Progress in Materials Science 55 (2010) 428–475
Contents lists available at ScienceDirect
Progress in Materials Science
journal homepage: www.elsevier.com/locate/pmatsci
Failure mechanisms of solder interconnects under current
stressing in advanced electronic packages
Y.C. Chan *, D. Yang
EPA Centre, Department of Electronic Engineering, City University of Hong Kong, Tat Chee Avenue, Kowloon Tong, Hong Kong
a r t i c l e
i n f o
Article history:
Received 18 November 2008
Received in revised form 5 November 2009
Accepted 6 January 2010
a b s t r a c t
The pursuit of greater performance in microelectronic devices has
led to a shrinkage of bump size and a significant increase in electrical current. This has resulted in a high current density and accompanying Joule heating in solder interconnects, which places great
challenges on the reliability of advanced electronic packages. A
review of current stressing-induced failures of solder interconnects
is thus timely. This review is devoted to five types of physical failure mechanisms occurring in high current density applications,
which include electromigration (EM), Joule heating-induced failures, interfacial reactions, stress-related damage, and thermomigration (TM). In practice, some of these failure mechanisms are
mixed together so that the real root cause cannot be easily
detected and understood. Reliability designers need to be well
informed to evaluate the electrical characteristics, thermal characteristics and mechanical strength for solder interconnects in
advance. This review summarizes recent progress and presents a
critical overview of the basis of atomic transport, diffusion kinetics,
morphological evolution, and numerical simulation. Special
emphasis is on the understanding of the interactions of EM with
other failure mechanisms. Aside from the review of the current status of knowledge, the remaining challenges as well as future directions are also discussed.
Ó 2010 Elsevier Ltd. All rights reserved.
* Corresponding author. Tel.: +852 2788 7130; fax: +852 2788 8803.
E-mail address: eeycchan@cityu.edu.hk (Y.C. Chan).
0079-6425/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.pmatsci.2010.01.001
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
429
Contents
1.
2.
3.
4.
5.
6.
7.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1.
Solder interconnects for advanced electronic packaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.
Challenges of high current density applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.
The scope of this review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
EM in solder interconnects under an electron wind force. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.
Phase separation and atomic transport under current stressing . . . . . . . . . . . . . . . . . . . . . . . .
2.2.
Current stressing-enhanced phase coarsening. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.
Nucleation and growth of voids at the interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.
Lifetime statistics and the reliability of EM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Joule heating-enhanced dissolution of UBM and the diffusion of on-chip metal interconnects . . . . . .
3.1.
Effect of Joule heating due to current stressing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.
Dissolution of UBM layers and possible solutions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.
Melting of solder interconnects due to Al diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Effect of current stressing on the formation of IMCs and kinetic analysis . . . . . . . . . . . . . . . . . . . . . . .
4.1.
Polarity effect of current stressing and enhanced growth of IMCs at the anode. . . . . . . . . . . .
4.2.
Dynamic equilibrium of IMCs at the cathode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.
Abnormal polarity effect of current stressing on the formation of IMCs . . . . . . . . . . . . . . . . . .
Stress-related degradation of solder interconnects under EM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.
Morphological evolution due to EM and a back stress in solder interconnects . . . . . . . . . . . .
5.2.
Mechanical deformation and degradation under current stressing . . . . . . . . . . . . . . . . . . . . . .
TM behavior in solder interconnects under a thermal gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.1.
TM in tin–lead solder interconnects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2.
TM in Sn-based lead-free solder interconnects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.1.
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.
Further problems which need to be addressed in the near future. . . . . . . . . . . . . . . . . . . . . . .
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
429
429
429
432
432
432
434
436
439
441
441
442
445
448
448
450
451
453
453
458
462
462
468
470
470
471
472
472
1. Introduction
1.1. Solder interconnects for advanced electronic packaging
With the trend towards higher integration and further miniaturization of Si-based devices, electronic packaging is successively requiring a higher input/output (I/O) density, smaller feature size,
and better performance. Concurrently, the flip chip solder interconnect has established its leadership
role for high current density packages as thousands of solder bumps are fabricated onto one single
chip. To meet an even higher demand for device performance, the I/O number is expected to increase,
while the dimensions of each individual bump will accordingly need to shrink.
According to the 2003 International Technology Roadmap for Semiconductors (ITRS), a significant
downsizing in flip chip packaging is anticipated [1]. Fig. 1 shows the anticipated variation in pad diameter, pad pitch and line width.
Also the bump size is expected to be reduced along with the pad size and pitch. At present, the
diameter of a solder bump in use is about 100 lm or less [2]. In 2007, the diameter of micro-bumps
had been decreased to 20 lm [3]. This relentless scaling-down will inevitably place severe challenges
on the reliability of micro-devices, as will be discussed later.
1.2. Challenges of high current density applications
At present, in the microelectronic industry, each solder joint is designed to carry 0.2 A, and this will
be doubled in the near future [2]. This means that the average current density through a 50 lm diameter solder joint may approach 104 A/cm2. This demands a reduced cross-section of the conductive
lines and solder interconnects, whilst on the other hand, they are expected to conduct such a high
430
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
160
pad pitch
pad diameter
line width
140
Size (micrometre)
120
100
80
60
40
20
0
2002
2004
2006
2008
2010
2012
2014
2016
2018
2020
Year
Fig. 1. A downsizing in flip chip packaging, based on 2003 ITRS edition.
current density. Meanwhile, since Joule heating is proportional to the square of the current density,
the local temperature of conductive lines and solder bumps will rise substantially. Also, during field
service, the solder joints will experience a temperature rise of at least 100 °C, to approximately 82%
and 76% of the melting temperatures of eutectic SnPb and SnAgCu, respectively. As a consequence, under the combined effect of a high current density and a high homologous temperature, easy diffusion
of atoms in the lattice is anticipated [4]. This renders electromigration (EM) a serious reliability issue
in the application of high current density packages.
EM means a diffusion controlled mass transport phenomenon due to the application of electrical
current. In 1961, Huntington and Grone proposed that thermally-activated metal ion becomes essentially free in the lattice and is acted upon by two opposing forces (a direct force and an electron wind
force) in a metal [5]. Also, they identified the electron wind force as the primary driving force responsible for the EM failure observed in interconnects. The electron wind force is further explained as one
force experienced by a metal ion in the direction of the electron flow due to the momentum exchange
between the moving electrons and the ion. Therefore, the phenomenological equation for the atomic
flux due to EM (Jem) is described as:
J em ¼ C m ¼ C
DF
D
D
¼ C Z eE ¼ C Z eqj;
kT
kT
kT
ð1Þ
where Z is a dimensionless quantity known as the effective charge or the effective valence that reflects the direction and the magnitude of the momentum exchange, e is the electron charge, E is the
electric field, q is the resistivity, j is the current density, C is the concentration of diffusing atoms, v
is the drift velocity of these atoms, D is the thermally-activated diffusivity, and kT is the average thermal energy.
As regards the EM of tin and lead in solder alloys, the first published work was given by Brandenburg and Yeh in 1998 [6]. EM causes the net atom transport of solder material along the direction of
the electron flow. Since 2002, ITRS has started to include this reliability problem for industrial attention [7]. Table 1 lists the near-term reliability challenges requiring concern in current assembly and
packaging techniques [3]. According to the 2007 ITRS, EM will become a more limiting factor of high
current density packages, such as wafer-level packaging (WLP) for micro-electro-mechanical systems
(MEMS). It is suggested that physical failure mechanisms such as EM, and thermal migration in combination with mechanical stresses should be understood and modeled for practical life assessment. In
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
431
Table 1
Assembly and packaging – difficult challenges [3].
Difficult challenges
Summary of issues
High current density
packages
Electromigration will become a more limiting factor. It must be addressed through changes in
materials together with thermal/mechanical reliability modeling
Whisker growth
Thermal dissipation
particular, solder and the under bump metallurgy (UBM) need to be well designed to support a high
current density and minimize or avoid EM.
Also, thermal dissipation is addressed as a critical factor of reliability considering the large Joule
heating generated by the on-chip metal interconnects. One should appreciate that the cross-sectional
area of conductive lines on the chip has been decreased significantly with the trend towards miniaturization, as shown in Fig. 1. It is true to say that the on-chip interconnects used in electronic packages
may be based on Al or Cu materials. However, most concern is with the interconnects based on Al,
where Joule heating is more pronounced (the resistivity of Cu is about 60% of that of Al). Fig. 2 provides
a schematic diagram of a flip chip interconnect. Normally the electrical resistance of the Al traces is at
least one order of magnitude higher than that of the solder joints and Cu conductors, and thus the Al
traces are the primary heat source. A significant Joule heating will accelerate the EM process in the
neighboring solder joints, but also result in the degradation of UBM layers and even the Al trace itself.
With the passage of current in service, the momentum transfer from electrons to atoms may also
play a significant role in interfacial reactions. The formation of an intermetallic compound (IMC) layer
at the interface is crucial for the good adhesion between a solder bump and a UBM, of which the driving force comes from the chemical potential gradient between different contacting materials. However, a rapid interaction may occur to form a substantial amount of IMC, which is known to be
brittle and has a negative effect on the mechanical integrity of interconnects. Therefore, the growth
of IMC particles/layer accelerated by current stressing and the role of EM in interfacial reactions is
becoming an important and challenging problem in solder reliability.
Another concern is the formation of compression and tension regions inside a solder joint, when
atoms are driven from the cathode to the anode by the electron wind force. Stress generation and
relaxation are issues under exploration, and stress-related damage under a current density should
be paid considerable attention. In addition, the mechanical properties are direct indicators of strength
and long-term durability. It is understandable that EM would exert a certain effect on the mechanical
transition of solder interconnects. As an important reliability factor, the mechanical behavior of solder
interconnects for high current density applications also needs to be carefully considered.
Moreover, owing to the significant heat accumulation, atomic migration process, thermomigration
(TM), may be triggered and influence the reliability of packages. Due to differences in electrical resistance and thermal dissipation of individual parts within the flip chip interconnect structure (see
Fig. 2), it is predicted that the heat accumulated at the chip side will be larger than that at the substrate side. This will inevitably lead to a considerable temperature gradient across solder joints, which
can provide a driving force for atomic diffusion to trigger TM. More exactly, the driving force of TM
comes from the energy transported by the moving atoms and the interactions with the usual heat carriers in the lattice [8]. The phenomenological equation for the atomic flux due to TM (Jtm) is [9]:
Fig. 2. A schematic diagram of a flip chip interconnect including Al traces, solder joints and Cu conductors.
432
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
J tm ¼ C m ¼ C
DF
DQ =N
ðDTÞ;
¼C
2
kT
kT
ð2Þ
where Q is the heat of transport and is the heat flow per mole that must be supplied to maintain unit
molar flow in the steady state, N is Avogadro’s number, and DT is the thermal gradient. The other
terms have been defined under Eq. (1). Being a potential reliability concern for flip chip solder interconnects, TM induced-void or pore formation has also been introduced in the 2007 ITRS [3].
1.3. The scope of this review
During field service, all the factors discussed are supposed to combine and act concurrently, which
will further complicate the failure processes. Therefore, a quantitative understanding of the physics
and mechanics of each failure mechanism will bring about a good appreciation which will help in
the design and life prediction in flip chip solder interconnects, which should be of particular interest
to both those in industry and in academe.
Five types of failure mechanism are presented in this review. Section 2 addresses the dominant
physical mechanisms during EM in solder interconnects, including phase separation, phase coarsening, and void formation. Also, from an engineering point-of-view, this section gives a summary of
the lifetime statistics and reliability evaluation of the EM of solders. Section 3 reviews the dissolution
of the UBM due to an accelerated interstitial diffusion, and the diffusion of on-chip Al trace under Joule
heating, then introduces research results on the time-dependant melting behavior of solder interconnects under current stressing. Section 4 presents a critical literature review related to interfacial reactions. The kinetics dominating in these interfacial reactions are investigated and discussed. A
summary of stress-related degradation in solder interconnects is presented in Section 5. The morphological evolution due to EM, and the back stress induced are described. Also some mechanical deformation and degradation mechanisms under current stressing are summarized as a part of an overall
understanding of the mechanical behavior. Section 6 discusses the reliability concerns of TM. The thermo-transports of Pb, Sn, Cu and Bi in solder interconnects under a thermal gradient are introduced.
Lastly but importantly, some issues that need to be clarified in the near future are proposed in
Section 7.
2. EM in solder interconnects under an electron wind force
2.1. Phase separation and atomic transport under current stressing
In eutectic two-phase solder joints, phase separation is likely to occur under EM due to the different atomic diffusivities over a range of operational temperatures. It was found that in a eutectic
Sn37Pb solder, Sn and Pb are the dominant diffusing species during EM at room temperature and
above 100 °C, respectively. This is in agreement with a study of tracer diffusion, where the result
showed that Sn has a larger diffusivity than Pb below 100 °C whereas Pb is the faster moving species
above this temperature [10]. Brandenburg and Yeh observed Pb migration with the electron flow
above 100 °C accompanied by phase separation in tin–lead solder [6]. This was also supported by later
findings in EM experiments above 100 °C [11,12]. In the case of EM at 100 °C, Agarwal et al. examined
the EM behavior and identified that Pb is still the dominant diffusion species [13]. Compared to the
results above, EM tests were carried out at room temperature by Liu et al. [14]. It was found that near
room temperature Sn is the dominant diffusing atom, and interfaces serve as the fastest kinetic path of
mass transport. However, a recent study through X-ray fluorescence spectroscopy has shown that Sn
atoms move faster than Pb in the initial stages even above 100 °C [15]. Further investigations on the
movement of Sn and Pb undergoing EM over a wide temperature range need to be developed to identify the in situ atomic transport. On the whole, this variation of migration behavior imposes a serious
limitation on an understanding of the physics of failure, and on conducting and interpreting accelerated tests of EM in tin–lead solder.
Since eutectic SnBi is very close to eutectic SnPb in its microstructure and constitution, SnBi solder
exhibits similar EM characteristic [16,17]. Fig. 3 illustrates the mass accumulation of Bi in a Sn58Bi
433
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 3. SEM image of mass accumulation of Bi in a Sn58Bi solder joint under a current density of 5 103 A/cm2 at 75 °C.
solder joint under a current density of 5 103 A/cm2 at 75 °C. As the dominant effective charge carrier,
Bi migrated along with the electron flow and substantially accumulated at the anode side. Also, as a
result of the large concentration of Bi, a phase segregation between Bi and Sn was evident, which behaved like the Pb segregation with Sn in Sn37Pb solder.
For most lead-free solders, e.g., Sn3.5Ag, Sn0.7Cu, and Sn4Ag0.5Cu, since the amount of elements
substituted is not as high as that of Bi in eutectic Sn58Bi, the migration of the substitutional elements
(Ag, Cu) is insignificant as compared with the self diffusion of Sn and thus the phase separation may be
negligible. Instead, a fast interstitial diffusion of Cu, Ni and Ag – induced dissolution of UBM and IMC
plays a dominant role in the electrical failure of the lead-free solder systems above, as will be mentioned in Section 3.2.
To understand the migration mechanism, the atomic flux due to EM was estimated from the phase
displacement, and the product of diffusivity and effective charge number was also calculated. According to Eq. (1), the product of the diffusivity and the effective charge number, DZ, should be:
DZ ¼ J em
kT
:
eqjC
ð3Þ
The total volume of atomic transport during the time in operation can be obtained approximately
from the product of the width and the cross-sectional area of the solder joint. According to the research by Lee et al. [18], the accumulated width of atoms increased linearly with time during EM.
In the case considered by us, the product DZ was calculated to be 6.5 1011 cm2/s for the migration
of Pb under a current density of 2.0 104 A/cm2 at 100 °C [19]. This is the same order of magnitude as
the results from other groups [18]. To measure the displacement driven by EM, inert particles or nanoindentations have been used as surface markers, which moved in the opposite direction to the EM flux
as expected. Also, the product for the migration of Bi was investigated [20]. Table 2 lists the DZ values
obtained for a Sn58Bi solder under different test conditions. However, since the diffusivity is strongly
related to the microstructural changes in the EM process, there exists a large uncertainty in determining the value of the effective charge number. A few studies have been conducted to unravel the diffusivity and explore the effective charge number [18,21].
On the other hand, for a SnAgCu solder with a higher melting temperature, under similar experimental conditions, the marker motion was found to be less significant, which indicates that the rate
of EM is lower in this lead-free solder [21,22].
Table 2
DZ values for a Sn58Bi solder under different test conditions [20].
Test conditions
3
Accumulation rate (cm/s)
2
35 °C/5 10 A/cm
55 °C/5 103 A/cm2
75 °C/5 103 A/cm2
10
3.02 10
4.70 1010
1.16 109
Atomic flux (atoms/cm2 s)
12
4.86 10
7.56 1012
1.87 1013
Product DZ (cm2/s)
6.60 1011
1.09 1010
2.84 1010
434
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
2.2. Current stressing-enhanced phase coarsening
During current stressing, another noticeable phenomenon of the microstructural evolution in solder joints is phase coarsening. Fig. 4 illustrates the effect of current stressing on phase coarsening in
Sn37Pb and Sn3.5Ag0.5Cu solder joints under a moderate current density of 6 102 A/cm2 at 125 °C
[23]. Fig. 4a and b shows the typical microstructures of Sn37Pb and Sn3.5Ag0.5Cu solder joints as-reflowed, in which a fine sized Pb-rich phase or Ag-rich IMC (Ag3Sn) is dispersed in a Sn-rich matrix,
respectively. After being stressed for 600 h, the coarsening of the Pb and Ag-rich phase was remarkable, as shown in Fig. 4c and d. Using electrothermal simulation and an infrared thermal technique,
it was found that the current stressing induced a substantial Joule heating in the solder interconnects.
Hence, it was proposed that the phase coarsening was enhanced by the current stressing as a result of
enhanced diffusion related to an elevated temperature and atomic stimulation [23,24].
Also, Fig. 5a presents the percentage distribution versus the particle size after current stressing for
Ag-rich phases. With an increase in stressing time, the heights of the peaks decreased and the curve
extended to a larger size with a wider range, which indicates a general tendency of overall coarsening.
The average size as a function of time is plotted in Fig. 5b. One recalls the typical equation of phase size
coarsening (Ostwald ripening) is [25]:
n
n
Q
d d0 ¼ At ¼ A0 e kT t;
ð4Þ
where d is the phase particle size after being annealed at a temperature of T for a time t, d0 is the initial
phase particle size, A0 is a pre-exponential term, Q is the rate-controlling activation energy, and n is
the phase size exponent related to a specific atomic transportation mechanism.
Fig. 4. (a) Typical microstructure of a Sn37Pb solder joint as-reflowed, (b) a Sn3.5Ag0.5Cu solder joint as-reflowed, (c) Pb-rich
phase coarsening in a Sn37Pb solder joint, and (d) Ag-rich IMC coarsening in a Sn3.5Ag0.5Cu solder joint [23].
435
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
(a)
40
Stressed for 600 hours(412 particles)
35
Stressed for 200 hours(302 particles)
Percentage (%)
30
As reflowed(540 particles)
25
20
15
10
5
0
0
0.4
0.8
1.2
1.6
2
2.4
2.8
3.2
3.6
Ag-rich Particle Size (micrometer)
(b) 0.7
Average Phase Size (micrometer)
0.6
0.5
0.4
0.3
0.2
Stressed at 125 degree Celsius
Aged at 125 degree Celsius
0.1
0
0
100
200
300
400
500
600
700
Aging or Stressing Time (hour)
Fig. 5. (a) The percentage distribution versus the particle size after current stressing for Ag-rich phases, and (b) the average size
as a function of time [23].
Then the derivative form of this equation is:
ln
dd
A0 Q
¼ ln þ ð1 nÞ ln d:
dt
n kT
ð5Þ
Through linear regression, the value of n for Ag-rich phase growth was estimated to be 2.96, corresponding to the controlling kinetics of volume diffusion. Also, the n for Pb-rich phase growth in our
study was approximately 3.28, which corresponded to the co-kinetics of grain boundary and volume
diffusion [26]. Similar to eutectic SnPb solder, current stressing-enhanced Bi coarsening was also detected in SnBi solder joints in our group [20].
Moreover, a phase-coarsening model including the influence of the current density was developed
by Ye et al. based on an experimental study of the coarsening of the Pb phase in eutectic Sn37Pb flip
chip solder joints [27]. It was found that the electric current had a greater influence on the Pb phase
436
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
growth than the temperature. Thus, a modified equation was proposed by adding a current density
term into Burke and Turnbull’s equation:
n
n
m
d d0 ¼ Aj t;
ð6Þ
where j is the current density, and m is the exponent for the current density. The n was found to be 5.5
for the Pb-rich phase growth, and the current density exponent m was about 3. However, a theoretical
explanation for the m is not yet clear.
It is understandable that a solder with a phase-coarsening microstructure is susceptible to damage
by EM. Solder normally operates at a high homologous temperature (TH > 0.5), where both lattice and
grain boundary diffusivities play important roles. The net contribution of these two diffusion mechanisms in solder is dependent on the grain size. Therefore, measures need to be taken to convolute the
arrangement of grain boundaries so that the process of diffusion is dominated by lattice rather than
grain boundary diffusion. For a better EM-resistant performance, the microstructure with triple points
or intersecting grain boundaries is preferred, since these intersections interrupt the diffusion path and
force the atoms to transport laterally following the grain boundaries or migrate into the blocking
grains [28].
Chen et al. found that doping with Cu could retard the phase coarsening of a solder under current
stressing [29]. The coarsening rate was significantly reduced from 4.6 to 1.4 lm3/h when 1 wt.% Cu
was added. They proposed that the Cu added reacted with Sn, and small Cu6Sn5 IMC precipitates were
formed inside the solder. These precipitates then acted as roadblocks which retarded the movement of
grain boundaries, leading to a reduced grain coarsening rate. Also, they investigated the effect of a
0.5 wt.% Ag addition on the EM behavior. It was found that some plate-like Ag3Sn precipitates were
formed inside the solder, and behaved as obstacles that intercepted the atomic migration [30].
2.3. Nucleation and growth of voids at the interface
During EM, atomic diffusion-induced microstructural evolution includes not only phase separation
and coarsening, but also void creation at interfaces. Fig. 6 illustrates the typical EM failure of a Sn37Pb
solder joint under a current density of 2.0 104 A/cm2 after 100 h at 100 °C [31]. A tilting effect of the
current [32–34] in such a line-to-bump structure was evident during the EM process. The Pb-rich
phase migrated towards and accumulated at the anode side corresponding to the entry direction of
the electron flow. Moreover, it is noticeable that these voids were primarily generated near the current
crowding regions because of the concentrated flux divergence and more serious UBM consumption.
A three-dimensional finite-element simulation was performed to demonstrate the current density
distribution in a solder interconnect with a 1.6 A current applied. As shown in Fig. 7, the local current
density at the entry location reached 7 109 A/m2 (i.e., 7 105 A/cm2), at least one order of magnitude larger than the average. Near this current crowding region, the atomic flux divergence may cause
vacancy accumulation and hence voids, as can be seen in Fig. 6.
Fig. 8 displays the progress of void growth in Sn3.5Ag1.0Cu solder joints under a current density of
1.5 104 A/cm2 at 125 °C [19]. Fig. 8a shows the typical morphology of the interface before the experiment. After a stressing time of 75 h, as shown in Fig. 8b, voids were initiated from the upper-right
corner, and gradually displaced the current to the surrounding areas which resulted in a lateral
growth. Since the growth of voids induced the redistribution of the current, it is also reasonable to find
that the voids were developed towards the periphery of the UBM opening, where the current density is
originally low. This experimental finding verifies the finite-element simulation by Liang et al. [35].
Fig. 8c shows the microstructural development after 280 h. It is evident that the voids continuously
extended from the right-hand to the left-hand regions. Fig. 8d displays further void growth after 425 h.
The propagation of voids decreased the effective contact area of the current path and induced a more
serious current crowding, and thus accelerated the void growth along the interface. This process continued till the voids finally spread across the complete contact window at 515 h, as shown in Fig. 8e.
From Fig. 8, it is believed that the first void nucleation took less than 14% of the failure time. The
failure time was then more dependent on the void growth than the void nucleation. Likewise, Chiu and
Chen monitored void formation and propagation in Sn37Pb solder joints under a current density of
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
437
Fig. 6. (a) SEM image of void formation at the current crowding region (the upper-right corner) of a Sn37Pb solder joint under a
current density of 2.0 104 A/cm2 after 100 h at 100 °C, and (b) local magnified micrograph [31].
Fig. 7. Current density distribution of a Sn37Pb solder joint stressed with a 1.6 A current (the current crowding effect is
apparent at the entry direction of the current) [31].
6.5 103 A/cm2 at 150 °C [36]. They found that voids started to form at approximately 10% of the failure time and they grew for the remaining of 90% of the failure time. This is different from the EM
behavior of Al and Cu interconnects. In Al and Cu interconnects, the failure is basically controlled
by the nucleation of voids, and the growth becomes very rapid once the voids are produced. By contrast, in the study by Yeh et al., they proposed that it took 88% of the failure time to initiate the first
438
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 8. SEM images of the morphological evolution in Sn3.5Ag1.0Cu solder joints under a current density of 1.5 104 A/cm2 at
125 °C (a) before the experiment, the time point A, (b) after 75 h, 14% of the failure time, the time point B, (c) after 280 h, 53% of
the failure time, the time point C, (d) after 425 h, 81% of the failure time, the time point D, and (e) after 515 h, 98% of the failure
time, the time point E.
few voids, whereas only 12% of the failure time was spent in void propagation until the final open failure [32]. According to their study, the incubation time for void nucleation was relatively long.
Based on the results from the time points A to E in Fig. 8, the growth rate was found to be about
0.32 lm/h for the whole process, which matches well with the result by Chiu and Chen [36]. They
found a void growth rate of 0.3 lm/h in the later stages in Sn37Pb solder joints. However, this is different from the investigation conducted by Zhang et al. [37]. They reported a void growth rate of
4.4 lm/h in Sn4.0Ag0.5Cu solder joints which experienced EM under a current density of
3.7 103 A/cm2 at 146 °C. A theoretical value was also calculated under a continuity condition according to the kinetic model they proposed, which was in accord with the experimental result. However,
similar to research in Al interconnects, thin-film test structures should be prepared to directly measure the material depletion in solder over an EM period, so that the atomic drift velocity may be precisely obtained. This method has been utilized to explore some EM parameters in Sn3.5Ag solder [38].
The growth of voids has been understood in relation to a variation in electrical resistance. A Kelvin
structure was designed and employed to monitor the resistance variation of a solder joint with the
propagation of voids [39,40]. A change in bump resistance as small as 0.01 mX could be detected in
this way, and it is effective to monitor how the void growth induced the resistance change in a single
solder joint. It was found that, when the percentage depletions of the contact opening were about 50%
and 80%, the maximum resistance increases could reach 70% and 250% of its initial value, respectively.
439
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
0.45
0.40
Voltage (V)
0.35
0.30
0.25
0.20
A
D
C
B
E
0.15
0
100
200
300
400
500
600
Time (h)
Fig. 9. The voltage as a function of time when an interconnect was stressed by 1.2 A at 125 °C (note this data arises from the
same experiments as Fig. 8, and the time points A–E refer to the same times in both figures).
Three-dimensional simulations for different stages of void propagation by Liang et al. also fitted well
with these results [35].
Fig. 9 shows the typical variation in the voltage as a function of time. Solder interconnects experiencing EM mostly exhibit such a characteristic of the variation of resistance [32,41,42]. There is a long
incubation time with very little resistance change below 90% of the stressing time. The solder joints
only contribute a minor part to the resistance of the whole interconnect structure as compared with
the Al traces and the Cu conductors, thus the effect of void propagation on the resistance of an interconnect was less significant before the UBM was completely detached from the solder. This is why solder interconnects mostly retain a low electrical resistance in the early stages of the stressing time,
although void accumulation has occurred in the solder joints. After that, the resistance rose abruptly
to an open circuit, since the expanding voids led to the final failure in the contact. It is also suggested
that the cause of the rapid increase of resistance in the later stages may involve Al degradation, which
will be emphasized in more detail in Section 3.3.
2.4. Lifetime statistics and the reliability of EM
From an engineering point-of-view, a mean-time-to-failure (MTTF) estimation of solder interconnects is of great interest, and a systematic reliability evaluation on EM is needed. For Al interconnects,
it is reported that the EM lifetime mostly followed a log-normal distribution. Recently, Black’s model
has been introduced to describe the EM lifetime of a solder based on the assumption that the failure is
controlled by void damage, and a log-normal function has been applied frequently [42–44]. However,
the reason for such a log-normal distribution has not been clarified. On the other hand, a Weibull analysis has been performed on time-to-failure (TTF) data in other research studies [6,45]. In our case, the
EM failure of Sn3.5Ag1.0Cu solder joints followed Weibull statistics very well with current densities
ranging from 1 104 to 2 104 A/cm2 at 100, 125, and 150 °C [19]. Fig. 10 illustrates the distribution
function under various current densities at 125 °C. As expected, the reliability of EM degraded with an
increase of current density.
Nevertheless, it is noticed that the predicted lifetimes did not match well with the measured ones,
particularly under a higher current density [46]. As a result, it is suggested that the model should be
440
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 10. Weibull cumulative distribution under various current densities at 125 °C (A: 2.0 104 A/cm2, B: 1.5 104 A/cm2, C:
1.0 104 A/cm2).
modified to include the effect of current crowding and Joule heating. A modified Black’s equation was
proposed by inserting a multiplying factor (c) of the current density and a temperature increment (DT)
into Black’s equation [34,46]:
MTTF ¼ A
1
Q
;
m exp
kðT þ DTÞ
ðcjÞ
ð7Þ
where A is a constant, j is the current density in the solder, m is an exponent for current density, Q is
the activation energy for EM, k is Boltzmann’s constant, and T is the average temperature.
By adding the current factor and a constant temperature increment, Choi et al. obtained activation
energies of 0.5 and 0.8 eV for SnPb and SnAgCu solders, respectively [46]. Chae et al. also considered
the effect of Joule heating, and the activation energy calculated was 0.86–0.94 eV and the current density exponent 2.1–2.2 for SnAg solder joints [42]. Also, by virtue of a numerical simulation and temperature coefficient resistance method (TCR ¼ DR0R D1T , R0 is the resistance of the Al trace at T0, DR is the
resistance variation, and DT is the temperature rise), we deduced the c and the DT (depending on the
applied current and ambient temperature), respectively. The average activation energy obtained was
about 0.62 eV, and the current density exponent ranged from 1.46 to 1.89 [19]. Table 3 summarizes
the EM reliability parameters for Sn-based lead-free solder interconnects from accelerated life tests
in different research groups [19,34,42,46].
More recently, Chiang et al. have compared the predicted values based on Black’s equation, the predicted values based on the modified equation, and the measured MTTFs under test conditions of
441
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Table 3
Statistics of EM reliability parameters of lead-free solders [19,34,42,46].
Research group
Test conditions,
T (°C), j (A/cm2)
Activation energy,
Q (eV)
Current density
exponent, m
Univ. of Texas at Austin (Sn3.5Ag) [42]
115–150,
4.12–5.67 104
125–160,
2.75–3 104
125–165,
0.74–1.68 104
100–150,
1.0–2.0 104
0.86–0.94
2.1–2.2
0.8
–
0.88
2.11
0.58–0.66
1.46–1.89
Univ. of California, LA (SnAgCu) [46]
National Tsing Hua Univ. (Sn3Ag0.5Cu) [34]
EPA, City Univ. of HK (Sn3.5Ag1.0Cu) [19]
different current densities and temperatures. They have found that the deviation of predicted values
from the experimental results have been reduced based on the modified equation [34]. Indeed, further
effort is necessary to identify the actual current density and bump temperature in solder interconnects. Also, the physics of failure after accelerated life tests needs to be established to confirm its consistency with the proposed failure mechanism. Otherwise, the EM reliability would be incorrectly
evaluated.
3. Joule heating-enhanced dissolution of UBM and the diffusion of on-chip metal interconnects
3.1. Effect of Joule heating due to current stressing
In high current density packages, heat accumulation cannot be ignored since Joule heating is proportional to the square of the current density. As mentioned in Section 2.3, the current crowding effect
inevitably leads to a local temperature rise which in turn accelerates the nucleation and growth of
voids inside the solder joint. More significantly, as the foremost heat source, Joule heating from the
on-chip metal interconnects is of particular concern. This has been verified with thermal infrared measurements [47]. Fig. 11 shows the temperature distribution in a flip chip interconnect when stressed
Fig. 11. Thermal infrared measurement for the chip side, with the Al trace exhibiting the highest temperature [47].
442
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 12. Temperature distribution within a solder interconnect when stressed by 3.7 104 A/cm2 at 75 °C (hot region occurred
in the Al trace) [48].
by 104 A/cm2 at an ambient temperature of 70 °C. The temperature in the middle of the Al traces was
much higher than that at the circular Al pads. The edges of the UBM (marked point A) and the passivation openings (marked point B) also exhibited higher temperatures than the Al pads above the solder
joints.
Also, Fig. 12 shows a numerical simulation result of the temperature distribution within a solder
interconnect, including an Al trace, the half-bump solder and Cu conductor, when stressed by
3.7 104 A/cm2 at 75 °C [48]. It is apparent that the highest temperature occurred in the Al trace. Similar results were found by other research groups [47,49].
Since the Al trace is the dominant heat source together with local Joule heating inside the solder
itself, it is expected that hot spots should occur where the Al traces enter the solder joint. Near this
hot spot region, atomic diffusion will be thermally accelerated so that the UBM layer will be damaged.
Also, lattice diffusion of Al atoms will possibly be initiated because of the local high current density
itself. These mechanisms may be combined and will be discussed later.
3.2. Dissolution of UBM layers and possible solutions
The dissolution of a Cu UBM in a eutectic Sn37Pb solder joint under current stressing has been detected [50,51]. Under a current density of 103 A/cm2 at 150 °C for 0.5 h, the solder joint failed with an
open circuit, as one of the corners of the Cu UBM was dissolved and replaced by solder according to the
microstructural analysis. Hu and co-workers also reported the rapid, asymmetrical, and localized dissolution of a Cu UBM at the cathode side [52,53]. The average dissolution rate was 1 lm/min when the
current density through the eutectic Sn37Pb solder joint was 2.5 104 A/cm2 at 100 °C. From the location and geometry of the dissolved Cu, it is believed that current crowding played a critical role in this
rapid dissolution. When the current density was increased to 4 104 A/cm2, extensive dissolution of
the Cu UBM occurred even at an ambient temperature of 30 °C.
The rapid dissolution of Cu UBM is attributed to an interstitial diffusion of noble and near-noble
metals enhanced by Joule heating. It is well known that the interstitial diffusion of dilute elements
in tin is significant [54–56]. A series of fundamental studies on diffusion and EM of Cu, Ni, Ag and
Au in lead–tin alloys have been developed since the 1980s [57–59]. As the lattice constants of a
and b in tin are much larger than that of the c axis, the open structure along the c axis facilitates faster
interstitial diffusion than along the other orthogonal directions. Taking Ni for example, the diffusivity
of Ni along the tetragonal c axis is about 7 104 times than that at right angles at 120 °C [56], and the
EM is relatively very fast.
Therefore, it is not difficult to understand why Ni was consumed during EM experiments too,
although Ni is used as a diffusion barrier in UBM application. Fig. 13 illustrates the effect of EM on
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
443
Fig. 13. Elemental mapping at the UBM/IMC interface in a Sn3Ag1.5Cu solder joint after 1967 h under a current density of
1 104 A/cm2 at 150 °C [60].
Fig. 14. (a) Schematic diagram of a Cu pillar bump with a solder cap, and (b) Focused Ion Beam (FIB) image of Cu pillar bumps
with a height of 80 lm [3].
a multi-layer UBM film of Ti (0.2–0.5 lm)/Ni(V) (0.325 lm)/Cu (0.5–1.0 lm) [60]. It was found that,
after experiencing a downward electron flow, the Ni and Cu constituents in the UBM began to spread
into the solder, and the UBM was gradually consumed. In this case, voids formed at the UBM/IMC
interface due to UBM consumption under the combined effect of interstitial diffusion and large Joule
heating.
Recently, it has been proposed that a possible solution to the effects of EM in solder joints would be
a thick Cu pillar. The thick Cu pillar could be fabricated as the UBM, and a thin cap layer of solder
444
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
would be required for the bump, as shown in Fig. 14 [3]. An additional electroplated Ni layer has been
suggested to suppress Cu diffusion into the solder body, thus practically inhibiting IMC formation and
Kirkendall voiding [61]. It is expected that this will be effective for dealing with the problem of UBM
dissolution and void accumulation, since the thick Cu pillar is designed to spread the current from the
contact to an approximately uniform and low density. This proposal has been supported by experimental studies and numerical simulation [62,63]. However, a substantial formation of IMC at the
interface is becoming an issue. Also, TM may be initiated since a large temperature gradient is generated across the shallow solder interconnects.
It was noted that the rotation of b-Sn grains occurred in Sn-based solder under current stressing
because of their anisotropic properties [64,65]. This re-orientation resulted in a realignment of Sn
grains along with the current flow, thereby reducing the resistance of the solder. It is also known that
the diffusion of Ni/Cu in UBM was much enhanced along the c axis of Sn crystals, which contributed to
the dramatic dissolution of the UBM. Therefore, one should say that the orientation of Sn grains is
closely related to the reliability of Sn-base lead-free solders. Recently, Lu et al. have investigated
Fig. 15. Elemental mapping of the interface between a solder and the UBM (a) unfailed after 1000 h current stressing, and (b)
failed after 1711 h current stressing [68].
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
445
the effect of Ag in a Sn-based solder and concluded that the grain re-orientation of Sn was blocked due
to the presence of cyclic twinning and a stable Ag3Sn IMC network, in turn the dissolution of UBM was
comparatively mitigated [66]. They further explored the effect of an additional 0.6 wt.% Zn in Sn1.0Ag
solder and obtained a positive result [67]. The Zn doping stabilized the Ag3Sn and Cu6Sn5 IMC networks, and suppressed the formation of Cu3Sn IMC. More importantly, polycrystalline-like structures
were found to form at the solder/UBM interface. Although it seems that Zn doping could not control
the grain orientation in bulk solder, the strong binding with Cu effectively slowed down the Cu diffusion, and thus stabilized the solder microstructure. This is a creative study to explore the doping effect
on the microstructural evolution and thus the enhancement of EM resistance. A further nano-doping
into solder is also anticipated to support a higher current density and attenuate the EM damage.
3.3. Melting of solder interconnects due to Al diffusion
For flip chip solder joints with an Al/Ni(V)/Cu UBM, if the Ni layer is consumed completely the
adhesion of the UBM to the solder will be degraded. Also, Al diffusion in the Al trace will be triggered
as a result of the high current density and local heating. Liu and Lin reported an Al flux-induced failure
at the cathode side of a Sn97Pb and Sn37Pb composite solder joint with a downward electron flow
[68]. Fig. 15a shows that the location of the Ni layer matched well with that of the Cu layer. In the
downward electron flow case, the Ni(V) layer would be gradually consumed over a prolonged period
of time. As Fig. 15b shows, the Ni completely diffused into the solder and the V layer was also damaged. Furthermore, Al started to spread within the solder joint. EM and the accompanying Joule heating drove the Al away from the Al trace and pushed it into the solder. The diffusion of Al into a Sn3.5Ag
solder was also detected by Shao et al. [69]. They found that the solder filled in where a Ti/Cr–Cu/Cu
UBM had been located and some CuAl2 IMC formed in the region where the Al pad had been situated.
Using an infrared microscope Liang et al. detected the fracture of an Al trace while the current density through the Al trace was about 1.2 106 A/cm2 [70]. They speculated that EM damage had also
occurred in the Al trace, and that the degradation of the Al trace may be responsible for an abrupt temperature rise. Additionally, their thermoelectric simulations supported this. It was the degradation of
the Al trace, instead of void formation, that contributed to the formation of a hot spot.
It has been proposed that solder melting under current stressing is a time-dependent phenomenon
[28,71]. According to previous research, the principal reason for an incubation time was attributed to
the process of void generation and propagation, and solder lifetime was explained and approached
through modeling void accumulation [37,53]. However, Ouyang et al. have observed the melting of
eutectic Sn37Pb solder joints due to the Joule heating of the Al traces [71]. They suggested that Al dissolution expedited the rise of electrical resistance of a solder interconnect and hence led to the final
melting of the solder. Since the resistance change of the Al trace was dependent on the dissolution rate
of Al into the solder, an incubation period was required for a temperature rise which could provide
sufficient heat to melt the solder joint. In this way they explained why the melting of the solder exhibited a time-dependent characteristic.
Recently, the melting failure of Sn3.5Ag1.0Cu solder interconnects has been studied under a current density of 2.3 104 A/cm2 at 125 °C [72]. A new failure mechanism involving the combined effect
of solder EM and Al diffusion was proposed. Fig. 16 shows typical stages of the morphological evolution. Firstly, with a downward electron flow, voids occurred at the interface between the Cu–Sn IMC
layer and the solder, especially in the current crowding region (see Fig. 16a). Secondly, as Fig. 16b
shows, the voids gradually extended to the surrounding areas due to vacancy super-saturation.
Thirdly, the creation of pancake-type voids decreased the effective contact area, which led to more
serious current crowding. Meanwhile, the Joule heating due to current crowding was enhanced because of poor heat dissipation around the voids. Under such accumulated effects, the atomic diffusion
of Ni(V) in the UBM was accelerated, and the barrier that prevents the dissolution of Al into the solder
no longer existed. Therefore, the diffusion of Al was triggered and some voids were found in the Al pad,
as demonstrated in Fig. 16c. Also, from the local magnified micrograph shown in Fig. 16f, it is noticed
that the Ni(V) layer previously attached to the Al pad had disappeared as compared with Fig. 16e. Ni
atoms were dissolved and consumed to form a Cu–Ni–Sn ternary IMC, and the V layer above the voids
extruded and begun to lose its structural integrity, so that the dissolution of Al through this layer was
446
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 16. SEM images of different stages of the morphological evolution in Sn3.5Ag1.0Cu solder joints under a current density of
2.3 104 A/cm2 at 125 °C after (a) 92 h, 25% of the failure time, the time point A, (b) 245 h, 66%, B, (c) 295 h, 80%, C, (d) 361 h,
98%, D, (e) local magnified micrograph of the interface at the time point B (the dotted region), and (f) C [72].
more rapid. Fourthly, with the progress of Al dissolution, the EM in connecting Al trace was initiated
and expedited [19,72], so that further melting failure of solder interconnects was produced, as shown
in Fig. 16d.
A finite-element simulation was applied to understand the current density distribution in the flip
chip interconnects. Fig. 17 displays the evolution of the current density in Al interconnects alone (seen
from underneath). The current density reached more than 106 A/cm2, which is sufficiently high to
trigger the EM of the Al. According to Fig. 17a, the current density at the exit location of the Al pad
ranged from 1.2 1010 to 1.4 1010 A/m2 (i.e., from 1.2 106 to 1.4 106 A/cm2) before voids were
developed. The modeled maximum value occurred at the connecting corner of the Al pad and the Al
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
447
Fig. 17. Current density distribution in the Al interconnect alone (seen from underneath) (a) before void growth (the current
density at the exit location was 1.2 1010–1.4 1010 A/m2), and (b) after void growth (the current density at the exit location
was 1.5 1010–1.7 1010 A/m2).
trace. By contrast, when the voids propagated, the location of the maximum current density was
transferred to the exit location of the Al pad, and it reached 1.7 1010 A/m2 (i.e., 1.7 106 A/cm2),
as shown in Fig. 17b. This simulation indicates that the current density through the Al pad tended
to be enhanced due to the decrease of contacting area at the interface, and this result supports the
experiments well.
The total incubation time for melting the solder was found to be dependent on the rates of void
growth and Al diffusion in this case. Therefore, the solder melting exhibited a unique time-dependent
characteristic. In the initial stages, the rate of void growth varied from 0.24 to 0.53 lm/h. This was related to the void nucleation and propagation. In the later stage, before the final failure, the depletion of
the Al also exhibited a linear relationship with time, which was ascribed to the EM of the Al
interconnect.
It has been known that the change in the trace resistance is a linear function of the atomic drift
velocity [73]. In this case, the relationship between the rates of resistance change of the traces (o(D
R/R)/ot) and material depletion (o(DL)/ot) may be described as:
@ðDR=RÞ
qr SAl
1 @ðDLÞ @ðDLÞ
/
¼ td ;
1
@t
L @t
@t
qAl Sr
ð8Þ
where the subscripts r and Al refer to the under-layer and Al trace, respectively, q is the electrical
resistivity, S is the cross-sectional area of the specific layer, R is the initial trace resistance, L is the initial trace length, and md is the atomic drift velocity. Based on an electrical characteristic, the rate of
resistance change was estimated to be 0.9% h1. This rate of change then represents the drift of Al
atoms in the later stage.
448
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
4. Effect of current stressing on the formation of IMCs and kinetic analysis
The interactions at the anode and the cathode are different due to the polarity of the electric current. Generally, it is found that the growth of IMCs is enhanced at the anode while inhibited at the
cathode, resulting in an excessive growth of the IMC layer at the anode and a dissolution of the
IMC layer at the cathode under current stressing. Either of these effects may put the reliability of solder joints at risk and lead to the failure of the whole circuit in electronic devices.
4.1. Polarity effect of current stressing and enhanced growth of IMCs at the anode
Fig. 18 illustrates the typical interfacial evolution of Cu/Sn3.8Ag0.7Cu/Cu interconnects under current stressing [74]. Taking the IMC thickness as the total value (including both the Cu6Sn5 and Cu3Sn
layers), the thickness change as a function of time under various current densities at 120 °C has been
plotted, and is shown in Fig. 19. It is apparent that the growth rates of the IMC layer were very
different at the anode and the cathode. The IMC layer at the anode was always thicker than that at
Fig. 18. SEM images of IMC evolution at interfaces in Sn3.8Ag0.7Cu solder joints under a current density of 3.2 104 A/cm2
[74].
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
449
Fig. 19. Change of the IMC thickness with time under various current densities at 120 °C [74].
the cathode after the same time of current stressing. This result is in good agreement with the research
by Chen and co-workers in the Sn/Cu, Sn/Ag, and Sn/Ni systems [75–79]. Also, it is noted that the vacancy flux became supersaturated and condensed to form voids at the cathode, as shown in Fig. 18h.
Void formation at interfaces has already been covered in Section 2.3.
Since the occurrence of voids brings complexity into the analysis of the evolution of the IMC layer,
it is more reasonable to probe the kinetics that dominate the IMC growth at the anode. According to
Fig. 19, the growth of IMCs at the anode has a parabolic dependence on time since the square of thickness increased linearly with stressing time, which indicates that the formation of the IMC layer was
mainly controlled by a diffusion mechanism. It is suggested that a back stress resulting from the
EM, which will be discussed in detail in Section 5.1, was responsible for this growth behavior. As
Cu is the dominant diffusing species in Cu–Sn IMC formation [80], the IMC growth rate at the interface
(d(Dx)/dt) based on the transport of the Cu flux was expressed and further simplified to [74]:
DC g
dðDxÞ
D DrX
¼G D
Cg
GJ0em G0 J em
dt
kT Dx
Dx
a þ a
þ b ða ; b P 0; anode; a ; b 6 0; cathodeÞ;
¼
Dx
ð9Þ
where D is the diffusivity of Cu, Cg is the Cu concentration in the IMC, r is the stress, X is the atomic
volume, J 0em is the Cu EM flux in the IMC, Jem is the Cu EM flux in the solder, and G and G’ are known
constants related to Cu concentrations in the Cu anode, the IMC and the solder.
From Eq. (9), it is evident that the (a + a)/Dx term contributes to parabolic growth, while the b
term contributes to linear growth. Without electric current (a = 0, b = 0), the IMC growth would follow a parabolic growth law. In addition, since a is positive, the growth rate of the IMC layer should be
relatively large at the anode. This is in accord with the experimental results above.
The parabolic law for the growth of the total IMC layer is also applicable to the evolution of a single
IMC layer. In Sn/Cu/Sn/Cu/Sn sandwich-type structures, uniform layers of Cu6Sn5 and Cu3Sn were observed at the interfaces where electrons flew from the Sn side to the Cu side at 170 °C [78]. Moreover,
the growth of both the Cu6Sn5 and Cu3Sn IMC layers exhibited a parabolic trend with the passage of
time.
Xu et al. reported the effect of EM on the IMC growth in Ni/Sn3.8Ag0.7Cu/Ni interconnects at different ambient temperatures [81]. They also found that the growth of the Ni–Cu–Sn IMC layer at the
anode was faster than that at the cathode, and the growth at the interface had a parabolic dependence
450
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
on the stressing time. However, both were faster than that due to isothermal aging because of the
Joule heating effect, which is distinct from that in the cases discussed above.
By contrast, Gurov and Gusak developed a model where the electric current contributes to the linear term in the growth of the IMC layer, suggesting a reaction-controlled mechanism for EM [82]. They
suggested that the phase growth changes gradually from parabolic to linear at a constant rate when
the electric current favors the growth. An investigation by Chae et al. showed a similar result in Cu/
SnAg/Cu and Ni/SnAg/Ni interconnects under a current density of 5.2 104 A/cm2 [42]. Also, a kinetic
model was formulated to verify the linear growth of the IMC layer when an EM driving force dominates the chemical interdiffusion [83,84].
It is well known that without current stressing, the growth of the IMC layer follows a parabolic
behavior [85,86]. However, exactly how the current influences the IMC growth is obviously quite complicated from the results cited above. More attempts need to be made to probe the underlying kinetics
which dominate the growth of the IMC layer under current stressing.
4.2. Dynamic equilibrium of IMCs at the cathode
As shown in Fig. 18b, d, f and h, the IMC growth at the cathode is retarded, in comparison with that
at the anode. This growth is accompanied by the consumption of Cu from the cathode side. The following discussion will be developed in the light of the flux of Cu.
Due to a larger electron scattering cross section, the Cu atoms experience a larger EM force than the
Sn atoms. In addition, the Cu solute atoms in solder have a larger diffusivity compared to that in the
Cu–Sn IMC and in the Cu substrate [87]. As a consequence, EM fluxes of Cu in the IMC and in the Cu
substrate are insignificant and thus negligible. IMC dissolution, and the Cu EM flux in the solder become the two dominant mechanisms in the process of Cu consumption. It is well known that both
the IMC dissolution flux and the Cu EM flux in the solder are dependent on the temperature. Therefore,
a critical temperature (Tc) can be determined by balancing the two fluxes [88]:
C Sn
Do expðQ Cu=Solder =kTÞ Q
Z Cu=Solder eE ¼ C Cu A exp ;
kT
kT
ð10Þ
where CSn and CCu are the atomic concentrations of Sn and Cu, respectively, QCu/Solder and Q are the
activation energies for Cu diffusion in the solder and Cu consumption, respectively, Z Cu=Solder is the
effective charge number for Cu diffusion in the solder, A is a constant, E is the electric field, T is the
EM temperature.
If define:
Y ¼ C Sn Do Z Cu=Solder eE=C Cu kA:
ð11Þ
Then, Eq. (10) can be expressed as:
Q Cu=Solder
Y
Q
exp :
¼ exp T
kT
kT
ð12Þ
If QCu/Solder = Q, the critical temperature is equivalent to Y. If QCu/Solder – Q, then the critical temperature needs to be resolved based on Eq. (12). More importantly, this indicates that when the EM temperature is above the critical temperature, Cu consumption is controlled by the IMC dissolution
process and is typically linear with time. On the other hand, when the temperature is below the critical temperature, the EM flux of Cu in the solder is dominant, and thus the Cu consumption has a parabolic relationship with time. Under this condition, a constant Cu–Sn IMC thickness can be obtained
by balancing the Cu chemical flux and the Cu EM flux in the solder, which suggests that a dynamic
equilibrium state is reached [2,89]. Liu et al. observed the interfacial evolution at the cathode in Sn
bumps under a current density of about 5 103 A/cm2 at 155 °C, and noticed that the IMC layer thickness approached a certain value in the later stage of current stressing [88]. A systematic experimental
investigation of this dynamic equilibrium was conducted by Tu [2]. He found that with a further
reduction of the current density, the IMC layer thickness at the cathode was stabilized after a period
of 55 h stressing time in the case of 5 103 A/cm2, which is shown in Fig. 20.
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
451
Fig. 20. Change of Cu6Sn5 IMC thickness at the cathode in Sn3.8Ag0.7Cu solder joints under various current densities at 150 °C
(a plateau was found after 55 h in the case of 5 103 A/cm2) [2].
4.3. Abnormal polarity effect of current stressing on the formation of IMCs
As mentioned in Section 2.1, under current stressing, Pb atoms will migrate with the electron flow
and be accumulated substantially at the anode side in SnPb solder joints. It is suggested that due to the
obstruction by the Pb barrier, the IMC layer growth at the interface would exhibit an abnormal characteristic. The IMC evolution in SnPb solder under a current density of 6.2 102 A/cm2 at 125 °C was
investigated by Wu et al. [90]. However, since the applied current was relatively moderate, the retardant effect on the evolution of the IMC layer due to the segregated Pb layer was insignificant.
As eutectic SnBi has been considered as a preferential solder for low temperature applications, the
behavior of SnBi solder was studied to verify the effect of current stressing on the IMC formation in
this type of two-phase solder joints. Fig. 21 shows the typical interfacial evolution in Cu/Sn58Bi/Cu
interconnects under a current density of 5 103 A/cm2 after different stressing times at 75 °C [91].
It was found that the growth of the IMC layers at both sides (anode and cathode) was enhanced by
the electric current, and the IMC layer at the cathode grew faster than that at the anode. This abnormal
phenomenon is different from what was discussed above in Sections 4.1 and 4.2. Also, systematic EM
experiments on this solder system at several ambient temperatures (35 °C, 55 °C and 75 °C) were conducted to establish the kinetic parameters. It is interesting to note that although the growth behavior
at the two sides showed the abnormal polarity effect, the IMC layer thickness at the cathode followed
a parabolic growth law, as shown in Fig. 22.
According to an empirical equation on interfacial reactions during solid-state aging [92], the IMC
layer thickness as a function of time and temperature can be described by:
pffiffi
Q pffiffi
x ¼ d t þ xo ¼ Do exp
t þ xo ;
kT
ð13Þ
where x is the IMC layer thickness at aging time t, x0 is the initial IMC layer thickness, Do is the intrinsic diffusivity of Cu in the IMC layer, Q is the activation energy for Cu diffusion, and T is the
temperature.
Based on Eq. (13), the intrinsic diffusivity was estimated to be 9.9 105 m2/s. Also, the activation
energy for Cu diffusion in the Cu–Sn IMC was about 0.92 eV, which agrees well with those reported for
the growth of Cu–Sn IMC [85].
452
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 21. SEM images of the interfacial evolution in Cu/Sn58Bi/Cu interconnects under a current density of 5 103 A/cm2 at
75 °C for (a) 192 h at the anode side, (b) 192 h at the cathode side, (c) 384 h at the anode side, (d) 384 h at the cathode side, (e)
576 h at the anode side, and (f) 576 h at the cathode side [91].
A possible explanation for the abnormal polarity effect in this case is as follows. At the anode, the
growth of the IMC layer was initially enhanced by the current stressing. However, due to the Bi EMinduced back stress inside the solder, the diffusion of Cu and Sn to the anode was inhibited. From
Fig. 21, some Cu–Sn IMC precipitates were detected within the Bi-rich layer, which may be attributed
to a Cu flux in the opposite direction to the electron flow. Also, with a long period of EM time, the Birich layer became thick enough to block the transfer of Sn atoms. Hence, further development of the
IMC layer was limited due to the lack of a Sn source at the anode. On the other hand, at the cathode,
the electron flow was in the same direction as the Cu chemical diffusion, so that the atomic flux of Cu
into the IMC was accelerated. Meanwhile, the outward Cu flux was significantly inhibited due to the
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
453
Fig. 22. Variation of the Cu–Sn IMC thickness at the cathode side of Cu/Sn58Bi/Cu interconnects with the square root of time at
various temperatures [91].
back stress and the obstruction by the Bi layer. Therefore, the IMC layer at the cathode exhibited preferential growth as the inward atomic fluxes were larger than those flowing out.
A back stress-induced abnormal phenomenon was also observed in eutectic SnZn solder. Because of
its low melting point, excellent mechanical properties and low cost, SnZn solder has been developed
as a substitute for SnPb. Zhang et al. detected an abnormal interfacial variation in eutectic Sn9Zn/Cu
interconnects under a current density of 4.3 103 A/cm2 at 140 °C [93]. They found that the Cu–Zn
IMC layer at the cathode was about 2.3 lm thicker than that at the anode after 166 h. Also, upon
increasing the temperature to 185 °C, the Cu–Zn IMC was replaced by Cu–Sn IMC at the anode and
by a mixture of Cu–Zn and Cu–Sn IMCs at the cathode. They speculated that under the effect of the
EM-induced back stress gradient, more Zn atoms migrated to the cathode. Then the Sn atoms close
to the anode reacted with Cu atoms to form a new Cu–Sn IMC layer. At longer times, the Zn migration
to the cathode was accelerated so that the initial Cu–Zn IMC was completely replaced by a Cu–Sn IMC
layer. The investigation by Kuo and Lin supported this finding [94]. They noticed a similar abnormal
polarity effect of Cu–Zn IMC, i.e., the growth rate at the cathode was a little faster than that at the anode. However, the rationale behind the experimental findings should be further verified.
5. Stress-related degradation of solder interconnects under EM
5.1. Morphological evolution due to EM and a back stress in solder interconnects
When atoms are driven from the cathode to the anode by the electron wind force, the latter will be
in compression and the former in tension. In a cross-sectioned solder joint for an in situ observation, it
is expected that the compressive stress will be released from the free surface causing hillocks or whiskers to occur at the anode. Using thin film solder strips, Liu et al. first investigated the formation of
atomic hillocks in pure tin under a current density of 105 A/cm2 at room temperature [14]. As can
be seen in Fig. 23, with a prolonged current stressing, Sn grains were extruded out as hillocks. An
explanation was given in terms of a stress relief mechanism that a hillock or whisker grows from
the surface under compression [95].
Fig. 24 illustrates EM in the form of hillocks at the anode in a eutectic Sn37Pb solder joint under a
current density of 4.4 104 A/cm2 after 1050 h. Here again, the surface morphology became rough because of a non-uniform stress distribution. Interestingly, it can be seen that a dimple occurred at the
cathode side, while a bulge was formed at the anode side. This is because the accumulated stress had
454
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 23. Hillocks at the anode side of a pure Sn strip under a current density of 105 A/cm2 after 80 h at room temperature [14].
Fig. 24. Hillocks at the anode in a eutectic Sn37Pb solder joint under a current density of 4.4 104 A/cm2 after 1050 h.
been relaxed in the matrix by the dimple and bulge of the solder surface. In addition, the hillocks and
dimples were observed in Cu/Sn9Zn/Cu solder interconnects under current stressing [96]. The formation of hillocks in the middle of the bulk solder was ascribed to the compressive stress driven by EM.
The average growth rate of the hillocks was measured to be approximately 1.3 108 cm3/h.
Synchrotron X-ray microscopy has been applied to provide information regarding a depth profile
for the accumulated stresses [97,98]. Fig. 25 demonstrates the hillock and valley formation in a eutectic tin–lead solder joint under a current density of 104 A/cm2 after 72 h. The depth profile obtained
with confocal laser microscopy for the joint, which is shown next to the micrograph, indicates that
the maximum height of the hillocks near the anode was about 16 lm, and the depth of the valley near
the cathode was about 34 lm. It can also be seen that the surface features of the hillock region exhibited rows of striations with a spacing of several micrometres. These striations were on the side of the
hillock facing the anode, and initiated from the anode end then propagated in the direction opposite to
the electron flow during current stressing. Such markings can be considered as an indication of the
material that was being squeezed out as a result of the compressive stress.
Also, whiskers in thin film solder strips have been detected under prolonged current stressing
[14,99]. Lin et al. investigated Sn whisker growth under current densities ranging from 4.5 104 A/
cm2 to 3.6 105 A/cm2 [99]. It was found that a higher current density accelerated the growth of
Sn whiskers. In addition, the current crowding effect played an important role in the growth of whiskers as expected.
Recently, Ouyang et al. have reported the formation of whiskers in solder joints [100]. Fig. 26a and
b shows the growth of whiskers at the upper-right corner (anode side) in eutectic SnPb and SnAgCu
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
455
Fig. 25. Hillock and valley formation in a eutectic tin–lead solder joint under a current density of 104 A/cm2 after 72 h (the
depth profile is shown to the left) [98].
Fig. 26. Whisker growth at the anode (chip side) (a) a eutectic SnPb solder joint under a current density of 104 A/cm2 after 48 h,
and (b) SnAgCu solder joint under a current density of 1.4 104 A/cm2 after 248.5 h [100].
solder joints under a current density of above 104 A/cm2 after 48 h and 248.5 h, respectively. The composition of the whiskers was confirmed as 93 wt.% Sn by EDX. These whiskers were initiated from the
cracked surface at the chip side. When Pb atoms were pushed towards the anode, a compressive stress
on Sn-rich grains was produced and then Sn whiskers were squeezed out. Moreover, it was noticed
that the cross-sectioned surface of the SnPb solder exhibited a dimple and bulge structure after EM,
while the surface of SnAgCu solder remained flat. This phenomenon suggests that the rate of EM in
SnAgCu was smaller than that in SnPb.
456
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
On the basis of the Nabarro-Herring model of the equilibrium vacancy concentration, more vacancies are generated in the tensile region, while less vacancies occur in the compressive region, so there
is a vacancy concentration gradient decreasing from the cathode to the anode [101]. The atomic flux
under a combined electrical and mechanical force can be expressed as:
J¼C
D D Xdr
Z eqj C
;
kT
kT dx
ð14Þ
where r is the hydrostatic stress, dr/dx is the stress gradient, and X is the atomic volume. The other
terms have been defined before. The first part represents the flux due to EM, whereas the second part
stands for the opposite flux due to back stress.
A plausible explanation for the above difference between the rates of EM of SnAgCu and SnPb is
then given in terms of the back stress [100]. Since the elastic modulus or stiffness of the SnAgCu solder
is larger than that of the SnPb, the back stress gradient in SnAgCu could be higher. Hence, the effect of
the back stress on the retardation of EM was relatively larger for the SnAgCu solder.
If the stress balances with the wind force at a critical length, there should be no net atomic flux
(J = 0), which has been well known as the Blech condition [102]. According to Eq. (14), the critical
length (Xc) can be obtained:
Xc ¼
rc X
Z eqj
ð15Þ
:
The effect of back stress and the critical length in a solder was investigated by Wei and Chen [103].
Eutectic SnPb solder strips with lengths ranging from 5 lm to 200 lm were prepared for the study and
a length-dependent EM behavior was identified. Fig. 27 shows the microstructural characteristics of
various solder strips under a current density of 2 104 A/cm2 after 490 h at 100 °C. No material depletion or voids could be detected for the 5 lm and 10 lm long strips. By taking the critical compressive
yield stress (rc) of SnPb solder (27 MPa), the critical length was estimated to be 11 lm under such
experimental conditions. This value agrees well with the experimental results.
The effect of the back stress was further studied by an area array of nanoindentation markers on
the cross-section of solder joints by Xu et al. [22]. Most markers moved against the EM-induced atomic
flux, indicating that the effect of the electron wind force was larger than that of the back stress in their
case. After 360 h of current stressing, the average marker movement from the cathode to the anode
was plotted and is shown in Fig. 28a.
Also, the atomic flux can be calculated as follows:
J¼
V
XðStÞ
¼
u
Xt
;
ð16Þ
where V is the total volume of atomic transport, u is the marker displacement, X is the atomic volume,
S is the cross-sectional area, and t is the operation time.
Fig. 27. Microstructural characteristics of various solder strips under a current density of 2 104 A/cm2 after 490 h at 100 °C
(no material depletion could be detected for the 5 lm and 10 lm long strips) [103].
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
457
Fig. 28. (a) Marker movement at different locations in a solder joint under current stressing after 360 h, and (b) the stress
gradient as a function of the location [22].
By combining Eqs. (14) and (16), and assuming that the effect of the back stress gradient on
the marker movement could be neglected when the marker was far enough from the anode, the stress
gradient as a function of marker displacement can be described as (when defining K = (1/C)(kt/D)(1/
X2t)):
dr
¼ Kðuo uÞ;
dx
ð17Þ
where uo is the marker displacement near the cathode, and K is a constant for a given temperature and
time.
Hence, the stress at any location is:
r¼
Z
x
Kðuo uÞdx:
ð18Þ
0
Substituting u(x) and the boundary conditions into Eq. (18), the stress gradient at any location
could be determined, which is shown in Fig. 28b. The stress gradient near the anode was 97 kPa/
lm, and it decreased gradually to zero with distance.
In addition, the damage under the combined effect of the electron wind force and the thermomechanical stress has been investigated using a coupled-field simulation [104]. It has been found that
a substantial thermal stress accumulated around the interface at the chip side, especially in the Ni
UBM and (Ni,Cu)3Sn4 IMC layers. As shown in Fig. 29a, the maximum stress was 138 MPa. Also, the
stress variation in direction y with the distance from point A is plotted in Fig. 29b, where a stress
gradient of 1.67 1013 Pa/m was found. Therefore, the thermal mismatch-induced force for Ni atomic
diffusion was estimated to be 1.82 1016 N, which was comparable to the electron wind force
(2.82 1016 N) in this case. At the initial stage, if the stress is not released effectively, stress migration would interact with EM and influence the reliability of solder joints, which needs more experimental studies.
458
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 29. Thermal stress distribution in Ni UBM and (Ni,Cu)3Sn4 layers (at the chip side) (a) stress distribution, and (b) variation
of the stress with the distance from point A in the direction y [104].
Fig. 30. U field fringe of a Sn4Ag0.5Cu solder joint (a) during current stressing, and (b) after the current was terminated [105].
5.2. Mechanical deformation and degradation under current stressing
To detect the EM-induced mechanical damage, a Moiré interferometric technique was used to obtain the in situ displacement evolution of solder joints under electric current stressing [105]. Large
deformations may be observed in solder joints under a current density of 104 A/cm2. Figs. 30a and
31a display the U field and V field fringes in a Sn4Ag0.5Cu solder joint after 1500 h of current stressing,
Fig. 31. V field fringe of a Sn4Ag0.5Cu solder joint (a) during current stressing, and (b) after the current was terminated [105].
459
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
respectively. The U field fringes were predominantly in the vertical direction with concentrations on
both vertical edges, indicating that a large normal deformation was developed in the horizontal
direction. Instead, the V field fringes were predominantly in the horizontal direction suggesting a large
62
anodic joint
cathodic joint
60
58
Modulus (GPa)
56
54
52
50
48
46
44
42
0
200
400
600
800
1000
1200
Time (h)
Fig. 32. Variation of modulus of Sn3.5Ag1.0Cu solder joints under a current density of 2.0 104 A/cm2 after different times at
125 °C.
Fig. 33. SEM images of solder joints after mechanical shear testing under a current density of 2.55 104 A/cm2 after 10 h at
140 °C (a) chip side, and (b) substrate side [106].
460
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
normal deformation in the vertical direction. Also, Figs. 30b and 31b show the field fringes after the
current was switched off. Although the fringes became less clear, there were little changes in both
U and V field fringes. This means that the deformations created by the high current density were irreversible, which is attributed to the re-arrangement of defects and atoms in the material and accompanying local volumetric change.
Nano-indentation tests were conducted to explore the mechanical behavior of solder joints after
EM [19]. Fig. 32 illustrates the variation of modulus of Sn3.5Ag1.0Cu solder joints under a current density of 2.0 104 A/cm2 after different times at 125 °C. It is apparent that the modulus of solder joints
tended to decrease after current stressing (except for that of the cathodic joint after 305 h), and the
mechanical properties were degraded as compared with the original values. When interfacial voids
are initiated due to flux divergence under EM, bond damage occurs in solder joints. From a physical
Fig. 34. Typical fractographs of Sn3.5Ag1.0Cu solder joints (substrate side) (a) as reflowed, (b) under current stressing after
144 h, and (c) under current stressing after 288 h [48].
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
461
perspective, the modulus is directly related to the atomic bonds. Hence, it is understandable that the
modulus decreased with the passage of time under current stressing.
The effect of EM on the shear behavior of flip chip solder joints was studied by Nah et al. [106]. It
was found that the mode of shear failure changed after EM and depended on the direction of electron
flow. Originally, shear-induced fracture occurred in the bulk of the solder without current stressing.
However, under a current density of 2.55 104 A/cm2 after 10 h at 140 °C, as shown in Fig. 33, fracture
occurred instead at the cathode interfaces between the solder and IMCs. This is because EM dissolved
and drove Cu or Ni atoms from the UBM or bond pad into the solder, and this resulted in the large
growth of brittle Sn-based IMCs at the cathode side. Therefore, shear failure occurred predominantly
at the cathode interface.
Also, fractographs of solder joints before and after EM were examined for a comparison [24]. Fig. 34
shows the typical sheared fracture surfaces of Sn3.5Ag1.0Cu solder joints (substrate side) under a
current density of 2.1 104 A/cm2 at room temperature. For the case without current stressing, the
fracture mode was in the bulk solder cutting through the region just near the (Cu,Ni)6Sn5 IMC layer,
and the fracture surface exhibited large amounts of ductile deformation with big dimples. At longer
stressing times, the interface became brittle and less plastic deformation was observed. This
mechanical deterioration with stressing time is attributed to void formation and stress accumulation
at the interface.
In addition, Ren et al. examined the tensile behavior of Cu/Sn3.8Ag0.7Cu/Cu solder interconnects
under current densities ranging from 1 103 A/cm2 to 5 103 A/cm2 at 100 °C [107]. They observed
the ductile-to-brittle transition in which the fracture migrated from the middle to the cathode interface of solder joints with increasing current density. Fig. 35 illustrates this movement of rupture position under a creep stress of 7 MPa. Likewise, the transition is explained by the polarity effect of EM,
Fig. 35. SEM images of Cu/Sn3.8Ag0.7Cu/Cu solder interconnects after tensile testing (a) without EM, (b) under a current
density of 3.3 103 A/cm2, and (c) under a current density of 5 103 A/cm2 [107].
462
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
especially the accumulation of vacancies at the cathode interface. As a whole, these EM-induced
mechanical property changes of flip chip solder joints would strongly impact on the reliability of flip
chip technology.
6. TM behavior in solder interconnects under a thermal gradient
With the trend towards greater integration and further miniaturization in the microelectronics
industry, the cross-sectional area of conductive lines on chips has been decreased significantly. This
has led to a dramatic accumulation of Joule heating in first-level interconnects, as discussed in Section
3.1. Then a considerable thermal gradient will possibly be built up across solder joints, which can provide a driving force for atomic diffusion to trigger TM.
6.1. TM in tin–lead solder interconnects
The earliest report regarding the combined effects of EM and TM in solder joints was given by Ye
et al. [108]. With microstructural observations and marker measurements, they found that TM in flip
chip solder joints may assist or counter EM, depending on the direction of the thermal gradient and
electric field.
The individual contribution of TM to the failure of solder joints was described by Huang et al. [109].
They proposed the design of a test structure of flip chip solder joints which can be applied to conduct
TM without EM. Generally, in the interconnect structure, the Al traces are the primary heat sources
because of their larger resistance. Hence, it is believed that a certain thermal gradient exists in the
powered solder joints due to the temperature difference. Moreover, owing to the excellent thermal
conductivity of the silicon chip, a similar thermal gradient is also formed across the adjacent un-powered joints. These un-powered solder joints are thus investigated for a TM study since no current is
applied to them.
Fig. 36 shows the typical TM phenomenon of tin–lead composite solder joints (Sn97Pb and Sn37Pb)
under a temperature gradient of above 1000 °C/cm after 5 h at 150 °C. The arrows indicate the direction of the electron flow. According to the cross-sectional observations after current stressing, the effect of TM was clearly visible across the un-powered solder joints, since in both of them Sn (dark
regions) moved to the chip side, the higher temperature end, and Pb (light regions) to the substrate
side, the lower temperature end. The advantage of using a composite solder is that there is an inhomogeneous compositional distribution in such samples, and the phase redistribution by TM can be
easily recognized. Also, this set of samples was used to conduct experiments in situ by detecting
changes on the cross-sectioned surfaces directly during current stressing. Likewise, the phase redistribution of Sn and Pb was recognized in the neighboring un-powered solder joints [109].
Morphological evolution due to TM has also been detected in eutectic tin–lead solder joints.
Fig. 37a shows SEM images of a row of tin–lead solder joints after 50 h at 150 °C. Fig. 37b demonstrates the detailed microstructure of joint 4 at a higher magnification [110]. According to these
micrographs, it is believed that Pb migrated to the substrate side under the temperature gradient
across the un-powered solder joints. This was supported by the EDX analysis of local regions. As
Fig. 36. SEM images of TM in un-powered tin–lead composite solder joints (Sn97Pb and Sn37Pb) under a temperature gradient
after 5 h at 150 °C [109].
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
463
Fig. 37. (a) SEM images of a row of solder joints from 1 to 12, with solder joints from 5 to 8 under current stressing after 50 h at
150 °C (Pb accumulation in the un-powered solder joints 4 and 9), and (b) the detailed microstructure of joint 4 at a higher
magnification [110].
shown in Fig. 37b, the average concentration of accumulated Pb at the substrate side was about
65.2 at.%, and the concentration of Sn at the chip side approached 86.3 at.%. The width of the accumulated Pb-rich phase band reached approximately 15 lm, i.e., half of the joint standing height. This result agrees well with TM in tin–lead composite flip chip solder joints [109].
A possible explanation for the above is as follows. The flow of atoms under a thermal gradient depends on the heat of transport (Q), defined as the difference between the heat carried by the moving
atoms and the heat of the atoms in the initial state (the hot end or the cold end) [111]. For the atoms
which move from the hot end to the cold end, the Q is negative since they lose heat. For atoms moving
from the cold to the hot end, then the Q is positive. Pb atoms are the dominant diffusing species with
a higher diffusivity in eutectic tin–lead solder above 120 °C [10,112]. Therefore, based on the microstructural evolution as shown in Figs. 36 and 37, it is speculated that with a negative Q, Pb atoms
migrated from the higher temperature side to the lower temperature side under the temperature
gradient. Meanwhile, Sn atoms moved slowly and replenished the vacancies due to the depletion of
Pb atoms. Macroscopically, the Pb-rich phase migrated to one side and the Sn-rich phase was
‘‘pushed” towards the opposite side on the basis of a constant volume process. However, the
464
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
mechanism of the reversed Sn flux during TM remains unclear, and the sign of Q for Sn cannot be confirmed as yet.
It is important to obtain the temperature distribution of first-level solder interconnects to understand the TM behavior of solder joints. Finite element modeling and simulation were applied to predict the electrothermal characteristics in a TM study. Fig. 38a shows the temperature distribution of a
flip chip test structure with a 1.8 A current applied at 150 °C [110]. A temperature difference between
the chip side and the substrate side was visible. Significantly, Fig. 38b describes the detailed temperature distribution of an un-powered joint in the interconnect, where there existed a temperature difference above 3 °C between the chip side and the substrate side. This means that a temperature
gradient greater than 1000 °C/cm (3 °C/30 lm 1000 °C/cm) was built up across the un-powered
joint due to the Joule heating from the neighboring Al traces. A simulation by Ye et al. also showed
the existence of a thermal gradient of about 1500 °C/cm in their flip chip test structures [108]. In combination with the simulation, thermocouples, and the temperature coefficient of resistance method
were used to verify real temperatures in solder interconnects [31]. However, due to the unique interconnect structure and the limitation of these measurement methods, it is difficult to pinpoint the
Fig. 38. (a) Temperature distribution of a quarter of the flip chip test structure, and (b) an un-powered solder joint (joint 9)
[110].
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
465
temperature gradient across solder joints. Recently, a thermal infrared technique has been employed
by Hsiao and Chen to obtain the thermal gradient directly in cross-sectioned solder joints [113]. As can
be seen from Fig. 39a, a uniform temperature distribution occurred in the solder bump before current
stressing. Then Fig. 39b shows the temperature distribution of a solder joint under an alternating current density of 9.2 104 A/cm2. Since there is no EM effect under alternating current stressing, and the
alternating current produces a similar Joule heating as does the direct current, the alternating current
was applied to independently investigate the TM behavior in solder joints. Fig. 39c shows the temperature profile along the dashed line in Fig. 39b, in which the average temperature at the chip side was
16.0 °C higher than that at the substrate side. The thermal gradient was calculated to be approximately 2143 °C/cm. This trial is significant since it verified the existence of a large thermal gradient
across real flip chip solder joints with experimental data.
Fig. 39. (a) Temperature distribution of a solder joint before current stressing, (b) under an alternating current density of
9.2 104 A/cm2, and (c) the temperature profile along the dashed line in (b) (the thermal gradient was estimated to be 2143 °C/
cm) [113].
466
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Also, from Fig. 37b it is noticeable that the Pb-rich phase accumulated at the lower left side, i.e., the
lower temperature region of solder joint 4. Likewise, the Pb redistribution in solder joint 9 showed a
similar tendency, i.e., Pb migrated to the lower right side (the lower temperature region), as Fig. 37a
shows. Taking solder joint 4 for example, since its right side was closer to the heat source, it is possible
that a temperature gradient was established laterally from the right side to the left side. Thus the Pbrich phase not only migrated to the substrate side under the vertical temperature gradient, but also
moved to the lower temperature region driven by the lateral temperature gradient across this solder
joint. This lateral TM was also observed in composite solder joints [109,114]. As Fig. 36 shows, in the
un-powered joints near to the powered joints, e.g., joint 9, it is evident that the Sn redistribution was
tilted towards the powered joints.
Morphological variations at different cross-sectional planes of a solder joint during TM are involved
because of different thermal dissipations [31]. Fig. 40a shows the obvious TM of Pb at the periphery of
two solder joints, which is similar to that in Fig. 37. Then a stepwise cross-sectional analysis was conducted by gradually grinding the solder joints to the center of the passivation opening. Fig. 40b demonstrates the cross sections of solder joints 4 and 9 after re-polishing by about 50 lm. It is noted that
the TM of Pb was not as apparent as that in Fig. 40a. Pb-rich phases were uniformly distributed in solder joint 4. Pb accumulation in solder joint 9 was also slight and only the Pb-rich phase at the periphery of the solder exhibited a TM characteristic. This means that TM of the inner solder region was less
significant than that of the outer solder (the surface layer). One can understand that the temperature
distribution inside the center of a solder joint was more uniform. By contrast, it is easier to build up a
large thermal gradient across the surface layer of solder joints where a substantial heat dissipation is
achieved, since the outer solder is close to the ambient environment.
Fig. 40. SEM images of two cross-sectional planes for un-powered solder joints 4 and 9 (a) after the first polishing (outer
solder), and (b) after re-polishing 50 lm or so (inner solder) [31].
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
467
Fig. 41. (a) SEM image of TM in a eutectic tin–lead solder joint under a thermal gradient after 27 h at 100 °C, and (b) local
magnified micrograph (the finer lamellar microstructure occurred after TM) [116].
It is worth mentioning that during the TM, the Pb grains were even more uniformly dispersed in the
tin-matrix, although the bulk of the Pb had moved to the substrate side, as shown in Fig. 37b. This
means that the lamellar microstructure became much finer after the TM process. This has also been
detected in Sn58Bi solder joints under a TM-enhanced effect in our group [115]. Ouyang et al. found
a similar phenomenon, which is shown in Fig. 41 [116]. They suggested that the formation of this finer
lamellar structure created a more disordered higher entropy state [2,116]. Also, according to their estimate, entropy production by heat propagation was many orders of magnitude larger than that by
atomic migration, and it is thus conceivable that entropy production in TM could affect the microstructure substantially.
In order to understand the mechanism of atomic transport, the atomic flux and the heat of transport during the TM process were estimated. Taking a central displacement (Dx) of 7.5 lm in Fig. 37b,
the total volume of atomic transport (Vtm) during the operation time (t) can be approximately obtained
from the product of the displacement and the cross-sectional area (S) of the solder joint. Therefore,
taking the atomic volume of Pb (X) as 3.0 1023 cm3, the atomic flux of Pb due to TM (Jtm) can be
calculated as follows:
J tm ¼
V tm
A Dx
7:5 104
1:4 1014 ðatoms=cm2 sÞ:
¼
¼
XðStÞ XðStÞ 3:0 1023 50 3600
ð19Þ
Also, the atomic flux due to TM can be expressed as Eq. (2), which has been given in Section 1.2.
Hence, taking a predicted temperature gradient of 1100 °C/cm, an atomic diffusivity of Pb of
4.0 1013 cm2/s [10], we substituted these values into Eq. (2), and obtained the molar heat of transport as about 27.2 kJ/mole. Compared to the result reported by Ouyang et al. (25.3 kJ/mole) [116], the
molar heat of transport of Pb is slightly different.
In addition, Chuang and Liu estimated the molar heat of transport of Pb as 22.2 kJ/mole under a
thermal gradient of 1010 °C/cm, by measuring the displacement of artificial markers [117]. Significantly, they found that the average displacement of atoms increased almost linearly with time during
TM. More recently, markers fabricated by a FIB method have been employed to measure the rate of TM
by Hsiao and Chen [113]. With a thermal gradient of 2143 °C/cm, a molar heat of transport of 26.8 kJ/
mole has been obtained for the transport of Pb.
An enhancement effect of TM on the phase coarsening of eutectic tin–lead solder has also been
investigated [118]. A possible explanation is that the inward atomic flux of Pb across the thermal gradient was larger than the outward atomic flux for any particular Pb-rich phase particles, and this accumulation of the atomic flux in the Pb phase was the major contributor to the enhancement of
coarsening.
468
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 42. (a) Original micrograph of a Sn3.5Ag solder joint and (b) micrograph of a solder joint after 800 h of TM testing with a
temperature gradient of 2829 °C/cm at 100 °C (note that the Sn whiskers were present at the chip side, and the markers moved
toward the substrate side) [119].
6.2. TM in Sn-based lead-free solder interconnects
As stated above, the migration of the Sn flux during TM has been unclear and has necessitated further investigations for lead-free solders.
A more recent study by Hsiao and Chen reveals the TM characteristic of Sn in lead-free solder [119].
They investigated the TM behavior of Sn in Sn3.5Ag solder joints under a temperature gradient of
2829 °C/cm at 100 °C. As mentioned in Section 6.1, an alternating current was used to eliminate the
EM effect, thus facilitating an independent study of TM. After 800 h of TM testing with a 0.57 A alternating current, it is interesting to find that hillocks were pushed out from the chip side, as shown in
Fig. 42b. These hillocks were generated by the mass transfer of the Sn at the hot side, providing direct
evidence that Sn was transported along the direction opposite to the thermal gradient. In addition, by
measuring the marker movement, they obtained the TM flux and molar heat of transport of Sn as
5.0 1012 atoms/cm2 s and 1.36 kJ/mole, respectively. Our studies also show the similar tendency that
Sn atoms migrate towards the higher temperature side in Sn3.0Ag0.5Cu solder joints under a thermal
gradient [19,120].
A TM of Cu atoms has been proposed. The microstructural evolution in Cu/Sn4Ag0.5Cu/Cu solder
interconnects has been studied under a thermal gradient of 1000–1200 °C/cm [121,122]. It has been
found that the two major microstructural differences between TM and isothermal samples were the
lack of a Cu3Sn layer at both the higher and lower temperature sides, and the thinning of the Cu6Sn5
layer at the higher temperature side for the TM samples. Supposedly, this thinning of the Cu6Sn5 layer
was due to its disintegration, during which the Cu atoms moved to the lower temperature side under
the thermal gradient. Meanwhile, the absence of the Cu3Sn layer was a result of an insufficient Cu concentration. More recently, the TM of interstitial Cu in SnAg flip chip solder joints has been reported by
Chen et al. [123]. It has been suggested that the void formation at the chip side was attributed to a fast
interstitial diffusion of Cu atoms from the UBM into the Sn matrix. The driving force of Cu diffusion
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
469
Fig. 43. (a) Schematic diagram of a line-type test structure (Ni/Sn58Bi/Cu), (b) temperature distribution in the test structure,
and (c) temperature distribution in the solder joint only (a thermal gradient of about 527 °C/cm existed across the solder joint)
[115].
was due to a large thermal gradient accredited to Joule heating across the solder bumps. Further attempts need to be made to verify the real characteristic of the TM of Cu.
More than that, a specific line-type test structure (Ni/Sn58Bi/Cu) has been applied to investigate
the combined effect of EM and TM of Bi, as shown in Fig. 43a [115]. As Ni shows a higher electrical
resistivity than Cu, a large temperature difference may be created at two sides of the solder joint during the current stressing (downward from the Ni to the Cu sides). By finite-element simulation, it was
shown that a thermal gradient of about 527 °C/cm existed in the solder joint under a current density of
5 103 A/cm2 at 50 °C, as demonstrated by Fig. 43b and c. Temperature measurements using thermocouples also supported this. By varying the direction of the electrical current, the counteracted and
enhanced effects by TM were detected separately. It can be seen from Fig. 44 that the migration of
Bi atoms was more pronounced when the Ni wire was used as the cathode. According to the experimental findings, we speculate that the Bi has a similar TM characteristic to Pb and shows a negative Q.
Then, if the direction of the thermal gradient was opposite to that of the electron flow, the TM counteracted the EM and retarded the diffusion of the Bi atoms (case 1). Otherwise, the TM assisted the EM,
and the diffusion of the Bi atoms was enhanced (case 2). In addition, based on the results from these
two cases, the atomic fluxes of Bi due to EM and TM were differentiated and estimated to be
1.48 1013 atoms/cm2 s and 5.38 1012 atoms/cm2 s, respectively.
470
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
Fig. 44. SEM images of Ni/Sn58Bi/Cu solder joints under a current density of 5 103 A/cm2 after 384 h at 50 °C (a) case 1: EM
counteracted by TM, (b) anode side in case 1, (c) cathode side in case 1, (d) case 2: EM enhanced by TM, (e) anode side in case 2,
and (f) cathode side in case 2 [115].
7. Concluding remarks
7.1. Summary
In this review, we have discussed five types of physical failure mechanisms of solder interconnects
for high current density applications. Phase separation and void growth, the essential physical processes occurring in EM, were introduced in Section 2. The pronounced effect of current crowding on
EM damage is worthy of attention. Also, EM reliability parameters for Sn-based lead-free solder interconnects in recent publications were collected for summary. One should say that the modification of
Black’s model is significant, which exerts a large effect on the lifetime statistics and reliability evaluation for EM failure.
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
471
Joule heating in first-level solder interconnects is substantial. This was demonstrated in Section 3
through experimental and numerical investigations. Due to interstitial diffusion enhanced by Joule
heating, the consumption of the UBM layer is noticeable so that new structural design (e.g., Cu/Ni pillar) and microstructural improvement need to be developed to support high current densities and
minimize EM. In addition, the Al diffusion-induced damage in flip chip interconnects has attracted
some interest. We proposed a failure mechanism involving the combined effect of solder EM and Al
diffusion, and thus offered an explanation for the time-dependent behavior of melting failure of the
solder under current stressing.
Vacancies are provided by the flux opposite to that of the lattice diffusion of atoms, but also come
from interfacial diffusion. As a result, another physical mechanism, interfacial reactions, was outlined
in Section 4. Due to the polarity effect of the electrical current, the formation of IMC layers is enhanced
at the anode while retarded at the cathode. The IMC layer at the anode grows preferentially with prolonged stressing time. On the other hand, IMC formation at the cathode can reach a dynamic equilibrium when the electron wind force is comparable to the chemical force, and the Cu consumption is
predominated by the EM process instead of dissolution. In addition, some abnormal polarity phenomena which were present in SnBi and SnZn solders have been summarized and included for comparison.
We reviewed the stress-related degradation of solder interconnects under a current density in Section 5. Due to the relief of compressive stresses, morphological evolution is apparent in the form of
hillocks or whiskers near the anode. One important factor, the back stress generated, was investigated
to understand the damage due to EM. Also, the mechanical deformation was identified through an
interferometric technique. In addition, the deterioration of solder interconnects under current stressing was detected through a series of mechanical tests.
An attempt to explore the TM behavior of solder interconnects has been made recently, and this
was covered in Section 6. By employing a testing method of differentiating TM from EM, the TM
behavior of Pb was understood in terms of morphological evolution, atomic transport and by numerical simulation. Pb shows a negative heat of transport. The TM of Sn has also been studied. On the basis
of experimental findings it is speculated that Sn atoms exhibit a different TM characteristic opposite to
that of Pb atoms. More recently, the TM of interstitial Cu has been reported, which states that the Cu
also has a negative heat of transport. In addition, a specific line-type test structure has been utilized to
investigate the combined effect of EM and TM of Bi. It was revealed that Bi has a similar TM behavior
to that of Pb.
7.2. Further problems which need to be addressed in the near future
The current carrying capability of Al/UBM/solder should be considered based on the limitation due
to EM in the design rules. Routing design of Al interconnects needs to be implemented to mitigate the
current crowding and Joule heating. A pad structure that produces a uniform current distribution
within the bump interconnect is recommended to avoid the dissolution of Al. In the case with EM
problems of Al, a relative enlargement of the cross-section of the Al trace is also a factor that could
be considered. For UBM, in Section 3.2, the solution of a thick Cu pillar with a Ni electroplated layer
has been suggested, which can be expected to alleviate the effect of current crowding and accompanying heat accumulation at the interface. For solders, new candidates with improved current density
capabilities and a higher operating temperature are expected. Considering the Sn-grain rotation under
current stressing, a further nano-doping into solder is anticipated to stabilize the microstructure as to
limit the fast interstitial diffusion of noble or near-noble metals. In addition, taking into account different applications in industry, EM studies in Sn-based solders should be extended to other lead-free
solders, such as InAu.
Although void formation at the UBM side has been confirmed as a major failure mechanism in solder interconnects for high current density applications, IMC growth cannot be ignored, which has been
discussed in Section 4. Moreover, as mentioned in Section 5.2, the extensive IMC growth at the substrate side exerts a great influence on the mechanical reliability. It is known that the kinetics dominating interfacial reactions have not been established yet, and the laws of IMC growth (parabolic or
linear) under current stressing are still under investigation. Therefore, the interfacial reactions under
current stressing are important and challenging problems, particularly for the application of micro-
472
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
bumps with Cu pillars in 3D packages, such as TSV (through silicon via) bonding, where the IMCs substantially form at the interface and hence a major phase transformation is induced [124,125].
Another issue of key importance is the TM of Sn. Although there have been a few studies made recently, the present understanding of the TM behavior of Sn is still limited. If one need differentiate TM
from EM further, it is suggested that each end of the connecting wires (same materials) in a line-type
test structure should be put at different temperatures [126]. In this way, the effect of current stressing
is completely removed and only the driving force due to TM is available. As compared with flip chip
samples, the advantages of line-type samples lie in the fact that the temperature gradient can be measured more easily.
So far, few studies on the combined effect of TM and mechanical stress have been reported [127].
For the mechanical characteristics, it should be mentioned that high strain rate fracture failure is as
important as low cycle fatigue failure, which has been of major concern in recent years. Shear tests
and tensile tests should be performed to evaluate the mechanical behavior of solder joints after TM.
If atoms migrate from the higher to the lower temperature side by the force due to the thermal gradient, a reversed flux of vacancies will move to the higher temperature side. Consequently, the interface at the higher temperature side will become mechanically degraded. This interaction between TM
and the applied stress is of considerable importance.
Acknowledgements
We would like to acknowledge Dr. M.O. Alam, Dr. B.Y. Wu and Dr. X. Gu at the EPA Centre, City University of Hong Kong. Without their significant contributions, the overview cannot be finished. Special
thanks are extended to Prof. B. Ralph at Brunel University, UK, and Dr. J. Shen at the EPA Centre, CityU,
HK, for their valuable suggestions.
We also wish to express our sincere gratitude to Prof. K.N. Tu’s group at University of California at
Los Angeles (Dr. H. Gan, Dr. C.Y. Liu, Dr. J.W. Nah, Dr. F. Ren, Dr. A.T. Huang, Dr. F.Y. Ouyang), Prof. C.
Basaran’s group at State University of New York at Buffalo (Dr. H. Ye), Prof. C. Chen’s group at National
Chiao Tung University (Dr. S.H. Chiu, Dr. C.C. Wei, Dr. H.Y. Hsiao), Prof. A. Lee at Michigan State University, Prof. M.H. Ren at National Sun Yat-sen University, Dr. K.L. Lin’s group at National Cheng Kung
University (Dr. Y.H. Liu), Dr. L.H. Xu at Intel, Chandler, AZ, for the permission in using their figures to
develop discussions in this review.
This work is supported by RGC of Hong Kong – GRF project: RGC project no. 111309 and CityU project no. 9041486.
References
[1] Assembly and packaging section. In: International technology roadmap for semiconductors. San Jose: Semiconductor
Industry Association; 2003.
[2] Tu KN. Solder joint technology: materials, properties, and reliability. 1st ed. New York: Springer Science + Business Media
LLC; 2007.
[3] Assembly and packaging section. In: International technology roadmap for semiconductors. San Jose: Semiconductor
Industry Association; 2007.
[4] Chen C, Liang SW. Electromigration issues in lead-free solder joints. J Mater Sci: Mater Electron 2007;18:259–68.
[5] Huntington HB, Grone AR. Current-induced marker motion in gold wires. J Phys Chem Solids 1961;20:76–87.
[6] Brandenburg S, Yeh S. Electromigration studies of flip chip bump solder joints. In: Proceedings of the surface mount
international conference and exhibition. San Jose (CA): SMI; 1998. p. 337–44.
[7] Assembly and packaging section. In: International technology roadmap for semiconductors. San Jose: Semiconductor
Industry Association; 2002.
[8] Ye H, Basaran C, Hopkins DC. Mechanical degradation of microelectronics solder joints under current stressing. Int J Solids
Struct 2003;40(26):7269–84.
[9] Campbell DR, Huntington HB. Thermomigration and electromigration in zirconium. Phys Rev 1969;179(3):601–12.
[10] Gupta D, Vieregge K, Gust W. Interface diffusion in eutectic Pb–Sn solder. Acta Mater 1999;47(1):5–12.
[11] Nah JW, Kim JH, Lee HM, Paik KW. Electromigration in flip chip solder bump of 97Pb3Sn/37Pb63Sn combination
structure. Acta Mater 2004;52(1):129–36.
[12] Yoon MS, Lee SB, Kim OH, Park YB, Joo YC. Relationship between edge drift and atomic migration during electromigration
of eutectic SnPb lines. J Appl Phys 2006;100:033715.
[13] Agarwal R, Ou SQ, Tu KN. Electromigration and critical product in eutectic SnPb solder lines at 100 C. J Appl Phys
2006;100:024909.
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
473
[14] Liu CY, Chen C, Tu KN. Electromigration in Sn–Pb solder strips as a function of alloy composition. J Appl Phys
2000;88:5703–9.
[15] Ho CE, Lee A, Subramanian KN, Liu W. Early stage of material movements in eutectic SnPb solder joint undergoing current
stressing at 150 °C. Appl Phys Lett 2007;91:021906.
[16] Yang QL, Shang JK. Interfacial segregation of Bi during current stressing of Sn–Bi/Cu solder interconnect. J Electron Mater
2005;34:1363–7.
[17] Chen LT, Chen CM. Electromigration study in the eutectic SnBi solder joint on the Ni/Au metallization. J Mater Res
2006;21(4):962–9.
[18] Lee TY, Tu KN, Kuo SM, Frear DR. Electromigration of eutectic SnPb solder interconnects for flip chip technology. J Appl
Phys 2001;89:3189–94.
[19] Yang D. Study on reliability of flip chip solder interconnects for high current density packaging. Doctor of philosophy
thesis. Hong Kong: City University of Hong Kong; 2008.
[20] Gu X, Chan YC. Electromigration in line-type Cu/Sn–Bi/Cu solder joints. J Electron Mater 2008;37(11):1721–6.
[21] Lee TY, Tu KN, Frear DR. Electromigration of eutectic SnPb and SnAg3.8Cu0.7 flip chip solder bumps and under-bump
metallization. J Appl Phys 2001;90(9):4502–8.
[22] Xu LH, Pang HL, Tu KN. Effect of electromigration-induced back stress gradient on nanoindentation marker movement in
SnAgCu solder joints. Appl Phys Lett 2006;89:221909.
[23] Wu BY, Alam MO, Chan YC, Zhong HW. Joule heating enhanced phase coarsening in the Sn37Pb and Sn3.5Ag0.5Cu solder
joints during current stressing. J Electron Mater 2008;37(4):469–76.
[24] Wu BY, Zhong HW, Chan YC, Alam MO. Degradation of Sn37Pb and Sn3.5Ag0.5Cu solder joints between Au/Ni (P)/Cu pads
stressed with moderate current density. J Mater Sci: Mater Electron 2006;17(11):943–50.
[25] Burke JE, Turnbull D. Recrystallization and grain growth. Prog Metal Phys 1952;3:220–92.
[26] Matin MA, Vellinga WP, Geers MGD. Aspects of coarsening in eutectic Sn–Pb. Acta Mater 2004;52(12):3475–82.
[27] Ye H, Basaran C, Hopkins DC. Pb phase coarsening in eutectic Pb/Sn flip chip solder joints under electric current stressing.
Int J Solids Struct 2004;41:2743–55.
[28] Rinne GA. Issues in accelerated electromigration of solder bumps. Microelectron Reliab 2003;43(12):1975–80.
[29] Chen CM, Huang CC, Liao CN, Liou KM. Effects of copper doping on microstructural evolution in eutectic SnBi solder strips
under annealing and current stressing. J Electron Mater 2007;36(7):760–5.
[30] Chen CM, Huang CC. Effects of silver doping on electromigration of eutectic SnBi solder. J Alloy Compd 2008;461:235–41.
[31] Yang D, Chan YC, Wu BY, Pecht M. Electromigration and thermomigration behavior of flip chip solder joints in high
current density packages. J Mater Res 2008;23(9):2333–9.
[32] Yeh ECC, Choi WJ, Tu KN, Elenius P, Balkan H. Current-crowding-induced electromigration failure in flip chip solder joints.
Appl Phys Lett 2002;80(4):580–2.
[33] Nah JW, Paik KW, Suh JO, Tu KN. Mechanism of electromigration-induced failure in the 97Pb3Sn and 37Pb63Sn composite
solder joints. J Appl Phys 2003;94:7560–6.
[34] Chiang KN, Lee CC, Lee CC, Chen KM. Current crowding-induced electromigration in SnAg3.0Cu0.5 microbumps. Appl
Phys Lett 2006;88:072102.
[35] Liang SW, Chang YW, Shao TL, Chen C, Tu KN. Effect of three-dimensional current and temperature distributions on void
formation and propagation in flip chip solder joints during electromigration. Appl Phys Lett 2006;89:022117.
[36] Chiu SH, Chen C. Investigation of void nucleation and propagation during electromigration of flip-chip solder joints using
X-ray microscopy. Appl Phys Lett 2006;89:262106.
[37] Zhang LY, Ou SQ, Huang J, Tu KN, Gee S, Nguyen L. Effect of current crowding on void propagation at the interface between
intermetallic compound and solder in flip chip solder joints. Appl Phys Lett 2006;88:012106.
[38] Hsu YC, Chen DC, Liu PC, Chen C. Measurement of electromigration parameters of lead-free SnAg3.5 solder using U-groove
lines. J Mater Res 2005;20(10):2831–7.
[39] Chang YW, Liang SW, Chen C. Study of void formation due to electromigration in flip-chip solder joints using Kelvin bump
probes. Appl Phys Lett 2006;89:032103.
[40] Chang YW, Chiang TH, Chen C. Effect of void propagation on bump resistance due to electromigration in flip-chip solder
joints using Kelvin structure. Appl Phys Lett 2007;91:132113.
[41] Tsai CM, Lin YL, Tsai JY, Lai YS, Kao CR. Local melting induced by electromigration in flip-chip solder joints. J Electron
Mater 2006;35(5):1005–9.
[42] Chae SH, Zhang XF, Lu KH, Chao HL, Ho PS, Ding M, et al. Electromigration statistics and damage evolution for Pb-free
solder joints with Cu and Ni UBM in plastic flip-chip packages. J Mater Sci: Mater Electron 2007;18:247–58.
[43] Ding M, Wang GT, Chao B, Ho PS. Effect of contact metallization on electromigration reliability of Pb-free solder joints. J
Appl Phys 2006;99:094906.
[44] Lee JH, Lee YD, Park YB, Yang ST, Suh MS, Chung QH, et al. Joule heating effect on the electromigration lifetimes and failure
mechanisms of Sn–3.5Ag solder bump. In: Proceedings of the 57th electronic components and technology conference.
Reno (NV): IEEE; 2007. p. 1436–41.
[45] Lai YS, Lee CW. Electromigration reliability and morphologies of 62Sn–36Pb–2Ni and 62Sn–36Pb–2Cu flip-chip solder
joints. IEEE Trans Compon Package Technol 2007;30(3):526–31.
[46] Choi WJ, Yeh ECC, Tu KN. Mean-time-to-failure study of flip chip solder joints on CuNi(V)Al thin film under bump
metallization. J Appl Phys 2003;94:5665–71.
[47] Chiu SH, Shao TL, Chen C, Yao DJ, Hsu CY. Infrared microscopy of hot spots induced by Joule heating in flip-chip SnAg
solder joints under accelerated electromigration. Appl Phys Lett 2006;88:022110.
[48] Wu BY. Reliability studies of flip-chip and ball-grid-array solder joints under current stressing. Doctor of philosophy
thesis. Hong Kong: City University of Hong Kong; 2007.
[49] Liang SW, Chang YW, Chen C. Effect of Al-trace dimension on Joule heating and current crowding in flip-chip solder joints
under accelerated electromigration. Appl Phys Lett 2006;88:172108.
[50] Alam MO, Wu BY, Chan YC, Tu KN. High electric current density-induced interfacial reactions in micro ball grid array
(lBGA) solder joints. Acta Mater 2006;54(3):613–21.
474
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
[51] Alam MO. Study of interfacial reactions in ball grid array (BGA) solder joints for advanced integrated circuit (IC)
packaging. Doctor of philosophy thesis. Hong Kong: City University of Hong Kong; 2004.
[52] Hu YC, Lin YH, Kao CR, Tu KN. Electromigration failure in flip chip solder joints due to rapid dissolution of copper. J Mater
Res 2003;18(11):2544–8.
[53] Lin YH, Hu YC, Tsai CM, Kao CR, Tu KN. In situ observation of the void formation and propagation mechanism in solder
joints under current stressing. Acta Mater 2005;53(7):2029–35.
[54] Dyson BF. Diffusion of gold and silver in tin single crystals. J Appl Phys 1966;37:2375–7.
[55] Dyson BF, Anthony TR, Turnbull D. Interstitial diffusion of copper in tin. J Appl Phys 1967;38:3408.
[56] Yeh DC, Huntington HB. Extreme fast-diffusion system nickel in single-crystal tin. Phys Rev Lett 1984;53:1469–72.
[57] Hu CK, Huntington HB. Diffusion and electromigration of silver and nickel in lead–tin alloys. Phys Rev B 1982;26:2782–9.
[58] Hu CK, Huntington HB, Gruzalski GR. Atom motions of copper dissolved in lead–tin alloys. Phys Rev B 1983;28:579–85.
[59] Hu CK, Huntington HB. Atom movements of gold in lead–tin solders. J Appl Phys 1985;58:2564–9.
[60] Jen MHR, Liu LC, Lai YS. Electromigration on void formation of Sn3Ag1.5Cu FCBGA solder joints. Microelectron Reliab
2009;49:734–45.
[61] Zhao JH, Gupta V, Zeng KJ. Electromigration-resistant flip-chip solder joints. US patent; 2008/0251927 A1.
[62] Nah JW, Suh JO, Tu KN, Yoon SW, Rao VS, Kripesh V, et al. Electromigration in flip chip solder joints having a thick Cu
column bump and a shallow solder interconnect. J Appl Phys 2006;100:123513.
[63] Kwon YM, Paik KW. Electromigration of Pb-free solder flip chip using electroless Ni-P/Au UBM. In: Proceedings of the 57th
electronics components and technology conference. Nevada (US): IEEE; 2007. p. 1472–7.
[64] Wu AT, Gusak AM, Tu KN, Kao CR. Electromigration-induced grain rotation in anisotropic conducting beta tin. Appl Phys
Lett 2005;86:241902.
[65] Wu AT, Hsieh YC. Direct observation and kinetic analysis of grain rotation in anisotropic tin under electromigration. Appl
Phys Lett 2008;92:121921.
[66] Lu MH, Shih DY, Lauro P, Goldsmith C, Henderson DW. Effect of Sn grain orientation on electromigration degradation
mechanism in high Sn-based Pb-free solders. Appl Phys Lett 2008;92:211909.
[67] Lu MH, Shih DY, Kang SK, Goldsmith C, Flaitz P. Effect of Zn doping on SnAg solder microstructure and electromigration
stability. J Appl Phys 2009;106:053509.
[68] Liu YH, Lin KL. Damages and microstructural variation of high-lead and eutectic SnPb composite flip chip solder bumps
induced by electromigration. J Mater Res 2005;20(8):2184–93.
[69] Shao TL, Chen YH, Chiu SH, Chen C. Electromigration failure mechanisms for SnAg3*5 solder bumps on Ti/Cr–Cu/Cu and
Ni(P)/Au metallization pads. J Appl Phys 2004;96(8):4518–24.
[70] Liang SW, Chiu SH, Chen C. Effect of Al-trace degradation on Joule heating during electromigration in flip-chip solder
joints. Appl Phys Lett 2007;90:082103.
[71] Ouyang FY, Tu KN, Kao CL, Lai YS. Effect of electromigration in the anodic Al interconnect on melting of flip chip solder
joints. Appl Phys Lett 2007;90:211914.
[72] Yang D, Chan YC, Tu KN. The time-dependent melting failure in flip chip lead-free solder interconnects under current
stressing. Appl Phys Lett 2008;93(4):027830.
[73] Hu CK, Rodbell KP, Sullivan TD, Lee KY, Bouldin DP. Electromigration and stress-induced voiding in fine Al and Al-alloy
thin-film lines. IBM J Res Dev 1995;39(4):465–97.
[74] Gan H, Tu KN. Polarity effect of electromigration on kinetics of intermetallic compound formation in Pb-free solder Vgroove samples. J Appl Phys 2005;97:063514.
[75] Chen CM, Chen SW. Electric current effects on Sn/Ag interfacial reactions. J Electron Mater 1999;28(7):902–6.
[76] Chen CM, Chen SW. Electromigration effect upon the Sn-0.7 wt% Cu/Ni and Sn-3.5 wt% Ag/Ni interfacial reactions. J Appl
Phys 2001;90(3):1208–14.
[77] Chen SW, Chen CM. Electromigration effects upon interfacial reactions. JOM 2003;55(2):62–7.
[78] Chen SW, Wang CH. Effects of electromigration on interfacial reactions in cast Sn/Cu joints. J Mater Res
2007;22(3):695–702.
[79] Chen CC, Chen SW, Chang CH. Electromigration effects upon the Sn/Ni-7 wt% V interfacial reactions. J Appl Phys
2008;103:063518.
[80] Tu KN, Thompson RD. Kinetics of interfacial reaction in bimetallic Cu–Sn thin films. Acta Metall 1982;30:947–52.
[81] Xu LH, Pang HL, Ren F, Tu KN. Electromigration effect on intermetallic growth and Young’s modulus in SAC solder joint. J
Electron Mater 2006;35(12):2116–25.
[82] Gurov KP, Gusak AM. On the theory of phase growth in the diffusion zone during mutual diffusion in an external electric
field. Phys Met Metallogr 1981;52(4):75–81.
[83] Chao B, Chae SH, Zhang XF, Lu KH, Im J, Ho PS. Investigation of diffusion and electromigration parameters for Cu–Sn
intermetallic compounds in Pb-free solders using simulated annealing. Acta Mater 2007;55(8):2805–14.
[84] Chao B, Zhang XF, Chae SH, Ho PS. Recent advances on kinetic analysis of electromigration enhanced intermetallic growth
and damage formation in Pb-free solder joints. Microelectron Reliab 2009;49:253–63.
[85] Lee TY, Choi WJ, Tu KN, Jang JW, Kuo SM, Lin JK, et al. Morphology, kinetics, and thermodynamics of solid-state aging of
eutectic SnPb and Pb-free solders (Sn–3.5Ag, Sn–3.8Ag–0.7Cu and Sn–0.7Cu) on Cu. J Mater Res 2002;17(2):291–301.
[86] Yoon JW, Jung SB. Effect of isothermal aging on intermetallic compound layer growth at the interface between Sn–3.5Ag–
0.75Cu solder and Cu substrate. J Mater Sci 2004;39(13):4211–7.
[87] Mei Z, Sunwoo AJ, Morris JW. Analysis of low-temperature intermetallic growth in copper–tin diffusion couples. Metall
Trans A 1992;23(3):857–64.
[88] Liu CY, Ke L, Chuang YC, Wang SJ. Study of electromigration-induced Cu consumption in the flip-chip SnCu solder bumps. J
Appl Phys 2006;100:083702.
[89] Orcharda HT, Greer AL. Electromigration effects on compound growth at interfaces. Appl Phys Lett 2005;86:231906.
[90] Wu BY, Chan YC, Zhong HW, Alam MO. Effect of current stressing on the reliability of 63Sn37Pb solder joints. J Mater Sci
2007;42(17):7415–22.
Y.C. Chan, D. Yang / Progress in Materials Science 55 (2010) 428–475
475
[91] Gu X, Yang D, Chan YC, Wu BY. Effects of electromigration on the growth of intermetallic compounds in Cu/SnBi/Cu solder
joint. J Mater Res 2008;23(10):2591–6.
[92] Frear DR, Jones WB, Kinsman KR. Solder mechanics: a state of the art assessment. Warrendale (PA): The Minerals, Metals
and Materials Society; 1990.
[93] Zhang XF, Guo JD, Shang JK. Abnormal polarity effect of electromigration on intermetallic compound formation in Sn–9Zn
solder interconnect. Scripta Mater 2007;57(6):513–6.
[94] Kuo SM, Lin KL. Polarity effect of electromigration on intermetallic compound formation in a Cu/Sn–9Zn/Cu sandwich. J
Mater Res 2008;23(4):1087–94.
[95] Tu KN. Irreversible processes of spontaneous whisker growth in bimetallic Cu–Sn thin-film reactions. Phys Rev B
1994;49(3):2030–4.
[96] Kuo SM, Lin KL. The hillock formation in a Cu/Sn–9Zn/Cu lamella upon current stressing. J Electron Mater
2007;36(10):1378–82.
[97] Lee A, Liu W, Ho CE, Subramanian KN. Synchrotron X-ray microscopy studies on electromigration of a two-phase material.
J Appl Phys 2007;102:053507.
[98] Lee A, Liu W, Subramanian KN. Electromigration induced microstructure and morphological changes in eutectic SnPb
solder joints. J Mater Res 2007;22(11):3265–572.
[99] Lin YW, Lai YS, Lin YL, Tu CT, Kao CR. Tin whisker growth induced by high electron current density. J Electron Mater
2007;36(10):1378–82.
[100] Ouyang FY, Chen K, Tu KN, Lai YS. Effect of current crowding on whisker growth at the anode in flip chip solder joints.
Appl Phys Lett 2007;91:231919.
[101] Herring C. Diffusional viscosity of a polycrystalline solid. J Appl Phys 1950;21:437–45.
[102] Blech IA. Electromigration in thin aluminum films on titanium nitride. J Appl Phys 1976;47:1203–8.
[103] Wei CC, Chen C. Critical length of electromigration for eutectic SnPb solder stripe. Appl Phys Lett 2006;88:182105.
[104] Wu BY, Chan YC, Zhong HW, Alam MO, Lai JKL. Study of the thermal stress in a Pb-free half-bump solder joint under
current stressing. Appl Phys Lett 2007;90:232112.
[105] Ye H, Basaran C, Hopkins DC. Deformation of solder joint under current stressing and numerical simulation – I. Int J Solids
Struct 2004;41:4939–58.
[106] Nah JW, Ren F, Paik KW, Tu KN. Effect of electromigration on mechanical shearing behavior of flip chip solder joints. J
Mater Res 2006;21(3):698–702.
[107] Ren F, Nah JW, Tu KN, Xiong BS, Xu LH, Pang HL. Electromigration induced ductile-to-brittle transition in lead-free solder
joints. Appl Phys Lett 2006;89:141914.
[108] Ye H, Basaran C, Hopkins DC. Thermomigration in Pb–Sn solder joints under Joule heating during electric current
stressing. Appl Phys Lett 2003;82(8):1045–7.
[109] Huang AT, Gusak AM, Tu KN, Lai YS. Thermomigration in SnPb composite flip chip solder joints. Appl Phys Lett
2006;88:141911.
[110] Yang D, Wu BY, Chan YC, Tu KN. Microstructural evolution and atomic transport by thermomigration in eutectic tin–lead
flip chip solder joints. J Appl Phys 2007;102:012716.
[111] Shewmon P. Diffusion in solids. 2nd ed. Warrendale: TMS; 1989.
[112] Tu KN. Recent advances on electromigration in very-large-scale-integration of interconnects. J Appl Phys
2003;94(9):5451–73.
[113] Hsiao HY, Chen C. Thermomigration in flip-chip SnPb solder joints under alternating current stressing. Appl Phys Lett
2007;90:152105.
[114] Huang AT, Tu KN, Lai YS. Effect of the combination of electromigration and thermomigration on phase migration and
partial melting in flip chip composite SnPb solder joints. J Appl Phys 2006;100:033512.
[115] Gu X, Chan YC. Thermomigration and electromigration in Sn58Bi solder joints. J Appl Phys 2009;105:093537.
[116] Ouyang FY, Tu KN, Lai YS, Gusak AM. Effect of entropy production on microstructure change in eutectic SnPb flip chip
solder joints by thermomigration. Appl Phys Lett 2006;89:221906.
[117] Chuang YC, Liu CY. Thermomigration in eutectic SnPb alloy. Appl Phys Lett 2006;88:174105.
[118] Chuang YC, Liu CY. The effect of thermomigration on phase coarsening in a eutectic SnPb alloy. J Electron Mater
2007;36(11):1495–500.
[119] Hsiao HY, Chen C. Thermomigration in Pb-free SnAg solder joint under alternating current stressing. Appl Phys Lett
2009;94:092107.
[120] Gu X. Electromigration and thermomigration studies of lead-free solder joints. Doctor of philosophy thesis. Hong Kong:
City University of Hong Kong; 2009.
[121] Li S, Abdulhamid MF, Basaran C. Damage mechanics of low temperature electromigration and thermomigration. IEEE
Trans Adv Package 2009;32:478–85.
[122] Abdulhamid MF, Basaran C. Influence of thermomigration on lead-free solder joint mechanical properties. J Electron
Package 2009;131:011002.
[123] Chen HY, Chen C, Tu KN. Failure induced by thermomigration of interstitial Cu in Pb-free flip chip solder joints. Appl Phys
Lett 2008;93:122103.
[124] Yu DQ, Chai TC, Thew ML, Ong YY, Rao VS, Wai LC, et al. Electromigration study of 50 lm pitch micro solder bumps using
four-point kelvin structure. In: Proceedings of the 59th electronic components and technology conference. San Diego
(CA): IEEE; 2009. p. 930–5.
[125] Xu LH, Han JK, Liang J, Tu KN, Lai YS. Electromigration induced high fraction of compound formation in SnAgCu flip chip
solder joints with copper column. Appl Phys Lett 2008;92:262104.
[126] Ouyang FY, Huang AT, Tu KN. Thermomigration in SnPb composite solder joints and wires. In: Proceedings of the 56th
electronic components and technology conference. San Diego (CA): IEEE; 2006. p. 1974–8.
[127] Yang D, Chan YC. The characteristics of electromigration and thermomigration in flip chip solder joints. In: Proceedings of
the 13th international workshop on thermal investigations of ICs and systelectromigrations. Budapest (Hungary): IEEE;
2007. p. 43–7.