Microelectronics Reliability 102 (2019) 113474

Contents lists available at ScienceDirect

Microelectronics Reliability
journal homepage: www.elsevier.com/locate/microrel

Vibration lifetime estimation of PBGA solder joints using Steinberg model T

a a,⁎ b,⁎ a a c d
Tong An , Fei Qin , Bin Zhou , Pei Chen , Yanwei Dai , Huaicheng Li , Tao Tang
a
Institute of Electronics Packaging Technology and Reliability, College of Mechanical Engineering and Applied Electronics Technology, Beijing University of Technology,
Beijing 100124, China
b
Science and Technology on Reliability Physics and Application of Electronic Component Laboratory, China Electronic Product Reliability and Environmental Testing
Research Institute, Guangzhou 510610, China
c
Qingdao Hisense Hitachi Air-Conditioning Systems Co., Ltd, China
d
Monolithic Power Systems (Chengdu) Co., Ltd, China

A R T I C LE I N FO A B S T R A C T

Keywords: The work studies the vibration lifetime modeling of Sn37Pb soldered plastic ball grid array (PBGA) assemblies on
Plastic ball grid array (PBGA) the basis of vibration tests, finite element analysis (FEA) and Steinberg's empirical formulation. The test vehicles
Random vibration for the vibration tests consist of twelve PBGA components with built-in daisy-chained circuits, which are as-
Vibration lifetime prediction sembled on a printed circuit board (PCB) symmetrically. First, the natural frequencies of the test vehicle were
Steinberg's model
determined by modal tests. The first three natural frequencies were obtained using FEA, which were compared
Solder joint
with those from the modal test. Then, narrow-band sinusoid vibration tests were conducted at the first natural
Finite element analysis (FEA)
frequency of the test vehicle using constant-amplitude excitation, and the number of failure cycles was recorded
by monitoring the overall electrical resistance. FEA was performed to obtain the stress of the critical solder joints
under various levels of sinusoidal vibration loadings. To construct the stress versus failure cycles (SeN) curve of
the solder joints, the stresses of solder joints were used in conjunction with the lifetime obtained from the
sinusoidal vibration tests. The random vibration test was conducted on the test vehicle to assess the lifetime of
the solder joints subjected to broad-band excitation. Random vibration analyses were performed numerically to
obtain the displacement responses of the PBGA assembly. Finally, a fatigue life prediction based on Steinberg's
model was deduced from all these procedures. A comparison of the lifetime between the experimental results
and the prediction suggests that the methodology is valid and practicable in predicting the vibration lifetime of
PBGA solder joints.

1. Introduction temperature cycling and 20% of the failures are due to shock and vi-
bration [2]. Thus, temperature cycling and vibration events appear to
The plastic ball grid array (PBGA) component was widely used in be two major environmental causes of solder joint fatigue failure. Be-
microelectronic packaging technologies because of its advantages in cause the majority failure of solder joint is caused by the thermal
low cost, high I/O density, miniaturization, and good electrical per- loading, most of the related works have considered the fatigue lifetime
formance. With increasing miniaturization and integration of electronic modeling under thermal loading [3–5]. However, relatively few in-
devices, the number of solder joints increases and the size of them vestigations have focused on the fatigue failure of solder joints in a
decreases. Solder joints function not only as electrical connections but vibration environment. In fact, for many real-world applications,
also as important mechanical supports between electronic components especially applications in the automobile, military, and avionic in-
and a printed circuit board (PCB) [1]. Therefore, the reliability of the dustries, electronic devices commonly operate in vibration environ-
solder joints is a key factor affecting the development of PBGA packa- ments; thus, enhanced understanding of the vibration lifetime of solder
ging technology. joints is very important.
In electronic components, thermal-induced, vibration-induced, and The current challenge is how to conduct the vibration lifetime
shock/impact-induced failures are the major causes of failure. A survey analysis rapidly and accurately. Numerous experimental studies have
compiled by the U.S. Air Force shows that approximately 55% of the been devoted to establishing an effective and valid methodology for the
failures of electronic devices are related to high temperatures and vibration lifetime prediction of solder joints. Two different vibration

⁎
Corresponding authors.
E-mail addresses: [email protected] (F. Qin), [email protected] (B. Zhou).

https://doi.org/10.1016/j.microrel.2019.113474
Received 16 November 2018; Received in revised form 24 May 2019; Accepted 30 July 2019
Available online 08 August 2019
0026-2714/ © 2019 Elsevier Ltd. All rights reserved.
T. An, et al. Microelectronics Reliability 102 (2019) 113474

tests have been developed: the sinusoid vibration test and the random law and the rain-flow algorithm, and three-band technology was ap-
vibration test. In the sinusoid vibration test, the test vehicle is subjected plied for the calculation of the CDI. Yu et al. [24] developed an as-
to a narrow-frequency-band excitation at its first natural frequency, sessment methodology on the basis of the SeN curve, the Miner's rule,
which is the most severe situation of vibration resonance loading. Yang and the rain-flow cycle counting results. Zhou et al. [7,25] developed a
et al. [6] studied the fatigue failure of solder joints in PBGA components fatigue life model that included fatigue damage contributions from both
using sinusoid vibration tests. Their results indicated that the resonant elastic deformation (Basquin's law) and plastic deformation (Coffin-
frequency of the PCB is often excited under sinusoid vibration loading, Manson's law). Steinberg [2] developed his empirical model to de-
which forces the PCB to bend back and forth; this action will produce termine component lifetime under vibration on the basis of substantial
very high stress levels in the solder joints, resulting in rapid fatigue experimental work. This model reveals the relationship between the
failure. Because it is difficult to measure the stresses of solder joints measured physical quantities and the fatigue lifetime of the component,
directly during tests, Zhou et al. [7] investigated the vibration lifetime and it has been widely utilized for the design of electronic systems
of both Sn37Pb and Sn3.0Ag0.5Cu solder joints in PBGA components which service in a high-vibration and high-shock environment [26,27].
using a combination of finite element analysis (FEA) and sinusoid vi- In this paper, the SeN fatigue curve of Sn37Pb solder joint is con-
bration tests. Their work showed that, during the sinusoid vibration structed by combining results from sinusoidal vibration tests and FEA.
tests, the stress amplitude in the solder joints was constant because of An assessment methodology for the vibration lifetime prediction of
the use of constant-amplitude excitation. Therefore, the stresses of PBGA solder joints based on the SeN fatigue curve, FEA and an ad-
solder joints subjected to sinusoid vibration loading conditions can be justed Steinberg model is developed, and the proposed prediction
obtained from FEA. In addition, the sinusoid vibration test is always model of vibration lifetime is confirmed by random vibration tests.
applied in determining the stress versus failure cycles (SeN) fatigue
curve of solder joints. Chen et al. [8] conducted sinusoid vibration tests 2. Experimental procedure and finite element analysis
with different excitation levels by varying the constant acceleration
amplitudes, and the failure time results were correlated with the 2.1. Vibration test bench
stresses from FEA to present the SeN fatigue curve of the solder joints.
To estimate the lifetime of solder joints under sinusoid vibration A test vehicle was designed and fabricated for this paper in ac-
loading, Miner's rule [9] is used to calculate the cumulative damage cordance with the requirements specified in IPC-7095C [28]. The test
index (CDI). Perkins and Sitaraman [10] applied Miner's rule to esti- vehicles consist of twelve specially designed PBGA components, which
mate the lifetime of a CCGA component under sinusoidal vibration are assembled on a PCB symmetrically. The PCB is 300 mm in length,
loading, and the failure lifetime predicted by the Miner's rule was va- 180 mm in width, 2 mm thick. The size of the PBGA component is
lidated by the test results. Chen et al. [8] used Miner's rule to predict 21 mm × 27 mm × 2.3 mm. The PBGA component is mounted with
the lifetime of a PBGA component under sinusoidal vibration loading, 256 Sn37Pb solder joints. The diameter, the standoff height and the
and the calculated CDIs were approximately equal to one (0.9–1.07). pitch of the solder joints are 0.75 mm, 0.7 mm and 1.27 mm, respec-
Che et al. [11] used Miner's rule to estimate the vibration lifetime of a tively. The daisy-chain designed are applied to all the solder joints, and
flip-chip component subjected to sinusoidal vibration loading, and they thus the overall electrical resistance can be measured for each com-
investigated the effect of different assumed CDI factors on the lifetime ponent. Fig. 1 shows a photograph of a test vehicle with the compo-
prediction. Basaran and Chandaroy [12] presented that when the stress nents labeled in consecutive order. The components are divided into
of solder joint was in the elastic range, the Miner's rule could be used three groups according to the symmetry of the test board.
for the fatigue lifetime prediction. While for plastic deformation Vibration tests were carried out on an electrodynamic shaker from
loading, Miner's rule may underestimate the result of cumulative da- SUSHI (see Fig. 2). For the tests, the PCB assembly was fixed onto the
mage. shaker through four screws. Two accelerometers placed on the center of
The other set of vibration tests are random vibration tests. The vi- the PCB and the shaker were used to record both the PCB's response and
bration environment is more random in nature than in real-word ap- the shaker's input. During the tests, the resistance of the daisy-chain
plications. When a PBGA assembly is excited by random vibration, the circuits was monitored continuously to record the number of failure
system responds by vibrating at its resonant frequencies with a varying cycles.
displacement amplitude; thus, the stress amplitude of the solder joints is
not constant. The randomness of the excitation in random vibration 2.2. Finite element model
tests makes it difficult to estimate the stress responses of the solder
joints. Usually, spectrum analysis is performed by FEA to acquire the The finite element model of the test vehicle was constructed using
statistical values, such as the root mean square (RMS) value, of the ABAQUS. The model consists of solder joints, Cu pad, solder mask,
stresses in the solder joint [13–15]. The fatigue life of the solder joints is substrate, mold compound (MC) and PCB, as presented in Fig. 3. The
then determined with these stress values. Generally, fatigue life models geometric dimensions of the model are the same as that of the test
are formulated for the constant amplitude excitation condition. For sample.
instance, the Coffin-Manson low cycle fatigue law [16,17] is used for To promote the efficiency of the computation, specific simplification
constant strain amplitude loading, while the Basquin high cycle fatigue methodology was used here. Three finite element models were estab-
law [18] is applied in the case of constant stress amplitude loading. For lished because of symmetry. In Model I, only the solder joints of com-
the random vibration condition, in which the stress amplitude is not ponent U1 were modeled in detail; continuous uniform solder layers
constant, the aforementioned fatigue models need improvement and were used to represent solder joints of all of the other components, as
should be used in conjunction with Miner's rule and the rain-flow cycle shown in Fig. 3. Thus, the stress of the solder joints in component U1
counting method [19] to present a vibration lifetime model for solder can be obtained using Model I. Similarly, in Model II and Model III, only
joints fatigue life prediction. Wu et al. [20] assumed the amplitude the solder joints of component U2 and component U3 were modeled in
distribution of random vibration is Gaussian and established a fatigue detail, respectively. The solder joints stresses in components U2 and U3
life prediction model based on Miner's cumulative damage rule and can be obtained using Model II and Model III, respectively.
Basquin power law relation. Wong et al. [21] and Wang et al. [22] The eight-node linear brick element with reduced integration, type
presented the lifetime prediction model of solder joints under random C3D8R in the ABAQUS, were applied to mesh thloade whole model.
vibration based on Miner's rule and Coffin-Manson's fatigue law. Yang There were totally 657,180 elements and 751,962 nodes in Model I,
et al. [23] estimated the lifetime of solder joints in a PBGA assembly 655,436 elements and 750,370 nodes in Model II and 639,606 elements
subjected to random vibration loading using Coffin–Manson's fatigue and 747,322 nodes in Model III.

2
T. An, et al. Microelectronics Reliability 102 (2019) 113474

0 X
Group II

Y
Group III

Group I

(a) Layout of the test vehicle (b) The loop within a

PBGA component
Fig. 1. The test vehicle.

In vibration studies, the linear elastic material properties were used are also shown in Table 2.
in the present work. Table 1 lists the elastic modulus E, Poisson's ratio ν
and the density ρ of the materials used in the FEA. The density of solder 2.4. Frequency-scanning test
layer and Cu layer were modified to ensure that the mass of the sim-
plified models was equivalent to that of the detailed model. The damping ratio ξ of the system was calculated through a fre-
quency-scanning test by the half-power bandwidth method in order to
2.3. Modal analysis and finite element modeling better correlate the finite element model with the experimental results.
The frequency-scanning test was conducted within a frequency range
Modal analysis was performed experimentally to measure the nat- from 5 Hz to 2000 Hz at 0.5 G input acceleration. The shaker input
ural frequencies of the test vehicle. The first three natural frequencies acceleration Gin and the PCB's output acceleration Gout were measured
obtained from the modal test were 92.67 Hz, 186.98 Hz and 233.98 Hz. by the accelerometers. The damping ratio ξ is defined as
Modal analysis was then performed numerically, and Table 2 compares Δf
the first three natural frequencies of the test vehicle determined from ξ=
2fn (1)
the modal test and those obtained from the simulation. The FEA results
and the experimental results are in good agreement, and the error where fn is the first natural frequency and Δf is the bandwidth of the
percentages of the results obtained from the FEA relative to those from half-power points. Here, we define the frequencies at which the re-
the modal test are all within 1%. The total masses of the test vehicle and sponse is 1/ 2 of its peak acceleration value Gpeak as the half-power
the finite element model are 274.8 g and 269.0 g, respectively, and the points. The output acceleration Gout under the frequency-scanning test
error percentage is 2.1%. These results indicate that the finite element is presented in Fig. 4. Using fn = 92.67 Hz and Δf = 6.43 Hz, the
model is feasible and reliable. The mode shapes obtained from the FEA damping ratio can be derived as ξ = 0.035 on the basis of Eq. (1).

test vehicle

Shaker
Accelerometer
Screw

Electrical
resistance monitor

Fig. 2. Experimental setup used for the vibration tests.

3
T. An, et al. Microelectronics Reliability 102 (2019) 113474

solder joints are modeled in detail

U1 U2 U3 U4 U5 U6
MC Substrate

Cu pad PCB
U7 U8 U9 U10 U11 U12
Solder joint

solder joints are modeled as a continuous uniform solder layer

Fig. 3. The finite element model of the test vehicle.

Table 1 15
Material properties of the PBGA model.
Material E (GPa) υ ρ (kg/m3)
12
MC 25 0.26 2.00 × 103
Output acceleration (G)

Δf
Substrate 22 0.28 4.00 × 103
Cu pad 129 0.38 8.94 × 103
PCB 29.5 0.3 2.32 × 103 9
Sn37Pb solder 43,250 0.363 2.32 × 103

6 0.707Gpeak
2.5. Vibration tests and finite element modeling

In this study, both sinusoid vibration tests and random vibration 3

tests were conducted. Twelve PBGA assemblies were used here, and the
test conditions are summarized in Table 3. The test vehicles were
subjected to uniaxial loading in their out-of-plane direction. All of the 0
test vehicles were tested until the failure criteria were reached, and 40 60 80 100 120 140
then the lifetime of solder joints was obtained. Frequency˄Hz˅
The narrow-band sinusoid vibration tests were conducted here.
Because the first natural frequency obtained from the modal test was Fig. 4. The output acceleration Gout under the frequency-scanning test.
found to be 92.67 Hz, the sinusoid vibration tests were conducted
within the frequency range of 87 Hz to 97 Hz. In addition, the accel- Fig. 5 shows the power spectral density (PSD) curve applied in the
eration magnitudes of constant input excitation were selected as 5 G, random vibration tests. The root-mean-square (RMS) acceleration Grms
7 G and 10 G. In the tests, failure was defined as a resistance value levels were 7 G, 11 G and 18 G, which are related to the area under the
greater than 1000 Ω for ten consecutive measurements, consistent with PSD curve. The random vibration tests were conducted over a fre-
the IPC standard [29]. quency range from 15 Hz to 2000 Hz. The failure criterion of the

Table 2
Comparison of the natural frequencies from the finite element analysis and the modal test.
Natural frequency (Hz) Mode shape

Experiment Model I Model II Model III Error (%) Finite element analysis

First mode 92.67 93.42 93.49 93.50 0.90

Second mode 186.98 185.14 186.34 188.49 0.98

Third mode 233.98 232.09 232.27 232.73 0.81

4
T. An, et al. Microelectronics Reliability 102 (2019) 113474

Table 3 where nine components failed in the test board. All the components in
Summary of the vibration tests. Group III and some of the components in Groups I and II of the test
Tests Loading Notes Number of Test board failed. The sinusoid vibration tests with an input constant G-level
test vehicles vehicle of 10 G were time terminated after 100% failure.
number A Weibull distribution [30] was adopted to describe the lifetime
results from the sinusoid vibration tests, and the cumulative failure
Sinusoid G-level = 5 G Frequency 1 #1
vibration G-level = 7 G from 87 to 1 #2
distribution function F(t) is expressed as:
G-level = 10 G 97 Hz 2 #3, #4
Random Grms-level = 7 G Frequency 2 #5, #6 t β
vibration Grms-level = 11 G from 15 to 2 #7, #8 F (t) = 1 − exp ⎡−⎛ ⎞ ⎤
⎢ ⎝α⎠ ⎥ (2)
Grms-level = 18 G 2000 Hz 4 #9, #10, ⎣ ⎦
#11, #12
where β is the shape parameter, and α is the lifetime at which 63.2% of
the components failed and is known as characteristic life.
Y The Weibull plot of the sinusoid vibration test results for three ex-
citation G-levels are shown in Fig. 6. The characteristic life of the solder
Power spectral density˄G2/Hz˅

joints in Group III under input excitations G-levels of 5 G, 7 G and 11 G

were calculated to be 947,645 cycles, 138,198 cycles and 104,811 cy-
cles, respectively. These results clearly show that the vibration lifetime
of the solder joints decreased with the increase of the input excitation.
To investigate the effect of component location on the vibration
4 dB/oct -6 dB/oct lifetime, a Weibull plot of the lifetime of the components in different
groups under the same G-level excitation was created, as shown in
Grms Fig. 7. At the excitation G-level of 10 G, the times to failure of Group I,
Group II and Group III are 194,275 cycles, 130,847 cycles and
104,811 cycles, respectively. For the components on the central region
of the test board (Group III), the solder joints fail with the shortest
lifetime, whereas the solder joints exhibit relatively good vibration
15 178 300 1000 2000 X
resistance for the components located farthest from the center of the
Frequency (Hz) test board (Group I). In addition, the vibration lifetime of the solder
joints in Group I is approximately twice that in Group III, indicating
Fig. 5. Power spectral density (PSD) profile for the random vibration tests.
that the component location strongly affects the lifetime of solder joints
and that closer proximity of a solder joint to the PCB center results in
random vibration test was based on the IPC standard, which was the shorter solder joint life.
same as that for the sinusoid vibration test.
Because it is difficult to measure the stresses of solder joints directly
during the vibration tests, FEA were performed to obtain the stress 3.1.2. Stress analysis of solder joints
amplitudes. Here, the sinusoid vibration analyses were performed to Table 5 lists the values of the maximum von Mises stress in the
obtain the values of von Mises stress in the critical solder joint subjected critical solder joint under different G-level sinusoid vibration excita-
to various vibration loading conditions. The G-levels used for the si- tions. The maximum stresses of the critical solder joints in component
nusoid vibration tests were 5 G, 7 G and 10 G, and the frequency range U3 at G-levels of 5 G, 7 G and 10 G were 189.7 MPa, 265.5 MPa and
was set from 87 Hz to 97 Hz. The boundary conditions and the loading 379.3 MPa, respectively, showing that the von Mises stress of the cri-
profile were exactly the same as those used in the sinusoid vibration tical solder joints in the same component increases with increasing
tests. The resulting stresses were then applied to develop the SeN fa- input excitation. Under the same G-level excitation, component U3,
tigue curve of the solder joints. which was located in the central region of the test board, exhibited a
The random vibration analyses were then performed numerically. higher stress level of the critical solder joint than the other components;
The first twenty mode shapes were obtained from modal analysis, and a by contrast, for the component U1, which was at the farthest location
frequency range of 15 Hz to 2000 Hz was set. The base excitation was from the center of the test board, the solder joints experienced lower
adopted when random vibration responses were computed. The PSD levels of stress. These results indicate that a component's location
curve used in the random vibration test was applied in the FEA. The strongly influences the stress level of the solder joints. The solder joints
input Grms excitation levels were 7 G, 11 G and 18 G. The boundary will experience higher stress levels when the components are located
conditions were similar to that of sinusoid vibration simulation. The closer to the central region of the test board. This result agrees with the
typical 1σ, 2σ and 3σ displacement levels at the component locations on experimental observation that the central components are more prone
the PCB were obtained from the simulations. to failure.
The histogram in Fig. 8(b) shows the stress distribution for each
3. Results and discussion solder joint of component U3 under the G-level = 10 G sinusoid vi-
bration loading condition. Because the flexure of PCB in the length
3.1. SeN fatigue curve for the PBGA solder joints direction exhibits much greater than in the width direction, the solder
joints at the outermost columns 1 and 16 in the PCB width direction flex
3.1.1. The fatigue life under sinusoid vibration loading more and experience a higher bending stress level in component U3.
Table 4 shows the statistics for the failed components for all the The calculated high stresses suggest that the solder joints at the out-
tested boards. The sinusoid vibration test with an input constant G-level ermost columns are likely to fail first.
of 5 G was time terminated after 58.33% failure, where seven compo- Fig. 8(a) shows the von Mises stress contour plot of the critical
nents failed in the test board. All the components on the 2 × 2 central solder joint under the G-level = 10 G sinusoid vibration loading con-
array of the test board (Group III) failed during the test, whereas the dition. The calculated maximum von Mises stress is located along the
components in Group I did not fail. The sinusoid vibration test with an Cu pad of the component/ solder joint interface, implying that the
input constant G-level of 7 G was time terminated after 75% failure, cracks of the failed solder joints will occur on the component side.

5
T. An, et al. Microelectronics Reliability 102 (2019) 113474

Table 4
Failed component statistics in the sinusoid vibration tests.
Tests Number of failed components/number of tested Failed component numbers Proportion of failed package
components

Sinusoid vibration G = 5 G 7/12 #1-U2, #1-U3, #1-U4, #1-U5, #1-U9, #1-U10, #1-U11 58.33%
Sinusoid vibration G = 7 G 9/12 #2-U1, #2-U2, #2-U3, #2-U4, #2-U8, #2-U9, #2-U10, #2- 75%
U11, #2-U12
Sinusoid vibration G = 10 G 24/24 #3-all; #4-all 100%
Random vibration Grms = 7 G 16/24 #5-U2, #5-U3, #5-U4, #5-U5, #5-U8, #5-U9, #5-U10, #5- 66.67%
U11;
#6-U2, #6-U3, #6-U4, #6-U5, #6-U8, #6-U9, #6-U10, #6-
U11
Random vibration Grms = 11 G 21/24 #7-U1, #7-U2, #7-U3, #7-U4, #7-U5, #7-U6, #7-U7, #7-U8, 87.5%
#7-U9, #7-U10, #7-U11;
#8-U2, #8-U3, #8-U4, #8-U5, #8-U7, #8-U8, #8-U9, #8-U10,
#8-U11
Random vibration Grms = 18 G 45/48 #9-all; #10-all; #11-all 93.75%
#12-U1, #12-U2, #12-U3, #12-U4, #12-U5, #12-U6, #12-U8,
#12-U9, #12-U10

99 Table 5
95 The calculated von Mises stresses of solder joints in sinusoid vibration excita-
90 tions.
80
70 Input G-level (G) Maximum von Mises stress (MPa)
60
50 Component U1 Component U2 Component U3
Unreliability (%)

40
30 5 77.09 149.7 189.7
7 106.8 209.6 265.5
20
10 152.5 299.4 379.3
α β
10
5 G, Group III 947645 1.19
5 7 G, Group III 138198 2.88 frequency of a typical PCB is on the order of 100 Hz. Therefore, the
lifetime of 100,000 cycles can be reached within 17 min. It indicates
10 G, Group III 104811 4.17
that the fatigue failure of the solder joints is in the high cycle fatigue life
2
regime. Thus, Basquin's power law can be used here
1 C
10000 100000 1000000 1E7 Nf =
(3)
σb
Time to failure (cycles)
where Nf is the number of failure cycles, C is the material constants, σ is
Fig. 6. Weibull distribution of time to failure for sinusoid vibration tests at the stress amplitude, and b is the fatigue exponent.
different G-levels. Fig. 9 shows the SeN fatigue curve of the Sn37Pb solder joints for
the PGBA assembly. This figure shows a log-log plot of the calculated
99 maximum von Mises stresses (S) of the critical solder joint from the FEA
95 and the number of cycles to failure (N) from sinusoid vibration tests.
90 The relation between the stress (σ) and the vibration lifetime (Nf) can be
80 described as
70
60 σ = 8887.59Nf−0.28 (4)
50
Unreliability (%)

40
30 Then,
20 1.27 × 1014
Nf =
σ 3.57 (5)
10
α β This equation results in a fatigue exponent of b = 3.57.
5 10 G, Group III 104811 4.17
10 G, Group II 130847 2.81 3.2. Vibration lifetime modeling
2 10 G, Group I 194275 3.25
3.2.1. The random vibration lifetime
1 The random vibration test with Grms = 7 G was time terminated
10000 100000 1000000 1E7 after 66.67% failure, or the failure of eight components in each test
Time to failure (cycles) board. All the components in Groups II and III of the two test boards
failed, whereas the components in Group I did not fail. The random
Fig. 7. Weibull distribution of the different group component lifetime for the
10 G sinusoid vibration tests.
vibration test with Grms = 11 G was time terminated after 87.5%
failure, at which point twenty-one components had failed in the two
test boards. All the components in Groups II and III and some of the
3.1.3. SeN fatigue curve development components in Group I of the test boards failed. The random vibration
The lifetime of the solder joints subjected to sinusoid vibration test with Grms = 18 G was time terminated after 93.75% failure, when
loading conditions exceeds 100,000 cycles. In addition, the natural forty-five components failed in the four test boards. All the components

6
T. An, et al. Microelectronics Reliability 102 (2019) 113474

PCB corner

PCB center

(a) Solder joint A1 in component U3 (b) Component U3

Fig. 8. The von Mises stress of solder joints predicted by finite element analysis under the G-level = 10 G sinusoid vibration condition.

400 in test boards #9, #10 and #11 failed, and all the components in Group
III and some of the components in Groups I and II of test board #12
Equation y = a*x^b
-- Value Standard Error
failed.
350 a 8887.588 4110.70559 A Weibull plot of the random vibration test results for the three Grms
von Mises stress (MPa)

b -0.2834 0.13311 excitation levels is shown in Fig. 10. The characteristic life of the solder
Adj. R-S 0.64258
joints at excitations of Grms = 7 G, Grms = 11 G and Grms = 18 G were
300 calculated to be 38,354 s, 14,893 s and 1536 s, respectively. These re-
sults clearly show that the fatigue life of the solder joints decreased with
increasing input excitation.
250
Fig. 11 shows a Weibull plot of the lifetime of the components in
different groups for the same Grms excitation level. Under the
200 Grms = 18 G excitation condition, the time to failure for Groups I, II and
III are 3128 s, 743 s and 520 s, respectively. For the components in the
central region of the test board (Group III), the solder joints failed with
150 the shortest lifetime; by contrast, for the components located farthest
0 200000 400000 600000 800000 1000000 from the center of the test board (Group I), the solder joints exhibited
Time to failure (cycles) relatively good vibration resistance. Similarly, under the Grms = 7 G
Fig. 9. The stress versus failure cycles (SeN) curve of Sn37Pb solder joint solder
and Grms = 11 G excitation conditions, the vibration lifetime of the
joints. solder joints in Group III was clearly shorter than that in other com-
ponents. Under the random vibration test condition, a frequency range
of 15 Hz to 2000 Hz was set. Although the forcing frequencies under the
99 random vibration loading can excite many resonant modes in the test
95 board, the fundamental resonant mode which often has the greatest
90
80 displacement amplitudes still has great impact on the stress of the
70 solder joints. Thus, the results indicate that a component's location
60
Unreliability (%)

50 strongly affects the lifetime of the solder joints, meaning that the solder
40 joints fail sooner when the component is nearer the center of the test
30 board.
20

10 3.2.2. An adjusted Steinberg vibration lifetime prediction model

α β In the Steinberg model, the ratio of lifetimes is equal to the ratio of
5 stresses to the specific fatigue exponent b:
7G 38,354 1.66
11 G 14,893 1.24 b
2 S
N1 = N2 ⎛ 2 ⎞
⎜ ⎟

18 G 1536 1.27 ⎝ S1 ⎠ (6)

1
100 1000 10000 100000 where S2 is the ultimate tensile strength (stress reference point), N2 is
Time to failure (s) the failure cycle times at stress S2 (vibration life reference point), S1 is
the maximum stress of the material under the current conditions, and
Fig. 10. Weibull distribution of the lifetime in random vibration tests with
different Grms levels. N1 is the expected fatigue life. For linear systems, the number of fatigue
cycles will be directly related to time; thus, time can be used to replace

7
T. An, et al. Microelectronics Reliability 102 (2019) 113474

99 Table 6
95 The parameters of the Steinberg model.
90
80 N2 Z0 (in) A (in) B (in) t (in) L (in) c b
70
60 2 × 10 7
0.0552 9.685 6.614 0.079 0.827 1.75 3.53
Unreliability (%)

50
40
30
Table 7
20
The 3σ displacements at the component locations on the PCB for the random
10 vibration tests.
α β Input Grms- Component U1 Component U2 Component U3
5 level (G) (mm) (mm) (mm)
18 G, Group III 520 9.26
18 G, Group II 743 7.57 7 0.42 0.669 0.822
2
18 G, Group I 3128 3.65 11 0.666 1.062 1.305
1 18 1.083 1.722 2.121
100 1000 10000
Time to failure (s)
the number of cycles. In addition, the stress S will be directly related to
the displacement Z. Thus, the relation can be rewritten as
(a) Grms = 18 G
b
99 Z
N1 = N2 ⎛ 2 ⎞
⎜ ⎟

95 ⎝ Z1 ⎠ (7)
90
80 The relative displacements at different positions on the PCB will
70 determine the forces and stresses in the component solder joints, which
60
Unreliability (%)

50 will directly influence the lifetime of these components. As a result, the

40 relative displacements can be utilized to find the approximate vibration
30
lifetime of the component solder joints in different environments on the
20
basis of the component location on the PCB. Here, the vibration lifetime
10 of the solder joints in the component is related to the dynamic dis-
α β placements developed by the PCB during vibration. Thus, Z2 is the peak
5 11 G, Group III 7784 1.31 displacement expected at the center of the PCB which can be estimated
11 G, Group II 13263 2.47 by
2 11 G, Group I 31424 1.92 0.00022A
Z2 =
1 ct L (8)
1000 10000 100000
where L is the component length, t is the PCB thickness, and A is the
Time to failure (s)
PCB length parallel to the component. Here c is a constant for different
types of components, and for a PBGA component, c = 1.75. When the
(b) Grms = 11 G
maximum PCB displacement is calculated by Eq. (8), N2 in Eq. (7) is
2 × 107 cycles.
99
α β In Eq. (7), Z1 is the current maximum PCB displacement at the ex-
95
90 pected location. Assuming that the majority of the damage occurs in the
7 G, Group III 17568 3.85
80 first resonant mode, the displacement at any PCB location can be de-
70 7 G, Group II 58044 3.68 termined from the following relation:
60
Unreliability (%)

50
40 πX πY
Z1 = Z0 sin ⎛ ⎞ sin ⎛ ⎞
30 ⎝ A⎠ ⎝ B ⎠ (9)
20
where Z0 is the PCB displacement at the center and A and B are the
10 lengths of the PCB along the X- and Y-axes, respectively. As a result, the
Steinberg model is expressed as
5 b
0.00022A
Z
b ⎛ ct L
⎞
2 N1 = N2 ⎛ 2 ⎞ = N2 ⎜
⎜ ⎟
⎟
⎝ Z1 ⎠ ⎜ Z0 sin
⎝ ( ) ( )
πX
A
πY
sin B ⎟
⎠ (10)
1
1000 10000 100000 The values of the parameters needed to evaluate Eq. (10) are given
Time to failure (s) in Table 6.
Notably, Z1 in Eq. (7) represents the current maximum displacement
(c) Grms = 7 G at the component location on the PCB. Importantly, the broad-band
random excitation was used here, and all of the frequencies in the
Fig. 11. Weibull distribution of the vibration lifetime of the components in
bandwidth are present instantaneously and simultaneously. However,
different groups during the random tests.
only the fundamental resonant mode is considered in the assumed de-
flection curve Eq. (9). Thus, the Z1 calculated from Eq. (9) under-
estimates the PCB displacement under random vibration loading. To
overcome this intrinsic shortcoming, the 3σ displacement levels (Z3) at
the component locations on the PCB were calculated via FEA; the

8
T. An, et al. Microelectronics Reliability 102 (2019) 113474

Table 8
Comparison of the random vibration tests results and the predicted fatigue life.
Input Grms-level and group Fatigue life results

Experiment (s) Eq. (10) (s) Error (%) Eq. (11) (s) Error (%)

18 G, group I 3128 17,536 460.61 8167 161.09

18 G, group II 743 4028 442.13 1559 109.83
18 G, group III 520 2211 325.19 741 42.50
11 G, group I 31,424 98,828 214.50 46,336 47.45
11 G, group II 13,263 22,701 71.16 8760 33.95
11 G, group III 7784 12,459 60.06 4198 46.07
7 G, group II 58,044 117,017 101.60 45,599 21.44
7 G, group III 17,568 64,223 265.57 21,859 24.43

results are presented in Table 7. In addition, an adjusted fatigue life (2) The SeN fatigue curve for the Sn37Pb solder joints is determined
prediction model based on the Steinberg model is proposed here: with the von Mises stresses of the critical solder joints from the FEA
0.00022A b and the vibration lifetime from the sinusoidal vibration tests. The
b
Z ⎛ ct L ⎞ results showed that the SeN fatigue curve is sufficiently accurate
N1 = N2 ⎛ 2 ⎞ = N2 ⎜
⎜ ⎟
⎟
Z
⎝ ⎠3 Z3 for predicting the fatigue life.
⎝ ⎠ (11)
(3) An adjusted fatigue life predict model based on the Steinberg model
In this alternative model, the 3σ displacement Z3 is substituted for was proposed in this paper. The proposed fatigue life prediction
the displacement Z1 calculated from Eq. (9). Because the 3σ displace- model successfully substitutes the 3σ displacement levels at the
ment level is the maximum level expected under the random vibration component locations on the PCB for the PCB displacement eval-
loading condition, Z3 can adequately represent the actual maximum uated by considering only the fundamental resonant mode. The
PCB displacement under random vibration loading. proposed adjusted Steinberg model is effective in predicting the
Table 8 shows a comparison among the vibration lifetime results fatigue life of solder joints under random vibration condition.
from the random vibration tests, the Steinberg model (Eq. (10)) and the (4) A method for the vibration lifetime prediction of the PBGA solder
adjusted lifetime prediction model (Eq. (11)). The fatigue life predicted joint based on the SeN fatigue curve, FEA and an adjusted
by Eq. (10) does not agree well with the fatigue life results from the Steinberg fatigue life prediction model was proposed. The com-
random vibration tests. The error percentages for the fatigue life pre- parison between the random vibration test results and the predicted
dicted by Eq. (10) compared with the lifetime of the vibration tests fatigue life demonstrates the validity of the methodology.
exceed 200% and, in some cases, 400%, indicating that the Steinberg
model (Eq. (10)) is ineffective. By contrast, the fatigue life results ob- To close, it is necessary to mention that we did not consider non-
tained from Eq. (11) indicate a good vibration lifetime prediction of the linear deformation in the finite element analysis in this study, and the
solder joints under different Grms-level tests. The error percentages for next goal is to apply elastic-plastic material model for the solder joints
Grms = 7 G, Grms = 11 G and Group III in Grms = 18 G loadings are all in the future modeling work.
within 50%. Only in the case of two groups in Grms = 18 G does the
error percentages exceed 100%. These results show that the adjusted Declaration of competing interests
fatigue life prediction model shown in Eq. (11) is sufficiently accurate
for predicting the vibration lifetime of Sn37Pb solder joints in PBGA The authors declare that they have no known competing financial
components. interests or personal relationships that could have appeared to influ-
ence the work reported in this paper.
4. Conclusion
Acknowledgment
This study aims to improve the fatigue life prediction of Sn37Pb
soldered PBGA assemblies subjected to random vibration loading on the This research was supported by the National Natural Science
basis of vibration failure tests, FEA, and Steinberg's empirical for- Foundation of China (NSFC) No. 11872078, Beijing Natural Science
mulation. Narrow-band sinusoid vibration tests were conducted to ob- Foundation No. 2182011, and the Beijing Key Laboratory of Advanced
tain the vibration lifetime of solder joints. The finite element model was Manufacturing Technology.
established to obtain the maximum stresses at critical solder joints.
Based on the calculated stresses from the FEA and the fatigue life from References
the sinusoid vibration tests, the SeN fatigue curve of the PBGA solder
joints was obtained. Random vibration tests were conducted to assess [1] R.R. Tummala, E.J. Rymaszewski, A.G. Klopfenstein, Microelectronics Packaging
Handbook, Second edition, Chapman and Hall, New York, 1997.
the vibration lifetime of solder joints under broad-band excitation. The
[2] D.S. Steinberg, Vibration Analysis for Electronic Equipment, Third edition, Wiley,
finite element model was applied to determine the maximum PCB New York, 2000.
displacement under random vibration loading. Finally, an adjusted vi- [3] X.W. Liu, W.J. Plumbridge, Thermomechanical fatigue of Sn-37 wt.% Pb model
bration lifetime prediction model based on the Steinberg model was solder joints, Mater. Sci. Eng. A 362 (2003) 309–321.
[4] B. Zhang, H. Ding, X.J. Sheng, Reliability study of board-level lead-free inter-
used to assess the fatigue life of PBGA solder joints subjected to random connections under sequential thermal cycling and drop impact, Microelectron.
vibration loading. On the basis of the results of this study, the following Reliab. 49 (2009) 530–536.
conclusions can be drawn: [5] S. Ridout, C. Bailey, Review of methods to predict solder joint reliability under
thermo-mechanical cycling, Fatigue Fract. Eng. Mater. Struct. 30 (2007) 400–412.
[6] Q.J. Yang, Z.P. Wang, G.H. Lim, J.H.L. Pang, F.F. Yap, R.M. Lin, Reliability of PBGA
(1) The fatigue life of the solder joints decreases with increasing input assemblies under out-of-plane vibration excitations, IEEE Trans. Compon. Packag.
excitation level. The location of the components strongly influences Technol. 25 (2002) 293–300.
[7] Y. Zhou, M. Al-Bassyiouni, A. Dasgupta, Vibration durability assessment of
the vibration lifetime of solder joints, and the solder joint life de- Sn3.0Ag0.5Cu and Sn37Pb solders under harmonic excitation, Trans. ASME, J.
creases with increasing proximity to the PCB center. Electron. Packag. 131 (011016) (2009).
[8] Y.S. Chen, C.S. Wang, Y.J. Yang, Combining vibration test with finite element

9
T. An, et al. Microelectronics Reliability 102 (2019) 113474

analysis for the fatigue life estimation of PBGA components, Microelectron. Reliab. [20] M.L. Wu, Vibration-induced fatigue life estimation of ball grid array packaging, J.
48 (2008) 638–644. Micromech. Microengineering 19 (65005) (2009).
[9] M.A. Miner, Cumulative fatigue damage, Am. Soc. Mech. Eng. 12 (1945) A [21] T.E. Wong, F.W. Palmieri, B.A. Reed, H.S. Fenger, H.M. Cohen, K.T. Teshiba,
159–164. Durability/reliability of BGA solder joints under vibration environment, Proc.
[10] A. Perkins, S.K. Sitaraman, Analysis and prediction of vibration-induced solder joint Electron. Compon. Technol. Conf. (2000) 1083–1088.
failure for a ceramic column grid array package, Trans. ASME, J. Electron. Packag. [22] S. Wang, Y.H. Tian, X.G. Guo, Fatigue life prediction for CBGA under random vi-
130 (2008) 11012. bration loading by finite element method, Proc. Electron. Packag. Technol. Conf.
[11] F.X. Che, J.H.L. Pang, Vibration reliability test and finite element analysis for flip (2014) 535–537.
chip solder joints, Microelectron. Reliab. 49 (2009) 754–760. [23] P. Yang, Z.X. Chen, Experimental approach and evaluation on dynamic reliability of
[12] C. Basaran, R. Chandaroy, Thermomechanical analysis of solder joints under PBGA assembly, IEEE Trans. Electron. Devices 56 (2009) 2243–2249.
thermal and vibration loading, J. Electron. Packag., Trans. ASME 124 (2002) [24] D. Yu, A.Y. Abdullah, T.T. Nguyen, S. Park, S. Chung, High-cycle fatigue life pre-
60–66. diction for Pb-free BGA under random vibration loading, Microelectron. Reliab. 51
[13] F. Liu, Y. Lu, Z. Wang, Z.M. Zhang, Numerical simulation and fatigue life estimation (2011) 649–656.
of BGA packages under random vibration loading, Microelectron. Reliab. 55 (2015) [25] Y. Zhou, M. Al-Bassyiouni, A. Dasgupta, Harmonic and random vibration durability
2777–2785. of SAC305 and Sn37Pb solder alloys, IEEE Trans. Compon. Packag. Technol. 33
[14] J. Xia, G.Y. Li, B. Li, L.X. Cheng, B. Zhou, Fatigue life prediction of package-on- (2010) 319–328.
package stacking assembly under random vibration loading, Microelectron. Reliab. [26] A. Dehbi, Y. Ousten, Y. Danto, W. Wondrak, Vibration lifetime modelling of PCB
71 (2017) 111–118. assemblies using Steinberg model, Microelectron. Reliab. 45 (2005) 1658–1661.
[15] T. An, C. Fang, F. Qin, H.C. Li, T. Tang, P. Chen, Failure study of Sn37Pb PBGA [27] S. Qin, Z.Q. Li, X.B. Chen, H.D. Shen, Comparing and modifying estimation methods
solder joints using temperature cycling, random vibration and combined tempera- of fatigue life for PCBA under random vibration loading by finite element analysis,
ture cycling and random vibration tests, Microelectron. Reliab. 91 (2018) 213–226. IEEE Conf. Progn. Health Manag. (2015) 7380066.
[16] Jr. L. F. Coffin, A study of the effects of cyclic thermal stresses on a ductile metal, [28] IPC Test Standards, IPC-7095C, Design and Assembly Process Implementation for
Transactions American Society of Mechanical Engineers 76 (1954) 931–950. BGAs, IPC-Association Connecting Electronics Industries, (2013).
[17] S.S. Manson, Fatigue: a complex subject-some simple approximation, Exp. Mech. 6 [29] IPC Test Standards, IPC-SM-785, Guidelines for Accelerated Reliability Testing of
(1965) 193–226. Surface Mount Attachments, IPC-association connecting electronics industries,
[18] O.H. Basquin, The exponential law of endurance tests, American Society of 1992.
Mechanical Engineers Proceedings (1910) 625–630. [30] W. Weibull, S. Sweden, A statistical distribution function of wide applicability, Am.
[19] M. Matsuishi, T. Endo, Fatigue of metals subjected to varying stress-fatigue lives Soc. Mech. Eng. 18 (1951) 293–297.
under random loading, Proc. Kyushu District Meeting, 1968, pp. 37–40.

10