible reasons being enhanced sidewall
diffusion and grain boundary diffusion through IMC grains in small-diametered joints, as
well as shorter diffusion path in lower joints. The shear strength of chip-to-chip bonding
vehicles was also found to be size-dependent. In addition, artificial neural network (ANN)
approach was applied to establish the numerical correlation of structural factors and
bonding strength. It was found that a properly constructed ANN model can successfully
predict the influence of the complex combinations of geometric and interfacial factors with
highly accurate output, providing an effective tool for design and optimization of micro-
bump interconnection structures and processes.
© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

* Corresponding author.
** Corresponding author.
E-mail addresses: zhuochen@csu.edu.cn (Z. Chen), zhuwenhui@csu.edu.cn (W. Zhu).
https://doi.org/10.1016/j.jmrt.2022.03.110
2238-7854/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).
1866 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

that the smaller the height/radius ratio (h/r) of the solder joint,
1. Introduction the higher the strength of the solder joint. Another factor that
has a significant impact on the strength of solder joints is the
Facing the urgent need to improve the performance of elec- interfacial IMC thickness. Aging [14e17] and multiple reflows
tronic products and chip process scaling, three-dimensional [18] increase the IMC thickness and thus reduce the strength
integrated circuit (3D IC) technology, which significantly pro- of the solder joint. The growth of IMC is also strongly affected
motes the off-chip Input/Output (I/O) performance by using by the size of the solder joints. Reducing the solder joint height
stacked silicon chips, has gained interests rapidly [1,2]. [19,20] and solder joint diameter [21e23] leads to faster growth
Microbump technology is the key for mechanical and elec- of IMC. It has also been studied that Cu/SAC305 (sandwich) in
trical connections between stacked chips. Today, the size of liquid thermal aging (300  C) found that the interfacial IMC
micro interconnection joints for stacked chips is approaching thickness decreased with decreasing solder joint height, while
10 mm level [1,3]. As a consequence, the amount of solder in a solid-state thermal aging (210  C) found that the interfacial
micro interconnect is very small and is prone to be fully IMC thickness increased with decreasing solder joint height
transformed to brittle intermetallic compounds (IMCs) during [24]. Magnien et al. [25] studied Cu/Sn3.5Ag0.75Cu thermal
bonding process and final product service life. Excessive aging (150  C) interface found that the IMC growth rate
amounts of brittle IMC are thermally mismatched to the sol- decreased with decreasing solder joint height. Table 1 sum-
der and the substrate, predisposing the solder joint to weak- marizes the previously reported effects of solder joint size on
ening and fatigue failure [4]. Industrial applications typically the interfacial IMC thickness and shear strength. It can be
use a Ni barrier layer inserted between Cu metallization and seen that previous studies [11e13,24,26] have mainly focused
Sn-based solder to retard the IMC growth in a joint [5]. on the size effect in large size ball grid array (BGA) solder joints
In order to study the growth characteristics of IMC and the or sandwich structures, and the impact of microbump joint
method of inhibiting its overgrowth, many studies [2,6,7] have (100 mm) size on shear strength as well as IMC growth are not
been conducted previously to explore new Sn-based alloy explored deeply enough and not clear, which can only be
solders. The basic systems of Sn-Cu, Sn-Ag and Sn-Ag-Cu clarified by the side-by-side comparison of joints with
have become the front-runners to replace Sn-Pb solders different size, under the same fabrication process and aging
[2,6]. Among them, the Sn-Ag solder is considered most suit- condition.
able for the small sized and Cu pillar-based joints, since the Developing a physical model to accurately predict the
Sn-Ag eutectic alloy can be easily prepared through electro- mechanical properties and reliability of microbump joints is
plating after the Cu pillar, and Ag was found effective in difficult because the strength of microbump joints is influ-
slowing the IMC growth [7]. enced by multiple factors. Some researchers [26,27] have
Many studies [8e10] have shown that solder joint size is a attempted to develop predictive models based on experi-
sensitive factor affecting the joint strength. For relatively large mental data, but these models consider relatively few influ-
joints, the shear strength exhibited an increase followed by a encing factors. Artificial neural network (ANN) modeling
decrease when the solder joint height was reduced from 500 to techniques are a promising artificial intelligence tool that has
100 mm [8,9]. In addition, a study found that the strength of been widely used to solve various scientific problems with
solder joints continued to increase as the diameter of solder classification and prediction capabilities [28,29]. ANN
joints decreased from 600 to 200 mm, exhibiting a size effect modeling techniques have applications in reflow soldering
[10]. Previous studies have concluded from the theoretical processes, the mechanical strength of solder joints and IMC
analysis [11,12] and experimental combined simulation [13] layer thickness of Sn-5Sb-xCNT/Cu composite solder joints

Table 1 e Size effect of the microsolder joint on shear strength and IMCs thickness.
Ref Sample size range Joint/Reaction type Shear strength with
reduced solder joint sizea
Qin et al. [11] 100e500 mm (height) Cu/SAC305(BGA) [þY(250 CReflow)
Li et al. [12]
Tian et al. [13] 200e600 mm (diameter) Cu/SAC305(BGA) [(150 CAging)
Qin et al. [16] 300 mm (diameter), 25e100 mm (height) Ni/SAC305(sandwich) [(250 CReflow)

IMCs thickness with


reduced solder joint sizea

Wang et al. [22] 1000 mm (diameter), 10e100 mm (height) Cu/Sn(sandwich) [(260 CReflow)
Zhu et al. [23] 10e50 mm (height) Cu/Sn(sandwich) [(160 CAging)
Wang et al. [27] 50e2000 mm (height) Cu/SAC305(sandwich) Y(300 CAging)
[(210 CAging)
Magnien et al. [28] 50e800 mm (height) Cu/Sn3.5Ag0.75Cu (sandwich) Y(150 CAging)
Huang et al. [24] 200e500 mm (diameter) Cu/Sn-xAg-yCu(BGA) [(250 CReflow)
Wang et al. [25] 6e11 mm (diameter) Cu/Ni/Sn-3.5Ag(CPB) [(150 CAging)
Wong et al. [26] 40e250 mm (diameter) Ni/Sn-3.8Ag-0.7Cu (BGA) [(150 CAging)
a
Increase ¼ [,decrease ¼ Y, Increase first and then decrease ¼ [þY.
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1867

[30e32]. However, a prediction model is lacking for the me- geometric designs of different size Cu/Ni/Sn-3.5Ag pillar
chanical strength of Cu/Ni/Sn-3.5Ag microbump joints of bumps. For chips with 100 mm diameter microbumps, the Sn-
different sizes (100 mm) after aging. Similarly, the growth 3.5Ag layers heights of both the top and bottom chips are
prediction model for IMC rarely considers the effect of size available in 15 and 25 mm, and the desired height of the
pffiffiffiffiffiffi
and mainly uses the diffusion equation (x ¼ x0 þ Dt) to microbumps can be obtained by controlling the gap during the
calculate it [33e37]. thermocompression bonding process. The Sn-3.5Ag layer is
In this paper, we experimentally characterized the IMC reflowed into a ball shape after electroplating. The thickness
layer and shear strength of isothermally aged Cu/Ni/Sn-3.5Ag of both chips is the same, 0.5 mm. For the chips containing 25
microsolder joints of same configuration but different in size. and 40 mm diameter microbumps, the Sn-3.5Ag layers height
Moreover, the Ni concentration distribution in solder joints was 15 mm for both the top and bottom chips. Figure 2(a)
was measured and the growth mechanism of IMC affected by shows the cross-sectional schematic of the microbump joints.
the joint size during isothermal aging was explored. Finally, to Figure 2(b) and (c) show the fabrication of 100 mm diameter Cu/
extend the scope of computational intelligence-based Ni/Sn-3.5Ag pillar bumps for the top and bottom chip. Sam-
methods for developing prediction models in engineering ples used in this article were provided by National Center for
applications, ANN models were successfully deployed to Advanced Packaging Co., Ltd. (NCAP China), which were
predict the mechanical properties of microbump joints by fabricated after a series of standard manufacturing processes
back-propagation convergence techniques. To verify the re- including but not limited to photolithography, electroplating
sults of the developed model, confirmatory experiments were and etching.
conducted.
2.2. Experimental setup and characterization

2. Material and methods The chip-on-chip bonding was carried out through thermo-
compression bonding process by an Athlete CB-600 flip chip
2.1. Materials bonding machine. Five microbump joint sizes were designed
for the test, 100 mm in diameter and 15, 25 and 35 mm in height.
The test vehicles consist of top and bottom chip with Cu/Ni/ 25 and 40 mm diameter samples were also prepared for the
Sn-3.5Ag microbumps and daisy chain circuits for electrical 15 mm joint height, respectively. The substrate is heated to a
test. Figure 1 shows the design drawing of bottom chips with constant temperature of 150  C during the entire bonding pro-
different size microbumps. All chips were fabricated through cess. The bonding parameters were optimized and finally
a standard bumping process patterned by photolithography, determined to be a bonding temperature of 325  C, a bonding
including first lithography for Cu trace etching, second pressure of 2 N, and a bonding time maintained between 6 and
lithography for defining the microbump area, and sequential 10 s. The bonding time is slightly longer for samples with higher
electroplating of Cu, Ni, and Sn-3.5Ag in the photoresist- solder joint heights. Before bonding, the microbumps on the
defined holes. The Ni barrier layer between the Cu pillar and top and bottom chips were coated with flux. The height of the
the Sn-3.5Ag layers was 2 mm for all chips. Table 2 shows the solder joint is controlled by adjusting the compression distance

Fig. 1 e Graphics of chip design pattern with different size microbumps ((a) 25 mm diameter and (b) 100 mm diameter) on the
bottom chip.
1868 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

Table 2 e Geometric designs of Cu/Ni/Sn-3.5Ag pillar bumps.


Layer Pillar structure Bump diameter (mm) Bump height (mm) Cu/Ni/Sn-3.5Ag Thickness (mm)
Bump Top chip Cu/Ni/Sn-3.5Ag 25 15 23/2/15
40 15 30/2/15
100 15 20/2/15
100 25 20/2/15
100 35 20/2/25
Bottom chip Cu/Ni/Sn-3.5Ag 25 15 8/2/15
40 15 8/2/15
100 15 15/2/15
100 25 15/2/25
100 35 15/2/25

Fig. 2 e Schematic diagram (a) SEM micrographs, (b) top chip, (c) bottom chip of Cu/Ni/Sn pillar bumps.

of the welding head during the thermocompression bonding. tests. The concentration values were obtained by averaging
After bonding, isothermal aging tests were performed in a non- four data points around the measurement locations. The Ni
vacuum oven at 150  C for 0, 100, 200 and 400 h. Die shear tests concentration variation curves were also plotted for micro-
were performed using a global bond test (Dage 4000). As shown bump joints of different sizes. The image processing software
in Fig. 3, the conditions for the low-speed shear test included a Photoshop was also used to measure the IMC thickness. The
height of 30 mm and a shear rate of 0.1 mm/s. The shear test was average thickness of IMC can be obtained by dividing the total
performed at least three times and the strength values were area of IMC by the length of the observed area.
averaged. The specimens were fixed with epoxy after bonding
and then sequentially ground to the maximum cross-section of 2.3. Artificial neural network (ANN) modelling
the microbump joints with 200, 800, 2500 and 4000 # SiC
sandpaper, and then polished with Al2O3 suspension and SiO2 ANN is a nonlinear statistical data modeling tool that is well
suspension. Finally, sample cross-sections and fracture sur- suited to solve nonlinear stochastic problems. In solving com-
faces were analyzed by field emission electron microscopy (FE- plex problems, researchers value its ability to learn and
SEM) and energy dispersive X-ray spectrometry (EDS) attached generalize reasonable outputs to new inputs that are not
to the microscope. The concentration distribution of Ni atoms encountered during training [38]. They did this by constructing
at different locations in the solder layer was scanned by Elec- input, hidden, and output layers, which were interconnected by
tron probe microanalysis (EPMA) after 400 h of isothermal aging neurons, and by learning and adopting information from the
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1869

Fig. 3 e Schematic diagram of the die shear test.

examples. Seventy percent of the sample parameters are used N 


P 2
to train the neural network, and the other 30 percent of the data yipre  yitag
are separated out and used to test and validate the developed R2 ¼ 1  i¼1N  2 (1)
P i
model. In this way, we take care to cover the entire range of ypre  y
i¼1
experiments to develop and test the neural network models.
The ANN model developed in this work uses the Levenberg vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uN  2
uP i
u
ti¼1 ypre  ytag
Marquardt back-propagation (LMBP) algorithm as the training i

algorithm because it is robust and has the fastest convergence RMSE ¼ (2)
N
rate relative to other training algorithms, which is essential for
effective training [39e41]. Previous studies [42,43] have pro- where yipre is the predicted output, yitag is the experimental
posed many methods to explore the optimal number of hidden (target) output, y is the average of the experimental output
layers and neurons, but they are not applicable to all research and N is the total number of data observation.
subjects. In this paper, a supervised trial-and-error method was
used to find that two hidden layers gave the best predictions.
Therefore, the ANN model in this paper determines the number 3. Results and discussion
of hidden layers is two, thus further exploring the optimal
number of neurons. In addition, the network performance and 3.1. IMC morphological analysis and growth kinetics
the degree of fit of the various ANN topologies studied in this
work were analyzed by using two statistical parameters, the Figure 4 shows SEM micrographs of the fracture surfaces, Figs.
coefficient of determination (R2) and root mean square error 5e7 show the cross-sectional microstructure of Cu/Ni/Sn-
(RMSE) [30,43]. 3.5Ag sample after isothermal aging at 150  C. As shown in

Fig. 4 e SEM micrographs of the fracture surfaces of 100 mm diameter samples after isothermal aging at 150  C for (a), (e) 0 h,
(b), (f) 100 h, (c), (g) 200 h, (d), (h) 400 h (h ¼ 15).
1870 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

Fig. 5 e Cross-sectional microstructure of Cu/Ni/Sn-3.5Ag sample after isothermal aging at 150  C for (a), (a1) 0 h, (b), (b1)
100 h, (c), (c1) 200 h, (d), (d1) 400 h (d ¼ 100 mm, h ¼ 15 mm).

Figs. 4(e) and 5(a1), a thin layer of scallop-shaped IMC with a Magnien et al. [25] and Kim et al. [45], while studying the
thickness of about 0.72 mm grew at the Ni/Sn-3.5Ag interface morphology of Cu/Sn-Ag-Cu/Cu microbump joints of different
after bonding for a microbump joint of 100 mm diameter and sizes after isothermal aging, found that the gradual coarsening
15 mm height. The EDS analysis showed that the IMC was of b-Sn grains and Ag3Sn particles was observed with
Ni3Sn4. Some Ag3Sn particles were also found to precipitate in increasing solder gap size (volume), due to slower cooling rate
the solder near the IMC. When the specimen was isothermally in joints with larger solder volume. Figures 8 and 9 show cross-
aged at 150  C, the Ni3Sn4 IMC grew symmetrically at the sectional microstructure of Cu/Ni/Sn-3.5Ag sample after
upper and lower interfaces of the microbump joints (Fig. 5(b)). isothermal aging at 150  C, Figs. 10 and 11 show SEM micro-
With increasing aging time, the morphology of Ni3Sn4 IMC graphs of the fracture surfaces. As shown in Figs. 8(a1) and
changes from scallop type to layer type. The subsequent 10(e), the microbump joints shrink from 100 to 40 mm, and
microbump joints fracture morphology analysis depict the after thermocompression bonding, a thin needle-like IMC layer
growth process of IMC grains with aging time. After aging for is formed at the Ni/Sn-3.5Ag interface, which is more slender
400 h, the thickness of Ni3Sn4 is 2.63 mm, which is 1.91 mm than 100 mm in diameter, with a layer thickness of about
higher than the original thickness. The Ag3Sn particles were 0.75 mm. This may be due to the fact that smaller diameter
observed attached to the Ni3Sn4 surface, which was preferred microbump joints dissipate heat more quickly during the
site for the heterogeneous nucleation Ag3Sn [44]. The IMC cooling phase of thermocompression bonded joints, with
growth in each type of microbump joints was slowing down greater subcooling and easier IMC nucleation, resulting in the
with increasing aging time, as shown in Figs. 6 and 7. formation of more and finer grains. After 100 h aging, the
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1871

Fig. 6 e Cross-sectional microstructure of Cu/Ni/Sn-3.5Ag sample after isothermal aging at 150  C for (a), (a1) 0 h, (b), (b1)
100 h, (c), (c1) 200 h, (d), (d1) 400 h (d ¼ 100 mm, h ¼ 25 mm).

spaces between the needle-like IMCs are filled by the repreci- adjacent Ni3Sn4 grains have formed holes between them due
pitated IMCs, causing the IMCs to grow from needle-like to to growth shrinkage, which is one of the reasons for the
shell-like. After aging for 200 h, Ni, Sn atoms diffusing through reduced strength of the microbump joints, as shown in Figs.
the IMC layers become less easy, the growth rate slows down, 4(g), (h), 10(g) and 10(h) [48,49].
and the shell-like IMC becomes a flatter layer. The thickness of Size effects play a role in aging and reflowing microbump
Ni3Sn4 is 2.47 mm, which is 1.72 mm higher than the original joints [21,46,50]. We found that the IMC thicknesses of the aged
thickness. This phenomenon is also seen in 25 mm diameter microbump joints were also significantly different for different
microbump joints, as shown in Figs. 9(b) and 11(e). Figure 12 sizes. After isothermal aging at 150  C, the IMC thickness of the
shows a plot of IMC thickness versus aging time for micro- microbump joints increases with the reduction of the joint size,
bump interfaces with diameters of 25, 40 and 100 mm. as shown in Fig. 12. Among the microbump joints with a
Furthermore, the small Ag3Sn grains were observed after diameter of 100 mm, the thickness of the IMC is 1.40 mm for the
bonding and had the effect of pinning at the IMC/Sn interface microbump joints with a height of 35 mm aged for 100 h. The
[46] and filling the voids [47], thus increasing the solder joint IMC thickness of the 25 mm height microbump joints is 1.60 mm,
shear strength. However, finely dispersed Ag3Sn particles which is 14.3% higher than that of the 35 mm height microbump
agglomerated into a few large grains during aging, as shown in joints. When the height of the microbump joints is further
Figs. 7 and 8. As the aging time increases, the pits in the reduced to 15 mm, the thickness of the IMC is 1.76 mm, which is
microbump joint sections are becoming less and deeper, and 10% higher than that of the microbump joints with a height of
the Ni3Sn4 grains are becoming coarser. Also, some of the 25 mm. When the height of the microbump joints is constant
1872 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

Fig. 7 e Cross-sectional microstructure of Cu/Ni/Sn-3.5Ag sample after isothermal aging at 150  C for (a), (a1) 0 h, (b), (b1)
100 h, (c), (c1) 200 h, (d), (d1) 400 h (d ¼ 100 mm, h ¼ 35 mm).

and the diameter of the microbump joints is reduced, the IMC pffiffiffiffiffiffi
x ¼ x0 þ Dt (3)
thickness of the microbump joints is increasing. For a micro-
bump joints with a diameter of 25 mm, the thickness of the IMC where x is the total thickness of the reaction layer at time t and
after 100 h aging is 1.87 mm, which is 6.3% higher than that of a x0 is the initial thickness, d is the growth rate constant. The
100 mm diameter. This shows that the size effect has a greater value of D at a specific temperature can be obtained by
influence on the height direction of the microbump joints than multivariate linear regression analysis.
in the diameter direction. A similar effect is manifested As seen in Fig. 13, the linear slope is the square root of the
throughout the aging process. growth rate constant D (h is the solder height and d is the
He et al. [36] and Alam et al. [37], while studying the growth microbump diameter). The growth constants for different size
kinetics of Ni3Sn4 in BGA microbump joints at the Sn-3.5Ag microbump joints ranged from 0.0066 to 0.0132, showing a size
solder/Ni interface, found that the initial growth of Ni3Sn4 is effect. The growth rate of IMCs increases with decreasing
greatly affected by the grain boundary diffusion, but after the height in microbump joints of 100 mm diameter with heights of
aging time increases, the grains become coarser and the 15, 25 and 35 mm. The equations for the growth of IMC thick-
number of grain boundaries decreases, so the volume diffu- ness for different solder heights in a 100 mm diameter micro-
sion gradually becomes dominant. Therefore, in order to bump are given in (6), (7) and (8).
study the kinetics of the interfacial reaction layer growth with Figure 13 also shows that the interfacial response increases
time at a particular temperature, it is appropriate to use the and the growth constant D of IMC increases as the diameter of
following empirical power law relationship. Cu/Ni/Sn-3.5Ag microbump decreases from 100 to 40 and then
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1873

Fig. 8 e Cross-sectional microstructure of Cu/Ni/Sn-3.5Ag sample after isothermal aging at 150  C for (a), (a1) 0 h, (b), (b1)
100 h, (c), (c1) 200 h, (d), (d1) 400 h (d ¼ 40 mm, h ¼ 15 mm).

to 25 mm in the isothermal aging test. The growth of IMC is 3.2. Ni concentration distribution in microbump joints
promoted by the increased grain boundary diffusion due to
the finer initial grain size of 25 mm diameter microbump joints To further clarify the direct cause of size effect on the IMC
during thermocompression bonding. The equations for the growth, the elemental concentration distribution in the
growth of IMC thickness in 25 mm diameter and 40 mm diam- microbump joints was analyzed. As shown in Fig. 14(a), Cu/Ni/
eter microbump are given in (4), (5). Sn-3.5Ag microbump joints (d ¼ 25 mm) formed a very thick
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi IMC layer at the Ni/Sn-3.5Ag interface. As indicated in
x1 ¼ 0:76 þ 0:0132t (4) Fig. 14(b), EPMA analysis shows that the Ni atom concentra-
tion in the solder gradually decreases from the IMC/Sn inter-
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x2 ¼ 0:75 þ 0:0121t (5) face to the middle of the solder.
As illustrated in Fig. 15, the Ni atom concentration de-
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x3 ¼ 0:72 þ 0:0104t (6) creases with increasing microbump joints height in micro-
bump joints of 100 mm diameter with three values of
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 0.907 ± 0.12, 0.510 ± 0.06, and 0.419 ± 0.08 wt.%, respectively.
x4 ¼ 0:67 þ 0:0089t (7)
This result is consistent with of Dybkov's diffusion kinetic
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi model: the lower the solder joint height, the faster the Ni
x5 ¼ 0:54 þ 0:0066t (8) barrier dissolves and the higher the Ni concentration in the Sn
1874 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

Fig. 9 e Cross-sectional microstructure of Cu/Ni/Sn-3.5Ag sample after isothermal aging at 150  C for (a), (b) 0 h, (c), (d) 100 h,
(e), (f) 200 h, (g), (h) 400 h (d ¼ 25 mm, h ¼ 15 mm).
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1875

Fig. 10 e SEM micrographs of the fracture surfaces of 40 mm diameter samples after isothermal aging at 150  C for (a), (e) 0 h,
(b), (f) 100 h, (c), (g) 200 h, (d), (h) 400 h (h ¼ 15).

Fig. 11 e SEM micrographs of the fracture surfaces of 25 mm diameter samples after isothermal aging at 150  C for (a), (e) 0 h,
(b), (f) 100 h, (c), (g) 200 h, (d), (h) 400 h (h ¼ 15).

solder [51]. Furthermore, when the height of the microbump microbump joints, the higher the concentration of Ni atoms.
joints was kept constant and the diameter was reduced to 40 To explain the Ni atom concentration distribution in different
and 25 mm, we found that the average Ni concentration size microbump joints, we suspected that on one hand, as the
increased from 0.907 ± 0.12 to 1.008 ± 0.18 and aging time increases, the IMC thickens, so that the volume
1.060 ± 0.15 wt.%, respectively. It can be deduced that the high diffusion path and the time required for Ni atoms to diffuse
Ni atom concentration in smaller sized microbump joints has through the IMC layer increases. Meanwhile, Ni atoms could
a positive effect on the oversaturation of Ni in the Sn solder. also diffuse from the sidewall surface through the surface
Figure 16 depicts the effect of diffusion on Ni atoms in diffusion, bypass the volume diffusion path in IMC and diffuse
microbump joints of different sizes. The smaller the size of the into the solder layer from the side to react with Sn, which has
1876 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

become an important driving force for the growth of IMC [52].


This proportion of surface diffusion increases when the
microbump joints diameter is reduced, so promoting the
growth of IMC [22]. On the other hand, as shown in Figs. 4(e)
and 11(e), during bonding, the smaller diameter microbump
joints dissipate heat faster and have higher subcooling, so
more IMC nucleation and finer IMC grains are produced. The
fine IMC grains increase the grain boundary diffusion of Ni
atoms, which enhances the Ni atom concentration in the
solder and promotes IMC growth. The grain boundary diffu-
sion is faster than the bulk diffusion and the IMC usually
grows along the grain boundary [33,53,54]. Another possible
reason is the thermal gradient as an additional driving force of
diffusion, or generally known as thermomigration, which is
higher in small diameter microbump joints [23]. Similarly, for
low height microbump joints, on the one hand, the low vol-
ume of solder and the fast heat dissipation during the ther-
Fig. 12 e Variation of average thickness of Ni3Sn4 IMC as a mocompression bonding promote solideliquid diffusion [20].
function of time. On the other hand Ni atoms have reduced diffusion paths and
tend to aggregate at the IMC/Sn-3.5Ag interface, both
increasing the Ni atom concentration and accelerating the
IMC generation.
During the aging process, the Ni barrier layer inhibits the
interdiffusion of Cu and Sn atoms significantly. The compo-
sition of the solder joints after isothermal aging at 150  C for 0,
100, 200 and 400 h was analyzed by EDS line scan. According to
Fig. 17, the Ni barrier layer thins with increasing aging time
but is always present. Meanwhile, as shown in Fig. 18, the EDS
element mapping of the microbump joints after isothermal
aging for 400 h shows that the Ni barrier layer remains
continuous and is not depleted or penetrated. In addition, an
EDS analysis of the specific composition of the IMC was per-
formed. The atomic ratios of the different compositions show
that the IMC is (NixCu1-x)3Sn4, as shown in Table 3. The con-
tent of Cu atoms in the IMC was 6.31 at.% after bonding and
increased to 9.10 at.% after 100 h of aging. The Ag content was
less than 0.1 at.% and was not included in the statistics.
Fig. 13 e Variation in the IMC thickness with the square Therefore, the strength of microbump joints in this paper is
root of the aging time. mainly influenced by the Ni3Sn4-structured IMC.

Fig. 14 e (a) Cross-section SEM images of the Cu/Ni/Sn-3.5Ag layers joint after aging 400 h and (b) concentration gradient of
Ni between the solder layers.
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1877

height 25 and 35 mm are close, larger than that with a height of


35 mm by 5.8 Mpa.
After 100 hours of aging, the strength of all microbump
joints decreased significantly. In the microbump joints with a
diameter of 100 mm, the shear strength of the joint with a
height of 15 mm dropped from 71.3 to 42.0 Mpa, and the shear
strength loss was 41.1%. The shear strength of microbump
joints of different sizes has a similar trend after aging. How-
ever, after 400 hours of aging, only the 100 mm diameter
microbump joints with a height of 15 mm have the largest
decrease in strength, 15.2 Mpa.
It can also be seen from Fig. 19 that when the microbump
joints height remains unchanged and the diameter is reduced
from 100 to 25 mm, the strength of the microbump joints after
bonding decreased from 71.3 to 26.0 Mpa, 63.5% shear strength
loss. After aging 400 h, the 25 mm diameter microbump joints
strength was slowly decreased to 22.8 MPa. It is close to the
Fig. 15 e Relationship between the concentration of Ni and strength (25.7 Mpa) of the microbump joints with a diameter
different size of bumps after aging 400 h. of 100 mm, both at a lower level.
Many previous studies [8,12,13] have shown that the solder
3.3. Shear strength analysis volume/substrate area ratio (V/A) or the microbump joints
height/substrate radius ratio (h/r) has a significant effect on
Figure 19 shows the shear strength of thermocompression the interfacial IMC growth and microbump joints strength.
bonded and aged microbump joints of different sizes studied Wong et al. [23] found that the solder ball strength decreases
in this paper. When the microbump joints size decreases in with decreasing pad diameter in the push ball test of Cu/Ni/
the height dimension, it shows a size effect, an increase in SAC solder balls. It is generally believed that the thinner the
shear strength. When decreasing the diameter dimension, the microbump joints are during aging, the faster the IMC growth
shear strength decreases. The shear strength of a 100 mm and the faster the microbump joint's strength decreases. In
diameter microbump joints with a height of 15 mm after this paper, the experimental analysis concluded that the ge-
bonding is 71.3 MPa, which is about 1.82 times higher than ometry of the microbump joints plays a dominant role in the
that with a height of 25 mm, 39.2 MPa. The shear strengths of strength of the microbump joints after bonding. The h/r value

Fig. 16 e Size effect of microsolder joints on Ni diffusion for (a) d¼ 100 mm, h ¼ 35 mm, (b) d ¼ 100 mm, h ¼ 25 mm, (c) d ¼ 100
mm, h ¼ 15 mm, (d) d ¼ 40 mm, h ¼ 15 mm, (e) d ¼ 25 mm, h ¼ 15 mm.
1878 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

Fig. 17 e (a), (c), (e) and (g) show SEM images of the Cu/Ni/Sn-3.5Ag interfaces, respectively; (b), (d), (f) and (h) show line-scan
results of elemental compositions at the Cu/Ni/Sn-3.5Ag interfaces, respectively.(d ¼ 40 mm, h ¼ 15 mm).
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1879

Fig. 18 e Element Mapping of Solder Joint after isothermal aging at 150  C for 400 h: Electron back scattered diffraction (BSE)
image(a) and EDS mappings of Ni(b), Cu(c), Sn(d) and mixed elements distributions (e).(d ¼ 40 mm, h ¼ 15 mm).

is smaller the higher the strength. However, after isothermal ductileebrittle combination, while in the others fractures all
aging at 150  C, the IMC increases linearly with the 0.5 th occurred in the soldered region and had dimple structures at
power of time. After aging for 100 h, the thickness of IMC the fracture site, indicating ductile fracture characteristics.
increased by about 1 mm for different sizes of microbump After 100 h of aging, in the 100 mm diameter joints, with the
joints. The weakening effect of IMC on the strength of height decreasing from 35 to 25 and then to 15 mm, the pits
microbump joints is obvious. The shear strength of the containing Ni3Sn4 particles are increasing, the fracture form is
microbump joints still shows a size effect. However, after ductileebrittle combination, but the brittle fracture constituent
400 h of aging, the average increase in the thickness of IMC of increases with the decrease of the height of the microbump
different sizes of joints was 1.89 mm. At this time, the weak-
ening of IMC on shear strength dominated, and the strength of
different sizes of joints was reduced to about 26.1 MPa. The
size effect of the shear strength of the microbump joints is not
obvious anymore.
To investigate the relationship between the shear fracture
pattern and the shear strength of the microbump joints, the
fracture morphology of the samples was observed by SEM, as
shown in Figs. 4, 20 and 21. After bonding, the fracture form of
the sample with 100 mm diameter and 15 mm height was a

Table 3 e Compositions of the IMCs marked by the red


box in Fig. 17 (d ¼ 40 mm,h ¼ 15 mm).
Aging Sn Ni Cu IMC type
time (h) (at.%) (at.%) (at.%)
0 54.77 38.92 6.31 (Ni0.86,Cu0.14)3Sn4
100 54.20 36.70 9.10 (Ni0.80,Cu0.20)3Sn4
200 53.74 36.96 9.30 (Ni0.80,Cu0.20)3Sn4 Fig. 19 e Shear strengths of the Cu/Ni/Sn-3.5Ag
400 55.79 34.79 9.42 (Ni0.79,Cu0.21)3Sn4
microbumps as functions of the aging time.
1880 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

Fig. 20 e SEM micrographs of the fracture surfaces of 100 mm diameter samples after isothermal aging at 150  C for (a), (e) 0 h,
(b), (f) 100 h, (c), (g) 200 h, (d), (h) 400 h (h ¼ 25).

Fig. 21 e SEM micrographs of the fracture surfaces of 100 mm diameter samples after isothermal aging at 150  C for (a), (e) 0 h,
(b), (f) 100 h, (c), (g) 200 h, (d), (h) 400 h (h ¼ 35).

joints. After aging for 200 and 400 h, the fracture form of only dense Ni3Sn4 particles remained on the fracture surface.
microbump with a diameter of 100 mm and a height of 25 and The fractured form changes from a ductileebrittle combination
35 mm is still a ductileebrittle combination, but the brittle to a nearly complete brittle fracture, with the strength of the
fracture component is increasing with time. The number of pits microbump joints dropping to 1/5 of that after bonding.
on their fracture surfaces is increasing and the Ni3Sn4 particles An interesting phenomenon can also be found in the
at the bottom of the pits are becoming thicker. This all causes fracture morphology. The Ni3Sn4 grains located in the frac-
the strength of the microbump joints to decrease as well. For a ture surface of microbump joints follow a growth pattern of
microbump with a diameter of 100 mm and a height of 15 mm, fine grains growth after bonding (Fig. 4(e)), and coarsening
after the aging time was increased from 200 to 400 h, almost after aging for 100 h (Fig. 4(f)). The Ni3Sn4 grains grew into
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1881

Table 4 e Shear strength of samples.


Factors Shear strength (Mpa)
Solder diameter (mm) Solder height (mm) IMC thickness (mm) Trial1 Trial2 Trial3
25 15 0.76 25.0 26.2 26.8
25 15 1.87 24.4 25.6 25.9
25 15 2.61 25.5 25.8 22.8
25 15 2.91 22.1 24.3 22.0
40 15 0.75 42.6 41.0 42.7
40 15 1.82 36.6 39.6 36.3
40 15 2.47 36.2 32.4 34.3
40 15 2.85 25.3 24.4 27.4
100 15 0.72 70.3 73.1 70.5
100 15 1.76 42.8 42.4 40.8
100 15 2.32 36.9 37.0 34.7
100 15 2.63 25.1 28.4 26.9
100 25 0.67 38.7 38.6 40.3
100 25 1.6 32.7 29.6 32.2
100 25 2.14 28.3 30.7 30.7
100 25 2.47 28.7 28.6 27.3
100 35 0.54 32.2 33.8 34.2
100 35 1.4 31.8 30.9 27.9
100 35 1.8 30.6 27.5 29.2
100 35 2.07 28.8 27.4 25.9

relatively independent large grains after aging for 200 h are expected to solve this problem with the help of artificial
(Fig. 4(g)). After aging continued to 400 h, small grains of neural network models.
Ni3Sn4 grew on the independent Ni3Sn4 large grains (Fig. 4(h)) 60 sets of experiments were done to maximize the effect of
and grew toward the adjacent grains until fusion. In addition, input factors on the shear strength of Cu/Ni/Sn-3.5Ag micro-
three states of grain growth are found in Fig. 4(h); the first one bump joints by varying the geometry of the microbump joints
is that the small grains on adjacent large grains grow toward and the IMC thickness. The experimental results are shown in
each other (red arrow a, b); the second one is that the small Table 4. In this study, an artificial neural network model with
grains on adjacent large grains have grown together and microbump joints height, diameter and IMC thickness as
formed a “bridge” (red arrow c); the third one is that the input parameters and shear strength as the characterization
adjacent grains have fused together and grown into larger result was developed. The optimal structure was determined
particles (red arrow d). by using several ANN topologies to train the network and by
As shown in Fig. 10, for a microbump with a diameter of adjusting the number of neurons in the two hidden layers.
40 mm and a height of 15 mm after bonding, most of the fracture Table 5 lists the ANN topologies (hidden neuron structures)
surface is Sn-3.5Ag solder, with a few long pits having tiny used to develop the ANN models and their respective network
particles, which are still mainly ductile fractures at this time. performance. Figure 22 shows a visual representation of the
As the aging time increases, the fracture appears as a pit developed model. Approximately 9 independent ANN topol-
containing Ni3Sn4 particles, and the Ni3Sn4 particles become ogies were systematically examined by varying the number of
bigger with increasing aging time. The form of fracture is a neurons in the hidden layer in the range of 2e16. As can be
ductileebrittle combination, and the shear strength decreases seen from the group of “*” markers in Table 5, 16-4 (20
continuously as a result. In general, 40 mm diameter micro-
bump joints have lower shear strength than 100 mm diameter
joints due to the smaller volume of the solder. A similar Table 5 e Summary of ANN topologies used in the
pattern occurs in microbump joints with a diameter of 25 mm, development of the ANN model.
as illustrated in Fig. 11. The only difference is that the IMC Hidden Training Validation Test
particles are coarser in the fracture at the same aging time, neurons 2 2 2
R RMSE R RMSE R RMSE
and the shear strength is diminished even more.
2e2 0.9036 0.1241 0.6790 0.1364 0.9312 0.1473
4e2 0.9901 0.0469 0.9247 0.0632 0.8470 0.0774
3.4. ANN model development for shear strength 4e4 0.9892 0.0436 0.9868 0.0608 0.9229 0.0812
5e5 0.9790 0.0583 0.8977 0.0592 0.9778 0.0938
Through the above example, we found that the specific 8e2 0.9830 0.0608 0.9651 0.0539 0.6598 0.1334
structure of the microbump joints determines the growth of 8e4 0.9849 0.0735 0.8977 0.0714 0.9853 0.0917
16e4a 0.9843 0.0424 0.9919 0.0583 0.9903 0.0721
IMC. The IMC growth and microbump joints structure
12e10 0.9888 0.0520 0.9698 0.0624 0.8475 0.1015
together determine the shear strength of the microbump
16e8 0.9883 0.0436 0.9632 0.1204 0.9408 0.0837
joints. Therefore, there are too many factors that affect the
a
Optimum model on the basis of the training, validation and test
shear strength of microbump joints, so that it is difficult for us
data.
to decouple them into a series of single factors. In this case, we
1882 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

Fig. 22 e Schematic of artificial neural network (ANN) layers.

neurons) is clearly the best structure for the hidden layer due values of the shear strength of Cu/Ni/Sn-3.5Ag microbump
to the largest correlation coefficient R2 and the smallest root joints.
mean square error RMSE value. Therefore, the 3-16-4-1 com- It can be seen from the figure that the developed ANN
bination is the optimal ANN structure for predicting shear model performs satisfactorily by using the model with the
strength. The activation functions used for convergence are best ANN topology to obtain a good fit between the predicted
the hyperbolic tangent “tansig” in the hidden layer and the results and the experimental results. The overall correlation
linear transfer function “purelin” in the output layer. The coefficient R2 between the predicted and measured values is
training performance curve of the neural network is shown in 0.99266, indicating a significant correlation with the actual
Fig. 23, and convergence to the mean square error of values. This finding is very different in the field of microbump
3.4233  103 is achieved in 7 iterations or epochs. Figure 24 joints interconnection and reliability than previous studies
shows a comparative plot of the experimental and predicted [30,31] on bonding process on microbump joints reliability

Fig. 23 e MSE convergence during training of the ANN model 3-16-4-1 for the shear strength.
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1883

Fig. 24 e Comparison of the ANN model simulation results with the experimental results for shear strength.

prediction and can fill the blanks of microbump joints size and reveal the size effect on the interfacial Ni3Sn4 IMC growth
IMC thickness on microbump joints strength prediction very kinetics and the resulting chip-to-chip bonding strength. It
well. This shear strength prediction model integrates the was found that the IMC morphology and thickness were
microbump structure, especially the IMC thickness. It results dependent on the joint size, joints with smaller diameter and
in a more accurate prediction of the strength of microbump solder layer height produced thicker Ni3Sn4, as well as a
joints and a wider adaptation to the situation. It is possible to transition from scallop shape to needle-like shape of Ni3Sn4
adequately control the selection of dimensions (microbump was observed when downsizing the microbump joint from 100
joints height and diameter), optimize the service parameters to 25 mm. EPMA analyses showed the Ni concentration distri-
(isothermal aging time and IMC thickness), and predict the bution in the Sn-3.5Ag solder matrix was indeed higher in
shear strength of Cu/Ni/Sn-3.5Ag microbump joints with smaller or lower joints. A phenomenon directly linked to the
small errors. higher diffusion coefficient and enhanced IMC growth, due to
the reason of increased proportion of sidewall Ni diffusion and
higher IMC grain boundary density, as well as shorter diffu-
4. Conclusion sion path. The die shear strength was also size-dependent,
joints with larger diameter and thinner solder layer showed
In this study, the intermetallic compound growth in Cu/Ni/Sn- higher strength. To deal with the complex influence of various
3.5Ag microbump joints of different diameters from 100 to factors including the joint diameter, height and IMC thick-
25 mm, and solder layer heights from 35 to 15 mm, was sys- ness, an artificial neural network model with a 3-16-4-1 layer
tematically investigated under isothermal aging condition to structure was applied to generate accurate predictions on the
1884 j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5

bonding strength of chip-to-chip bonding samples, providing [9] Li XP, Xia JM, Zhou MB, Ma X, Zhang XP. Solder volume
a practical tool for the design and optimization of joint effects on the microstructure evolution and shear fracture
structure with highly complex factor combinations. behavior of ball grid array structure Sn-3.0Ag-0.5Cu solder
interconnects. J Electron Mater 2011;40(12):2425e35.
[10] Tian Y, Hang C, Wang C, Yang S, Lin P. Effects of bump size
on deformation and fracture behavior of Sn3.0Ag0.5Cu/Cu
CRediT authorship contribution statement solder joints during shear testing. Mater Sci Eng, A - Struct
Prop Microstruct Process 2011;529:468e78.
Chu Tang: Data curation, Writing - original draft. Zhuo Chen: [11] Suhir E. Analysis of a short beam with application to solder
Methodology, Writing-review and editing, Supervision. Min- joints: could larger stand-off heights relieve stress? Eur Phys
gang Fang: Investigation, Validation. Xiaoyu Xiao: Methodol- J Appl Phys 2015;71(3).
[12] Tang C, Zhu W, Chen Z, Wang L. Thermomechanical
ogy, Investigation. Gui Chen: Methodology, Investigation.
reliability of a Cu/Sn-3.5Ag solder joint with a Ni insertion
Wenhui Zhu: Conceptualization, Methodology.
layer in flip chip bonding for 3D interconnection. J Mater Sci
Mater Electron 2021;32(9):11893e909.
[13] Qin HB, Zhang XP, Zhou MB, Zeng JB, Mai YW. Size and
Declaration of Competing Interest constraint effects on mechanical and fracture behavior of
micro-scale Ni/Sn3.0Ag0.5Cu/Ni solder joints. Mater Sci Eng,
The authors declare there are no conflict of interest. A - Struct Prop Microstruct Process 2014;617:14e23.
[14] Li ZQ, Belyakov SA, Xian JW, Gourlay CM. The influence of
primary Cu6Sn5 size on the shear impact properties of Sn-Cu/
Cu BGA joints. J Electron Mater 2018;47(1):84e95.
Acknowledgments [15] Kim YH, Ma SW, Kim YH. Chip to chip bonding using Cu
bumps capped with thin Sn layers and the effect of
microstructure on the shear strength of joints. J Electron
This work is supported by National Natural Science Founda-
Mater 2014;43(9):3296e306.
tion of China (Grant No. U20A6004), the Fundamental
[16] Kim DG, Kim JW, Lee JG, Mori H, Quesnel DJ, Jung SB. Solid
Research Funds for the Central Universities of Central South state interfacial reaction and joint strength of Sn-37Pb solder
University, China, (Grant 2021zzts0144), Natural Science with Ni-P under bump metallization in flip chip application. J
Foundation of Hunan Province (Grant No. 2021JJ40734), State Alloys Compd 2005;395(1e2):80e7.
Key Laboratory of High Performance Complex Manufacturing [17] Bertheau J, Hodaj F, Hotellier N, Charbonnier J. Effect of
(Grant No. ZZYJKT2020-08), the Innovation-Driven Project of intermetallic compound thickness on shear strength of 25
mu m diameter Cu-pillars. Intermetallics 2014;51:37e47.
CSU (Grant No. 2020CX005). Key Project of Science and Tech-
[18] Ho CE, Fan CW, Hsieh WZ. Pronounced effects of Ni(P)
nology of Changsha (kq2102005). thickness on the interfacial reaction and high impact
resistance of the solder/Au/Pd(P)/Ni(P)/Cu reactive system.
Surf Coating Technol 2014;259:244e51.
references [19] Wang B, Wu FS, Wu YP, Liu H, Zhou LZ, Fang YB. Effect of
stand-off height on the microstructure and mechanical
behaviour of solder joints. Solder Surf Mt Technol
[1] Chuang HY, Chen WM, Shih WL, Lai YS, Kao CR. Critical new 2010;22(1):11e8.
issues relating to interfacial reactions arising from low solder [20] Zhu Y, Sun F. Geometric size effect on IMC growth and
volume in 3D IC packaging. In: IEEE 61st Electronic elements diffusion in Cu/Sn/Cu solder joints. Solder Surf Mt
Components and Technology Conference (ECTC), Electronic Technol 2017;29(2):85e91.
Components and Technology Conference. Lake Buena Vista, [21] Huang ML, Yang F. Size effect model on kinetics of interfacial
FL; 2011. p. 1723e8. reaction between Sn-xAg-yCu solders and Cu substrate. Sci
[2] Tu KN. Reliability challenges in 3D IC packaging technology. Rep 2014;4.
Microelectron Reliab 2011;51(3):517e23. [22] Wang DF, Ling HQ, Sun ML, Miao XY, Hu AM, Li M, et al.
[3] Yang CL, Ren SR, Zhang XD, Hu AM, Li M, Gao LM, et al. Effect Investigation of intermetallic compound and voids growth in
of Sn surface diffusion on growth behaviors of intermetallic fine-pitch Sn-3.5Ag/Ni/Cu microbumps. J Mater Sci Mater
compounds in cu/Ni/SnAg microbumps. Mater Char Electron 2018;29(3):1861e7.
2020;159. [23] Wong CK, Pang JHL, Tew JW, Lok BK, Lu HJ, Ng FL, et al. The
[4] Li GY, Chen BL. Formation and growth kinetics of interfacial influence of solder volume and pad area on Sn-3.8Ag-0.7Cu
intermetallics in Pb-free solder joint. IEEE Trans Compon and NiUBM reaction in reflow soldering and isothermal
Packag Technol 2003;26(3):651e8. aging. Microelectron Reliab 2008;48(4):611e21.
[5] Chuang HY, Yu JJ, Kuo MS, Tong HM, Kao CR. Elimination of [24] Wang SB, Yao Y, Long X. Size effect on microstructure and
voids in reactions between Ni and Sn: a novel effect of silver. tensile properties of Sn3.0Ag0.5Cu solder joints. J Mater Sci
Scripta Mater 2012;66(3e4):171e4. Mater Electron 2017;28(23):17682e92.
[6] Kotadia HR, Howes PD, Mannan SH. A review: on the [25] Magnien J, Khatibi G, Lederer M, Ipser H. Investigation of
development of low melting temperature Pb-free solders. interfacial behavior in miniaturized solder interconnects.
Microelectron Reliab 2014;54(6e7):1253e73. Mater Sci Eng, A - Struct Prop Microstruct Process
[7] Yu DQ, Wang L. The growth and roughness evolution of 2016;673:541e50.
intermetallic compounds of Sn-Ag-Cu/Cu interface during [26] Hu XW, Li YL, Liu Y, Liu Y, Min ZX. Microstructure and shear
soldering reaction. J Alloys Compd 2008;458(1e2):542e7. strength of Sn37Pb/Cu solder joints subjected to isothermal
[8] Qin HB, Zhang XP, Zhou MB, Li XP, Mai YW. Geometry effect aging. Microelectron Reliab 2014;54(8):1575e82.
on mechanical performance and fracture behavior of micro- [27] Yang LM, Zhang QK, Zhang ZF. Effects of solder dimension
scale ball grid array structure Cu/Sn-3.0Ag-0.5Cu/Cu solder on the interfacial shear strength and fracture behaviors of
joints. Microelectron Reliab 2015;55(8):1214e25. Cu/Sn-3Cu/Cu joints. Scripta Mater 2012;67(7e8):637e40.
j o u r n a l o f m a t e r i a l s r e s e a r c h a n d t e c h n o l o g y 2 0 2 2 ; 1 8 : 1 8 6 5 e1 8 8 5 1885

[28] Britto ASF, Raj RE, Mabel MC. Prediction of shear and tensile [41] Wilamowski BM, Yu H. Improved computation for
strength of the diffusion bonded AA5083 and AA7075 Levenberg-Marquardt training. IEEE Trans Neural Network
aluminium alloy using ANN. Mater Sci Eng, A - Struct Prop 2010;21(6):930e7.
Microstruct Process 2017;692:1e8. [42] Sheela KG, Deepa SN. New criteria to fix number of hidden
[29] Dewan MW, Huggett DJ, Liao TW, Wahab MA, Okeil AM. neurons in multilayer perceptron networks for wind speed
Prediction of tensile strength of friction stir weld joints with prediction. Wind Struct 2014;18(6):619e31.
adaptive neuro-fuzzy inference system (ANFIS) and neural [43] Sheela KG, Deepa SN. Review on methods to fix number of
network. Mater Des 2016;92:288e99. hidden neurons in neural networks. Math Probl Eng
[30] Dele-Afolabi TT, Hanim MAA, Ojo-Kupoluyi OJ, Jung DW, 2013;2013.
Nuraini AA, Erameh AA. Interfacial IMC evolution and shear [44] Yang Y-S, Yang C-J, Ouyang F-Y. Interfacial reaction of
strength of MWCNTs-reinforced Sn-5Sb composite solder Ni3Sn4 intermetallic compound in Ni/SnAg solder/Ni system
joints: experimental characterization and artificial neural under thermomigration. J Alloys Compd 2016;674:331e40.
network modelling. J Mater Res Technol 2021;13:1020e31. [45] Kim KS, Huh SH, Suganuma K. Effects of cooling speed on
[31] Lin YH, Deng WJ, Shie JR, Yang YK. Optimization of reflow microstructure and tensile properties of Sn-Ag-Cu alloys.
soldering process for BGA packages by artificial neural Mater Sci Eng, A - Struct Prop Microstruct Process
network. Microelectron Int 2007;24(2):64e70. 2002;333(1e2):106e14.
[32] Liukkonen M, Hiltunen T, Havia E, Leinonen H, Hiltunen Y. [46] Ma H, Kunwar A, Huang R, Chen J, Wang Y, Zhao N, et al. Size
Modeling of soldering quality by using artificial neural effect on IMC growth induced by Cu concentration gradient
networks. IEEE Trans Electron Packag Manuf 2009;32(2):89e96. and pinning of Ag3Sn particles during multiple reflows.
[33] Mita M, Kajihara M, Kurokawa N, Sakamoto K. Growth Intermetallics 2017;90:90e6.
behavior of Ni3Sn4 layer during reactive diffusion between Ni [47] Chu K, Lee C, Park S-H, Sohn Y. Effects of Ag addition and
and Sn at solid-state temperatures. Mater Sci Eng, A - Struct Ag3Sn formation on the mechanical reliability of Ni/Sn solder
Prop Microstruct Process 2005;403(1e2):269e75. joints. Microelectron Reliab 2017;75:53e8.
[34] Wang Y, Huang JH, Ye Z, Peng XW, Yang J, Chen SH, et al. [48] Hadian F, Schoeller H, Cotts E. Correlation between the
Growth kinetics of Ni3Sn4 in the solid-liquid interfacial growth of voids and Ni3Sn4 intermetallic compounds at
reaction. Metall Mater Trans Phys Metall Mater Sci SnAg/Ni and SnAgCuBiSbNi/Ni interfaces at temperatures up
2019;50A(7):3038e43. to 200 degrees C. J Electron Mater 2020;49(1):226e40.
[35] Shen J, Chan YC, Liu SY. Growth mechanism of Ni3Sn4 in a [49] Son K, Ryu H, Kim G, Lee J, Park Y-B. Ieee in electromigration
Sn/Ni liquid/solid interfacial reaction. Acta Mater polarity effect of Cu/Ni/Sn-Ag microbumps for three-
2009;57(17):5196e206. dimensional integrated circuits. In: 19th IEEE Electronics
[36] He M, Chen Z, Qi GJ. Solid state interfacial reaction of Sn- Packaging Technology Conference (EPTC), Singapore; 2017.
37Pb and Sn-3.5Ag solders with Ni-P under bump SINGAPORE.
metallization. Acta Mater 2004;52(7):2047e56. [50] Huang ML, Yang F. Size effect on interfacial reactions of Sn-
[37] Alam MO, Chan YC. Solid-state growth kinetics of Ni3Sn4 at 3.0Ag-0.5Cu solder balls on Cu and Ni-P pads. J Mater Sci
the Sn-3.5Ag solder/Ni interface. J Appl Phys 2005;98(12). Mater Electron 2015;26(2):933e42.
[38] Raj RE, Daniel BSS. Prediction of compressive properties of [51] Dybkov VI. Reaction diffusion and solid state chemical
closed-cell aluminum foam using artificial neural network. kinetics. Kyiv: IPMS; 2002.
Comput Mater Sci 2008;43(4):767e73. [52] Liu Y, Chu YC, Tu KN. Scaling effect of interfacial reaction on
[39] Akbari MK, Shirvanimoghaddam K, Hai Z, Zhuiykov S, intermetallic compound formation in Sn/Cu pillar down to
Khayyam H. Al-TiB2 micro/nanocomposites: particle capture 1mm diameter[J]. Acta Mater 2016;117:146e52.
investigations, strengthening mechanisms and [53] Tseng YC, Lee H, Tsai SC, Yen YW, Chen CM. Suppression
mathematical modelling of mechanical properties. Mater Sci effect of Ni grain size on the Ni3Sn4 growth at the Sn/Ni
Eng, A - Struct Prop Microstruct Process 2017;682:98e106. interface. Mater Char 2017;128:232e7.
[40] Mukherjee I, Routroy S. Comparing the performance of [54] Sun F, Yin Z, Guo M. The effect of temperature gradient on
neural networks developed by using Levenberg-Marquardt interfacial Cu6Sn5 growth during thermal compression
and Quasi-Newton with the gradient descent algorithm for bonding. J Mater Sci Mater Electron 2019;30(15):13957e63.
modelling a multiple response grinding process. Expert Syst
Appl 2012;39(3):2397e407.

You might also like