ck
k
y o
y, Q
in
ne
to
al
al
igat
rac
can take a very long time and it would be very useful to
the magnitude of cracking within the two Sn–3.5Ag solder
joints of a 2512 surface mount resistor (6 mm long, 2.6 mm
wide, 0.55 mm high), mounted at the centre of a strip of
each of the tests under investigation. Modelling methods
the cracks caused by thermal cycling. In a thermal cycling
test, the specimen will be placed in an oven where the tem-
perature will be repeatedly cycled between two extremes.
An example of a commonly used thermal cycle is 55 C
to +125 C with a period of 45 min, however many varia-
tions are used – e.g. a shorter period may be chosen in
* Corresponding author. Tel.: +44 20 8331 8761.
E-mail address: s.w.ridout@gre.ac.uk (S. Ridout).
Microelectronics Reliability 46determine the amount of damage which has been done
to the joint before it ultimately fails. One method used is
to cut the sample in half and inspect its cross-section with
a microscope [1–3]. Another method is to use dye penetra-
tion. In this way, the fatigue cracks can be observed and
measured. However, these methods are destructive to the
component and the solder joints.
The tests presented in this paper are intended to be a
quicker and cheaper, non-destructive method to detect
are then discussed, before presenting the results of the sim-
ulations. For each of the tests investigated, a prediction is
given of the sensitivity of each test to the different kinds
of cracks which occur.
2. Cracks caused by thermal cycling
In order to model the crack detection tests presented in
this paper, it is first necessary to determine the geometry of1. Introduction
Accelerated life tests are a widely used method in indus-
try to assess the reliability of soldered assemblies. Typi-
cally, a component will be exposed to a number of
environmental loading conditions (thermal, mechanical,
etc.) and monitored for electrical continuity. However, this
approach will only provide the time to complete failure, i.e.
when a crack has grown completely through the joint. This
FR4 PCB (360 mm long, 5 mm wide, 1.3 mm thick). We
envisage these tests being used to periodically test for dam-
age during accelerated testing, this will allow the growth of
a crack though a single specimen to be recorded which is
not possible using destructive methods such a cross-sec-
tioning. The tests are not intended to detect damage on a
PCB with many components.
The paper will first discuss the kinds of cracks which are
caused by thermal cycling, following this is a description ofAssessing the performance of cra
Stephen Ridout a,*, Milos Duse
a School of Computing and Mathematical Sciences, Universit
b NPL Materials Centre, National Physical Laborator
Received 19 May 2005; received
Available onli
Abstract
This paper presents both modelling and experimental test data
focus is on determining the presence and rough magnitude of therm
on a strip of FR4 PCB. The tests all operate by applying mechanic
the resistor. The modelling results show that of the four tests invest
magnitude. Hence these tests show promise in being able to detect c
these results although more validation is required.
 2006 Elsevier Ltd. All rights reserved.0026-2714/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.microrel.2006.05.001f Greenwich, Park Row, Greenwich, London SE10 9LS, UK
ueens Road, Teddington, Middlesex TW11 0LW, UK
revised form 25 October 2005
22 June 2006
characterise the performance of four non-destructive tests. The
fatigue cracks within the solder joints for a surface mount resistor
loads to the PCB and monitoring the strain response at the top of
ed, three are sensitive to the presence of a crack in the joint and its
king caused by accelerated testing. The experimental data supportsdetection tests for solder joints
b, Chris Bailey a, Chris Hunt b
www.elsevier.com/locate/microrel
(2006) 2122–2130

in the joints since the presence of a crack will affect the way
the load (and therefore strain) is distributed through the
specimen.
S. Ridout et al. / Microelectronics Reliability 46 (2006) 2122–2130 2123order to damage the joints more rapidly, or a milder tem-
perature range may be chosen to better represent the
intended use conditions.
A number of experiments have been reported in which
thermal cycled joints have been cross-sectioned [1–3], the
results show a number of different directions of crack prop-
agation, three of which are illustrated in Fig. 1. It is possi-
ble that there may be a correlation between the thermal
cycle used and the kind of crack generated. For instance,
recent work at the NPL (National Physical Laboratory,
UK) has indicated that with very fast ramp rates (thermal
shocking), a delamination of the solder from the compo-
nent’s vertical surface in the fillet region is observed. And
with slower ramp rates, cracks tend to form within the bulk
of the solder. However more evidence is required to con-
firm this phenomenon. For the purposes of this work, it
will be assumed that all three of the crack directions shown
in Fig. 1 are possible.
3. Test methods
As discussed above, the traditional manner of testing
solder joints in components during an accelerated test is
to obtain a cross-section for examination. A limitation of
this approach is that the crack is seen only as a 1D line,
not a 2D area. If the crack propagation direction was in
Fig. 1. Crack propagation directions.the plane of the SEM images then this would not be a prob-
lem, however dye-penetration experiments at the NPL have
shown this not to be the case. FEA simulations performed
at Greenwich using a simple damage model support these
results and show the prominent direction of crack propaga-
Fig. 2. Crack propagation in stand-off region (Ltion to be in the out-of-plane direction with respect to the
SEM images – as shown in Figs. 2 and 3. This implies that
the crack propagation time through the stand-off region as
measured from a cross-section [1] is significantly underesti-
mated. In most cross-sections the cracks are not symmetri-
cal, for example see Fig. 4.
The following tests have been designed so that cross-sec-
tioning is not required. All the tests work in a similar way,
by applying a mechanical loading to the specimen and
monitoring the strain response at the top of the resistor.
The strain in the direction of the longest component side
is measured with a strain gauge which is glued to the resis-
tor as seen in Fig. 5.
3.1. Pull test
The test specimen is subjected to a 100 N tensile load
using the apparatus shown in Fig. 5 and the strain on the
resistors top surface is monitored with a strain gauge. This
measured strain will depend on the magnitude of the cracks
Fig. 3. Result showing damage build-up in a 1206 resistor joint after
thermal cycling.3.2. 3-point bend test
In this test, displacements are imposed on the test spec-
imen at three points causing it to bend as illustrated in
HS – actual shape, RHS – idealised shape).

of
2124 S. Ridout et al. / Microelectronics Reliability 46 (2006) 2122–2130Fig. 4. An SEM image showing unsymmetrical crackingFig. 6. The apparatus used for this test is the modified
micrometer screw gauge shown in Fig. 7, the central point
rotates as the screw gauge is turned.
3.3. 4-point bend test
In this test, 4 displacements are imposed on the test
specimen causing it to bend in the opposite way to the 3-
point bend test, this is illustrated in Fig. 8. No experimental
apparatus has been built for this test, only modelling
results are presented.
Fig. 5. The pull test apparatus (a tensile testing machine).
Fig. 6. The 3-point bend test.the two joints in a component after 1200 thermal cycles.3.4. Reverse 3-point bend test
In this test, a similar deformation mode to the 4-point
bend test is produced by using a central point which is
pulled as shown in Fig. 9. Again, there is no experimental
apparatus built to perform the test, but modelling work has
been conducted to predict its sensitivity to cracks.
Fig. 7. The 3-point bend test apparatus (a modified micrometer screw
gauge).
Fig. 8. The 4-point bend test.
Fig. 9. Reverse 3-point bend test.

4. Finite element analysis
In order to predict the sensitivity of the various tests
described above, many FEA models were constructed, each
with a different crack length. An example of the FEA mesh
used is shown in Fig. 10, slight variations to this model
were required when introducing the different sized cracks.
The correct boundary conditions were imposed to simulate
each of the tests and the strain at the top of the resistor (the
location of the strain gauge) was recorded. The multi-phys-
ics code PHYSICA [4] was used to run all the simulations.
It was assumed in this study that the cracks in both
joints would be similar and the jagged details that would
be seen in a real joint are avoided. Although it may be
the case that only one of the two joints has a crack, for
the purposes of the sensitivity analysis both joints are
assumed to have identical cracks. This is not generally true
in practice [1] and further work would be necessary to pre-
is Boltzmann’s constant. The constants used are shown in
Table 2 and are from Ref. [6]. Although Sn–3.5Ag proper-
ties are used in this investigation, it is reasonable to expect
the predicted trends to be similar for other solders.
It is also assumed that the loading applied during the
test opens up the crack. Therefore contact analysis
between the crack interfaces was not performed. In certain
Fig. 11. The materials used in the model.
Table 1
Material properties
Material Young’s modulus
(GPa)
Poisson’s
ratio
Coefficient of thermal
expansion (ppm/C)
S. Ridout et al. / Microelectronics Reliability 46 (2006) 2122–2130 2125dict the effect of asymmetric cracking on these tests.
The materials which the specimen is made of are shown
in Fig. 11. For all the materials except the solder, the
behaviour can be assumed to be linear elastic at room tem-
perature. The material properties used are shown in Table
1. Note that two Elastic moduli are used for the FR4. Var-
ious sources report a modulus of between 12 GPa and
28 GPa. In this work, simulations have been performed
at 12.4 GPa (as measured by the NPL), and 22 GPa [5],
the effect of the different values is discussed later.
The behaviour of the solder is more complex. It is
known to exhibit instantaneous plasticity and creep at
room temperature. For most of the results presented, this
plasticity is ignored and the solder is treated as a linear-
elastic material. But a number of simulations have been
performed using a constitutive law for creep, the results
of which are compared to the elastic results. The constitu-
tive law used for creep in the Sn–3.5Ag solder is given in
below [6]Fig. 10. The Fdes
dt
¼ A sinhðarÞ½ n exp Qa
kT
 
where des/dt is the steady-state strain rate, A, a, n and Qa
are constants, r is the stress, T is the temperature, and k
FR4 12.4 or 22a 0.28 18
Copper 121 0.35 17
Palladium 117 0.39 11.5
Alumina 282 0.222 6
Sn–3.5Ag 45.7 0.31 20
a Two different values are used for the FR4 Young’s Modulus.EA mesh.

conditions this is not true and the crack surfaces are pushed
together, this is discussed further in the results section.
5. Results
5.1. Pull test
The elastic simulation with an FR4 Young’s Modulus of
12.4 GPa as measured by the NPL will be presented first.
On loading, the test piece deforms as shown in Fig. 12,
the left-hand-side of the images represents a symmetry
plane, so the position of the strain gauge is at the top left
corner of each diagram. With no crack present the bending
causes the top surface of the resistor to be in compression.
When a crack has grown across the stand-off region the
bending of the resistor is reduced by a small amount, lead-
ing to a 29% change in the strain at the strain gauge. When
the crack grows further, either vertically or diagonally into
the fillet region, then a more dramatic change is seen caus-
ing the strain to eventually turn from compressive to tensile
(>100% change). The sensitivity to the different crack direc-
tions is shown in Fig. 13. Note that in this and the follow-
ing graphs, a crack length of 50% represents a completely
cracked stand-off region. This test is reasonably sensitive
to horizontal cracks in the stand-off and fillet regions,
and very sensitive to vertical and diagonal cracks in the fil-
let region.
There is a problem in predicting the horizontal crack
sensitivity as the results show an overlap between the two
crack surfaces. This means that a contact analysis would
Table 2
Solder creep constants
Solder alloy A (1/s) a (1/MPa) n Qa/k (K)
Sn–3.5Ag 9.0E5 0.0653 5.5 8690.0
2126 S. Ridout et al. / Microelectronics Reliability 46 (2006) 2122–2130Fig. 12. Pull test results showing the strain contours at the symmetry
plane. Deformation exaggerated by 150 times. From top to bottom: no
crack, 100% stand-off crack, vertical crack, diagonal crack, horizontal
crack.be required for a proper prediction. However, by intuition
it can be seen that if the contact between the 2 crack sur-
faces is resisted then the strain at the strain gauge will
change by less than the value predicted in this simulation,
which is already quite low. Therefore, the sensitivity to hor-
izontal cracks is considerably lower than the sensitivity to
vertical and diagonal cracks.
The elastic simulation using a higher Young’s Modulus
of 22 GPa for FR4 will now be discussed. As can be seen
from Fig. 14, the overall shape of the graph is roughly sim-
ilar to that produced using the lower strength FR4. There
are two main differences. First, the magnitude of the strains
is lower in this case. Second, the sensitivity to crack length
in the stand-off region is 13%, this is a considerable
decrease compared to the 29% obtained using
E = 12.49 GPa.
The next result is from a simulation incorporating creep
in the solder and using the lower Young’s modulus of
12.4 GPa for the FR4. From Fig. 15 it is seen that creep
has a negligible effect when no crack is present but becomes
very significant for the larger crack sizes. This has the effect
of increasing the difference in the strains between the differ-
ent crack lengths, thereby improving the sensitivity of the
tests with respect to the elastic results. It is a concern that
significant damage may be done to the solder over theFig. 13. Sensitivity of pull test (FR4 E = 12.4 GPa).

correlation with the FEA prediction in Fig. 13 when no
Fig. 16. Experimental results for pull test.
cs Reliability 46 (2006) 2122–2130 2127Fig. 14. Sensitivity of pull test (FR4 E = 22 GPa).
S. Ridout et al. / Microelectronicourse of the test. The simulation of crack growth is
beyond the scope of this investigation but it is intuitively
obvious that with a large enough crack initially present in
the solder, the high stress concentration generated by the
test in the remaining solder will cause the crack to grow,
possibly completely destroying the joint.
To restrict creep and crack growth, the applied load
could be reduced. This would have the effect of lowering
the strains generated, therefore potentially reducing the
accuracy of the strain gauge measurements. Hence, a load
is required which offers an acceptable compromise between
the accuracy and destructiveness of the test.
This test has been applied experimentally to a number of
specimens exposed to different numbers of thermal cycles.
The results are shown in Fig. 16. It should be noted
that the number of specimens tested was very small – each
of the data points in the graph represents only one test.
Furthermore, each of the data points represents a different
specimen (in future work it would be interesting to test the
same specimen after different numbers of thermal cycles). It
was assumed that all of the samples were similar and there-
fore would all have the same strain gauge reading before
thermal cycling.
The strain gauge reading taken before any thermal
cycles (1st point on the graph in Fig. 16) shows reasonable
Fig. 15. Pull test – effect of creep in solder.crack is present (the correlation is not as good using the
higher value of 22 GPa for the FR4 Young’s Modulus).
By using the FEA results, the crack length for the thermal
cycled specimens can been predicted. Note that these pre-
dictions are made with the assumption (1) that all materials
apart from the solder have no cracks (e.g. the resistor) and
(2) that the cracks are symmetrical. As the second assump-
tion is unlikely to be true these predictions of crack length
should only be considered a rough estimate. In order to
make the predictions, the FEA results in Fig. 13 were cal-
ibrated so that the strain with no crack present matched the
experimental data. This calibrated graph was cross-refer-
enced with the experimental results in Fig. 16 to provide
Fig. 17, which shows the predicted crack length with
increasing thermal cycles. It seems reasonable to say that
the specimens tested at 900 and 1200 cycles could not have
contained symmetrical horizontal cracks and no fillet
cracks – the strains generated were outside the range which
two horizontal cracks are capable of generating according
to the FEA results (Fig. 13). It is possible that at 900 and
1200 cycles there were either symmetric vertical or diagonal
cracks, or unsymmetrical cracks. It is impossible to deter-
mine the exact length of a crack due to this ambiguity.
5.2. 3-point bend test results
With no crack present the deformation is as shown in
Fig. 18. If a crack is introduced in the stand-off region thenFig. 17. Crack length in experiment as predicted from pull test results.

no crack in solder. (Deformation exaggerated 30 times.)
cs Rno response will be seen because the crack is under com-
pression. A vertical or diagonal crack in the fillet region
would be under tension but it seems unlikely that the crack
surfaces would be able to separate significantly until the
crack was 100% complete. This has not been verified with
FEA simulations because it would require a contact analy-
sis for the parts of the crack in compression. In conclusion,
this test is definitely not sensitive to stand-off cracks at all,
and is unlikely to be sensitive to fillet cracks.
5.3. 4-point bend test results
Unlike the 3-point bend test, the 4-point bend test causes
the initial stand-off crack to be under tension. It will there-
fore be pried apart and will generate a large response in the
Fig. 18. Deformation mode in 3-point bend test with
2128 S. Ridout et al. / Microelectronistrain gauge reading as can be seen in Fig. 19 (these results
use the NPL measured value of 12.4 GPa for the FR4
Young’s Modulus). The test shows good sensitivity to hor-
izontal cracks. Unfortunately though, there is no sensitivity
to vertical or diagonal cracks in the fillet region. For the
diagonal crack the lack of sensitivity could be determined
from the FEA results shown in Fig. 20. For the vertical
crack, although Fig. 20 shows a drop in strain during verti-
cal crack propagation, it is seen from Fig. 19 that the crack
surfaces are being compressed and are overlapping. There-
fore, it is reasonable to assume that the true result will be
similar to the case where there is no vertical crack present
– which makes this test insensitive to vertical cracks.
When the FR4 strength is increased to 22 GPa (the
value found in the literature [5]) then the magnitude of
the strains is increased (Fig. 21). Other than that, the shape
of the curve is very similar to that given for the weaker
FR4, meaning that the strength of the FR4 does not greatly
affect the sensitivity of the test.
As for the pull test, 3 simulations were performed using
a constitutive law for creep and the Young’s Modulus of
12.4 GPa for the FR4. The response of the strain on theeliability 46 (2006) 2122–2130resistor is shown in Fig. 22. And just as for the pull test,
creep in the solder actually improves the sensitivity of this
Fig. 19. 4-point bend test results showing the strain contours at the y = 0
symmetry plane. Deformation exaggerated by 150 times. From top to
bottom: no crack, 100% stand-off crack, vertical crack, diagonal crack,
horizontal crack.

test, but at the expense of permanently deforming the sol-
der, invalidating its description as a non-destructive test.
The results presented above indicate that the test could
be useful in determining the length of a crack in the
stand-off region, or a horizontal crack in the fillet region.
No experiments have been performed on this test.
5.4. Reverse 3-point bend test results
The results are shown in Figs. 23–25 and everything that
has been said regarding the 4-point test results applies to
Fig. 20. Sensitivity of 4-point bend test (FR4 E = 12.4 GPa).
Fig. 21. Sensitivity of 4-point bend test (FR4 E = 22 GPa).
Fig. 23. Sensitivity of reverse 3-point bend test (FR4 E = 12.4 GPa).
Fig. 24. Sensitivity of reverse 3-point bend test (FR4 E = 22 GPa).
S. Ridout et al. / Microelectronics Reliability 46 (2006) 2122–2130 2129Fig. 22. 4-point bend test – effect of creep. Fig. 25. Reverse 3-point bend test – effect of creep.

Table 3
The sensitivity of all the tests
Test type FR4 Young’s modulus (GPa) Sensitivity to different crack shapes (crack length in parentheses)
Stand-off (50%) Vertical fillet (75%) Diagonal fillet (75%) Horizontal fillet (75%)
Pull test 12.4 29% 109% 70% 49%
Pull test 22 13%
tely
is th
2130 S. Ridout et al. / Microelectronics Reliability 46 (2006) 2122–2130this. It shows good sensitivity to stand-off cracks and hor-
izontal fillet cracks, but vertical and diagonal fillet cracks
have little effect on the strain gauge reading as the cracks
are under compression.
6. Conclusion
Table 3 shows a comparison of the sensitivities of all the
tests to the various directions of crack propagation. The
sensitivity is defined as the relative change in the strain
gauge reading when compared to the case of no crack.
The 3-point bend test has been shown not to work but
the other three tests have a good possibility of working.
The 4-point test and reverse 3-point test have shown good
sensitivity to initial stand-off cracks and horizontal fillet
cracks. The pull test has shown some sensitivity to stand-
off cracks and good sensitivity to vertical or diagonal fillet
cracks. Therefore, a combination of the pull test along with
either the reverse 3-point or 4-point test should allow any
of the possible stand-off and fillet cracks to be detected.
Further modelling work could be conducted to determine
the response of the test to an unsymmetrical crack.
The NPL measured Young’s Modulus of 12.4 GPa for
the FR4 provided a better match between simulation and
experiment than the 22 GPa value quoted in the literature
3-Point 12.4 or 22 n/a (contact)
4-Point 12.4 125%
4-Point 22 123%
Reverse 3-point 12.4 145%
Reverse 3-point 22 143%
As for the earlier sensitivity graphs, a 50% crack length indicates a comple
* An asterisk indicates that there is a crack contact and the value given[5].
All three tests require further experimental validation
with specimens whose crack length is known. Such an
investigation could also determine how destructive the tests
are – whether they cause the cracks to grow significantly
when using different forces (pull test) or displacements
(bend tests). This would help to find the optimum forceor displacements to apply in the tests – an acceptable com-
promise needs to be made between the size of the strains
generated and the damage done to the solder joints for
the tests to be classified as non-destructive.
Acknowledgements
The work was carried out as part of a project in the
Materials Processing Metrology Programme of the UK
Department of Trade and Industry. Other sponsors include
the EPSRC (Engineering and Physical Sciences Research
Council), Prime Faraday and the NPL (National Physical
Laboratory) in the UK.
References
[1] Shangguan D. Analysis of crack growth in solder joints. Solder Surf
Mount Technol 1999;11(3):27–32.
[2] Stam FA, Davitt E. Effects of thermomechanical cycling on lead and
lead-free (SnPb and SnAgCu) surface mount solder joints. Microelec-
tron Reliab 2001;41:1815–22.
[3] Andersson C, Andersson D, Tegehall PE, Liu J. Effect of different
temperature cycle profiles on the crack propagation and microstruc-
tural evolution of real lead free joints of different electronic compo-
nents. In: Proceedings of the 5th EuroSimE conference, Brussels,
Belgium, 2004. p. 455–64.
117% 58% 26%
n/a (contact) n/a (contact) n/a (contact)
125%* 125% 169%
123%* 126% 166%
145%* 145%* 191%
143%* 143%* 181%
cracked stand-off region.
e estimated sensitivity, assuming no shearing between the crack surfaces.[4] PHYSICA: Software for multi-physics simulations, Greenwich Uni-
versity. Available from: http://physica.gre.ac.uk.
[5] Lau J, Chang C, Lee R, Chen TY, Cheng D, Tseng TJ, et al. thermal-
fatigue life of solder bumped flip chip on micro via-in-pad (VIP) low
cost substrates. In: Proceedings of the NEPCON-West 2000, Anaheim,
CA. p. 554–62.
[6] Darveaux R, Banerji K, Mawer A, Dody G. Reliability of plastic ball
grid array assembly. In: Lau J, editor. Ball grid array technology. New
York: Mc-Graw Hill; 1995. p. 379–441.