Fatigue Response of Polyimide Thin Film under Cyclic Loading

Yu-Chen Chang1, Tz-Cheng Chiu1,*, Yu-Ting Yang2, Yi-Hsiu Tseng2, Xi-Hong Chen2, Pu-Shan Huang2
1
National Cheng Kung University
1 University Rd., Tainan, 701, Taiwan, ROC
*[email protected]
2
ASE Group Chung-Li
550 Chung-Hwa Rd., Sec. 1, Taoyuan 320, Taiwan, ROC

ABSTRACT assessing the PI dielectric cohesive cracking reliability in

microelectronic components is lacking. This study therefore focused
Fatigue characteristics of polyimide thin film strips prepared with on characterizing the fatigue response of PI thin film, and developing
realistic wafer-level redistribution processes were investigated a prediction model for assessing the failure life of PI thin film under
experimentally. Uniaxial tensile tests were first conducted on the thin cyclic loading condition. Uniaxial tensile experiments were first
film specimens to characterize the stress-strain relationship, and to performed on free-standing PI thin films to determine its basic
measure ultimate strength and elongation. Both strain- and stress- mechanical properties including Young’s modulus, and elongation at
controlled fatigue cycling experiments were then performed. Under film breakage. Strain- and stress-controlled fatigue experiments were
strain-controlled cyclic fatigue loading an obvious stress relaxation then conducted under test conditions based to the tensile test findings.
behavior was observed. The stress relaxation characteristic depends The fatigue experimental results were fitted statistically by using
only on the applied strain range, but not on the level of the average physics based mathematical models, which can be applied to
strain. Under stress-controlled cyclic fatigue loading the polyimide estimate the cycles to PI thin film cohesive fracture, and can be used
thin film exhibited both viscoelastic and plastic responses, and the as the basis of reliability prediction for PI-based interconnect
peak plastic strain followed a power-law increasing trend as the structures.
fatigue cycle increased. A fatigue strain evolution model was
developed by statistically fitting the stress-controlled fatigue EXPERIMENTAL
responses with a physics-based mathematical model. The fatigue
response model can be applied for developing design rules for The PI thin films used in this study are of dimension either
extending the polyimide based redistribution interconnect technology 20 mm × 4 mm, 20 mm × 5 mm, or 40 mm × 10 mm, with average
to the regimes of larger chip size and higher input/output densities. thickness around 5 µm. These films are fabricated on Si substrate by
using a typical photolithography process used for redistribution layer
INTRODUCTION process. After curing the PI film strips were separated from the
substrate by using HF. The glass transition temperature of the PI film
Polyimide (PI) is an engineering material with the advantages of is around 320 °C. Uniaxial tensile and fatigue tests of the PI thin film
low dielectric constant, high mechanical compliance, and good high- were performed by using an Instron 5565 tester with a low force load
temperature stability. PI thin films have been used extensively in cell. A custom test fixture with rubber shim is designed to hold the
flip-chips (FCs) and wafer-level packages (WLPs) as the passivation specimen. To improve film specimen handling and alignment in the
material for silicon die surface redistribution interconnect and bump tester, the PI film strip was first mounted on a 100 µm-thick PI
structures. While polyimide is clearly a favorable solution for die window frame, which was cut before the uniaxial test started. Shown
surface interconnect passivation, extending the technology into larger in Fig. 1 is the PI film specimen assembled in the test fixture. To
chip size and finer input/output (I/O) pitch is complicated by the evaluate the effects of film dimensions on the measurement results,
increasing risk of PI cracking due to thermal or mechanical stresses uniaxial tensile tests under room temperatures were first performed
during temperature cycling or drop conditions. Consequently, an with PI thin films of all 3 different in-plane dimensions. The stress-
important task in developing redistribution interconnect technology
is to characterize the resistance of PI thin film to cracking failure
such that the mechanical reliabilities of FC or WLP interconnect can
be accurately assessed and optimized. Because in the
microelectronics applications the PI thin films are boned to dissimilar
material surfaces, many of the PI related studies were focused on
adhesive properties. For example, Kim and Allen [1] investigated the
residual stress of PI thin film and the durability of the polyimide-
metal interface. The fatigue behavior of Cu thin films on PI
substrates was investigated by Kraft et al. [2]. Alaca et al. [3]
measured the mode I stress intensity factor of PI thin films on
compliant substrates. Li and Jiao [4] studied the effect of
temperatures and aging on the elastic property of PI by using
dynamic mechanical analysis. These studies focused on the basic
mechanical properties of PI and the interaction of PI thin film and
other materials. However, the fatigue characteristic of PI thin film for FIGURE 1. The PI thin film assembled in the test fixture

233
▲

IEEE Catalog Number: CFP1559B-ART ISBN: 978-1-4673-8356-1

 200 160

150 120

σ (MPa)
σ (MPa)

100 80

N=1
40 N = 10
50 4 mm
5 mm N = 100
10 mm N = 1000
0
0
0 5 10 15 20 0 1 2 3 4 5

ε (%) ε (%)
FIGURE 2. Effect of specimen width on tensile test results FIGURE 4. Evolution of hysteresis loops over cycles in
the 0.1 Hz, 0-4% strain-controlled cyclic test

200
200 4~6 %
4.2×10-3 1/s
6~8 %
150 8~10 %
150 10~12 %
σ (MPa)
σ (MPa)

100
100

50
50 25 C
25 °C
75 C
75 °C
C
125 °C 0
0 10-1 100 101 102 103 104 105 106
0 5 10 15 20
N
ε (%)
FIGURE 3. PI thin film responses under uniaxial tensile loads FIGURE 5. Evolution of stress ranges in the strain-controlled
fatigue tests
TABLE 1. PI thin film tensile properties

Parameter 25 °C 75 °C 125 °C characteristics as shown in Fig. 1 were also used as the base for both
room-temperature strain- and stress-controlled fatigue experiments.
Young’s Modulus (GPa) 4.19 3.71 2.84
In the strain-controlled fatigue experiment, cycling frequencies in
elongation at rupture (%) 19.5 18.2 19.3 the range of 0.1 Hz and 0.5 Hz were selected because the
ultimate strength (MPa) 167.8 142.5 96.5 corresponding strain rates are in the same order as those experienced
in the temperature cycling tests. The cycling strain ranges considered
were between 0 and a peak value between 1% and 10%. Total
strain responses of these tests are shown in Fig. 2, from which it can number of cycles performed for each loading condition was about
be seen that the differences in the test responses for specimens of 105. A typical stress-strain response of the strain-controlled fatigue
different in-plane dimensions may be negligible. In addition, test is shown in Fig. 4, in which the hysteresis loops at cycle numbers
displacement measurements based on the read-outs of cross-head and 1, 10, 100 and 1000 are presented. It can be seen from Fig. 4 that the
extensometer mounted on the test fixture are almost identical. Based cyclic stress range relaxes as the fatigue cycle increases. In addition,
on these observations, the effect of in-plane thin film dimensions the area of the hysteresis loop decreases rapidly as the number of
were not considered in subsequent tensile and fatigue test results. fatigue cycle increases, and is very close to zero as the cycle number
To determine the mechanical properties of the PI thin film, a reaches beyond the range of 103. It implies that, under fixed cyclic
series of uniaxial tensile tests were performed at constant strain range, the irreversible inelastic response of the PI thin film
temperatures between 25 °C and 125 °C with a constant strain rate of saturates quickly and the behavior of PI thin film as cycle number
4.2×10-3 1/s. Results of the tensile experiments are shown in Fig. 3, increases is governed by viscoelastic stress relaxation. The effect of
from which it can be seen that the compliance increases as mean strain level on the fatigue response is shown in Fig. 5, in which
temperature increases. The Young’s modulus, ultimate strength and the stress range changes over cycle numbers are given. It can be seen
elongation obtained from the tensile test are given in Table 1, from from Fig. 5 that, under a constant cyclic strain range (2%), the
which it can be seen that the Young’s modulus decreases 32 % to as relaxation trends of the maximum and minimum stresses are very
the temperature increases from 25 °C to 125 °C; the elongation, on similar and independent of the mean strain when the mean strain
the other hand, remains constant at around 19 %. The stress-strain value is 5% or higher.

234
▲

IEEE Catalog Number: CFP1559B-ART ISBN: 978-1-4673-8356-1

 μ1
10
130MPa
110MPa
8
90MPa μ2 η
60MPa
6
ε (%)

FIGURE 7. The Zener viscoelastic model

4
TABLE 2. Material parameters of the viscoelastic model
2
Parameter Value
µ1 1236 MPa
0
10-1 100 101 102 103 104 µ2 670.7 MPa
N t 18288 s

FIGURE 6. Envelope of the fatigue cycle peak strains in TABLE 3. Material parameters of the plastic model
the stress-controlled fatigue experiment
Parameter Value
The stress-controlled fatigue cycles were performed under room σ0 60 MPa
temperature with cyclic stress ranges between a minimal value of 5 C 6.283×10-6
MPa and a maximum value between 60 MPa and 130 MPa. Shown in n 9.56
Fig. 6 are the envelopes of the maximum strains measured in every D 5.603×10-4
fatigue cycles. It can be seen from Fig. 6 that the peak strain value m 0.221
increases monotonically as the fatigue cycle increases, and that the
increasing rate of the envelope strain accelerates as the fatigue cycles
reach beyond 103. Furthermore, the increasing trends in the envelope From the Boltzmann superposition principle, the viscoelastic strain
strain vs. cycle relationships for the peak stress values of 110 MPa response can be expressed as
and 130 MPa are more significant than those for the peak stress
values of 60 MPa and 90 MPa. The different trends are consistent t dσ ( ξ )
ε ve ( t=
) σ ( 0) J (t ) + ∫ J (t − ξ ) dξ (3)
with the finding from the uniaxial tensile experimental results (Fig. 0+ dξ
3), from which it can be seen that the response of the PI thin film
changes from linear to nonlinear at stress level around 100MPa. where σ denotes the uniaxial stress. By approximating the cyclic
fatigue stress loading as
FATIGUE MODEL
σ (=
t ) σ m + σ a sin(ωt ) (4)
From the experimental findings presented in the previous section,
it can be seen that the low-cycle fatigue failure is unlikely to occur
where σ m and σ a are the mean and amplitude stresses respectively.
under the stress- or strain-ranges considered in this study. As an
alternative approach, the evolution of the envelope strain is By substituting (2) and (4) into (3), the viscoelastic strain response
considered as the damage accumulation index, and fatigue failure can be expressed as
would occur when the envelope strain reaches a critical value.
− µ1t
For considering the PI thin film behavior under stress-controlled σm    
cyclic loading, it is assumed that the deformation consists of both ε ve ( t )
=  µ1 + µ 2 1 − e
(µ + µ 1 2 )t

viscoelastic and plastic responses. The strain can therefore be
µ1 ( µ1 + µ 2 )    

expressed as σa
+ 2
µ1 + ( µ1 + µ 2 ) t 2ω 2 
2 {
 µ1 + ( µ1 + µ 2 )t 2ω 2  sin(ωt ) (5)
ε ε ve + ε p
= (1)
− µ1t

− µ 2tω cos(ωt ) + µ 2tωe (
µ1 + µ2 )t
ve p
where the ε and ε denote the viscoelastic and plastic strains, 
respectively. The room-temperature viscoelasticity behavior of the PI 
thin film is considered by using a Zener model as depicted in Fig. 7.
The creep compliance of the Zener viscoelastic model can be The plastic component of the envelope strain under stress-
expressed as controlled fatigue cycling is considered by using a power-law
relationship to model the plastic strain, i.e.
  µ1t  
µ1 + µ 2 1 − exp  −  C (σ max σ 0 ) + DN (
εp = mσn σ0 )
, σ max > σ 0 (6)
 ( µ1 + µ 2 )t  
max


J (t ) = (2)
µ1 ( µ1 + µ 2 )
where C, D, m, n are material parameters, N is the cycle number, σ 0

where t is time, µ1 , µ 2 , η and t = η µ 2 are material parameters. is the yield stress above which the plastic deformation occurs, σ max
is the peak stress, and σ max
= σ m + σ a . The first term on the right

235
▲

IEEE Catalog Number: CFP1559B-ART ISBN: 978-1-4673-8356-1

 12 60
Exp. 60 MPa
model 90 MPa
45 110 MPa
9
130 MPa

ε (%)
ε (%)

6 30
130 MPa
3 110 MPa 15
90 MPa
60 MPa
0 0
10-1 100 101 102 103 104 10-1 100 101 102 103 104 105 106 107

N N
FIGURE 8. Comparison between the experimental results and FIGURE 9. Evolution of envelope strain under stress-controlled
fitted model fatigue cycling

hand side of (6) represents the initial plastic strain when the stress
reaches σ max in the first cycle, and the second term corresponds to ACKNOWLEDGMENTS
the subsequent plastic strain increment as the fatigue cycle
progresses. By combining (5) and (6) and best-fitting to the This study was partially supported by ASE Group and by the
experiment results given in Fig. 6, the values of the material Ministry of Science and Technology, ROC, under the grant MOST
parameters used in the viscoelastic and plastic models are given in 104-2221-E-006-056-MY2.
Tables 2 and 3. The R-squared value of the model fitting is 0.98.
Shown in Fig. 8 is the fitted model, which compares well to the REFERENCES
experiment data.
By using the strain model (1) with the model parameters given in [1] Y. J. Kim and M. G. Allen, “In situ measurement of
Tables 2 and 3, the envelope strain growth as the number of fatigue mechanical properties of polyimide films using
cycle increases is shown in Fig. 9. Assuming that the PI thin film micromachined resonant string structures,” IEEE Transactions
ruptures at the elongation strain value of 19 %, fatigue failure would on Components and Packaging Technology, Vol. 22, pp. 282-
not occur under 106 fatigue cycles with 90 MPa peak stress because 290, 1999.
the strain only reaches 9.3%, as can be seen in Fig. 9. On the other [2] O. Kraft, R. Schwaiger and P. Wellner, “Fatigue in thin films:
hand, under fatigue cycles with maximum stress values of 110 MPa lifetime and damage formation,” Materials Science and
and 130 MPa, fatigue failure would occur around 106 and 105 cycles, Engineering, pp. 919-923, 2001.
respectively. [3] B. E. Alaca, J. C. Selby, M. T. A. Saif and H. Sehitoglu,
“Biaxial testing of nanoscale films on compliant substrates:
CONCLUSIONS Fatigue and fracture,” Review of Scientific Instruments, Vol.
73, pp. 2963-2970, 2002.
In this study, both monotonic and cyclic tensile experiments were [4] R. S. Li and J. Jiao, “The effects of temperature and aging on
conducted to determine the mechanical and fatigue characteristics of Young’s moduli of polymeric based flexible substrates,” The
a 5 µm-thick PI thin film. Some conclusions are given as follows: International Journal of Microcircuits and Electronic
1. As the temperature increases from 25 °C to 125 °C, Young’s Packaging, Vol 23, pp. 456-461, 2000.
modulus of the thin film decreases from 4.19 to 2.84 GPa, but the
maximum elongation remains constant at around 19 %.
2. The PI thin film exhibited significant viscoelasticity relaxation
under room-temperature strain-controlled fatigue experiments.
Fatigue failure did not occur after 105 cycles when the cyclic peak
strain was around or lower than 10 %.
3. Under room-temperature stress-controlled fatigue condition, both
viscoelastic and plastic responses were observed. Plastic deformation
and strain accumulation occurred when the cyclic peak stress was set
at higher than 60 MPa.
4. A fatigue strain evolution model was developed to predict the
maximum strain as a function of fatigue cycle number. The
maximum strain can be used as the damage index to evaluate fatigue
life of the PI thin film.

236
▲

IEEE Catalog Number: CFP1559B-ART ISBN: 978-1-4673-8356-1