Engineering Fracture Mechanics 73 (2006) 2614–2626

www.elsevier.com/locate/engfracmech

Modeling of multiple cracking and decohesion of a thin film

on a polymer substrate
N.E. Jansson, Y. Leterrier, J.-A.E. Månson *

Laboratoire de Technologie des Composites et Polymères (LTC), Ecole Polytechnique Fédérale de Lausanne (EPFL),
CH-1015 Lausanne, Switzerland

Received 17 May 2005; received in revised form 23 March 2006; accepted 21 April 2006
Available online 21 June 2006

Abstract

Thin brittle films on polymer substrates are finding increasing use as gas barriers for example in the medical and food
packaging industries and also for the next generation of ultra-light displays based on flexible polymer substrates. In order
to determine the durability of the barrier under thermal and mechanical loads, test procedures and corresponding data
reduction methods are needed to feed the analysis models. One of the tests frequently employed for this kind of multi-layer
material systems is the fragmentation test, whose designation comes from the progressively denser pattern of parallel
cracks developing when the specimen is loaded under uniaxial tension. From the crack-density versus strain data obtained,
a critical strain for crack growth and an assessment of the adhesion of the coating to substrate can be obtained. However,
no accepted data reduction methods exist to extract material properties from the test or inversely, successfully predict the
crack density as a function of a set of material properties without fitting parameters. In an earlier paper, the authors pre-
sented a finite element based analysis methodology to determine the fracture toughness of both the coating and the inter-
face from the fragmentation data. In the simulations, the plastic constitutive behavior of the substrate and the debonding
of the coating from the substrate were explicitly included, the latter by use of a cohesive zone model. In this paper an exten-
sion of this methodology is presented that enables crack-density evolution with strain to be predicted. The results presented
comprise comparisons with experiments to validate the methodology and the influence of (i) coating toughness, (ii) inter-
face toughness and (iii) coating thickness on crack density versus strain.
Ó 2006 Elsevier Ltd. All rights reserved.

Keywords: Finite element method; Cohesive zone; Adhesion; Polymer substrate; Thin film

1. Introduction

Polymer films with thin barrier coatings are frequently used in the food and medical packaging industry to
protect the content against for instance moisture and oxygen [1,2]. A further potential application area for
such layered composite materials is in the electronics industry for use within flexible, lightweight displays

*
Corresponding author. Tel.: +41 21 693 42 81; fax: +41 21 693 58 80.
E-mail address: jan-anders.manson@epfl.ch (J.-A.E. Månson).

0013-7944/$ - see front matter Ó 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfracmech.2006.04.013
 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626 2615

Nomenclature

a, b prefactors (Eqs. (16) and (19))

c compression factor (Eq. (10))
B damage function (Eqs. (8), (13) and (14))
cd, cdsat crack density, and crack density at fragmentation saturation
F applied force
G shear modulus of the interface
Gc coating fracture toughness (Gc(x) and Gc are vectors (Eq. (6)))
GiI ; GiII interfacial mode I and mode II energy release rates
exp i
Gc experimentally determined fracture toughness (for calibration)
Gcc coating fracture toughness
Gic interface fracture toughness
hc coating thickness
h0c normalization term (Eqs. (16) and (19))
J opening displacement (Je elastic, Jp plastic parts)
k0 regularization width (interface thickness, Eq. (7))
lc fragment critical length at fragmentation saturation
n, m exponents (Eqs. (15), (16) and (19))
P dissipated energy (Pp plastic, Pd delamination, Pc cracking dissipations)
Qe interface stiffness (Qen ,Qet normal and shear terms)
T; T b interface tractions ( Tb n ; Tb t normal and shear terms)
U strain energy

Greek symbols
a isotropic damage variable
b mode mix function (Eq. (14))
c shear stress to normal fracture stress ratio
D indicates the change of a term
e strain
ec crack initiation strain
k_ plastic multiplier (Eq. (8))
m Poisson’s ratio of the interface
u, u* yield and plastic potentials
r
^f interfacial normal fracture stress
rcf coating fracture stress
s interfacial shear strength
X work of external loading

and solar cells that are manufactured on polymer substrates [3,4]. To be able to predict the reliability of the
structure under mechanical and thermal loads, a number of tests have been designed to measure the properties
of the coating and the interface, i.e., the strength and toughness of the coating and the adhesive strength and
toughness of the coating–substrate interface. Common tests are the Peel test [5,6], the scratch test and related
indention tests [7] and the tensile or fragmentation test [8,9]. The advantage of the last example as compared
with the first two is the exclusion of any third body interactions. The specimen is strained under uniaxial load
and, upon coating failure, the substrate is able to transfer the load across the crack. With increased strain fur-
ther cracking takes place and a progressively denser pattern of parallel cracks develop. At a high enough
strain, the crack density reaches a plateau value; the saturation level. The saturation level is often used as a
measure of the adhesion between the coating and the substrate, usually calculated using a shear lag approach
assuming a constant interfacial shear stress, IFSS and a fracture stress of the coating [10]. A measure of the
2616 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626

accuracy of the data reduction methodology can be obtained by comparing predictions of the whole fragmen-
tation process with experimental data. Approaches to modeling of the fragmentation process are usually based
on a stress analysis method in conjunction with a fracture criterion. The stress analysis can either be based on
linear-elastic theory [11–17] or shear-lag theories with the IFSS assumption [18,19]. Approaches to fracture
prediction can be divided into fracture mechanics theories, where failure is controlled by the toughness of
the coating [13,16,17,20], or stress based theories where a fracture stress is assigned to the coating. The coating
fracture stress has been treated as a constant property [17–19] or as dependent on fragment size using statis-
tical methods [12,15,21,22]. In the present case with an elastic coating on a polymer substrate, the assumption
of either elastic or elastic–ideal plastic behavior is a problem as the substrate generally strain hardens in the
strain range where fragmentation takes place. In the analytical formulation this can be handled to some extent
through factors calibrated against finite element (FE) calculations [17] or by totally relying on FE simulations
for the stress analysis [21].
Another difficulty in modeling the fragmentation process of a brittle coating on a ductile substrate, not
addressed in the above approaches, is the decohesion of the coating from the substrate. In this work a cohesive
zone model [23,24] is therefore used to explicitly include the progression of interfacial cracks during straining
in the FE simulation. The elastic–plastic behavior of the substrate under uniaxial tension is also incorporated
in the model. The analysis is based on a series of 2-D plane strain FE simulations in order to obtain strain
energy density, dissipation density and forces on the unit cell as function of strain and crack density. This data
is then used to calculate the crack density versus strain relation with the fracture toughness of the coating and
the interface as the two main controlling parameters aside from stiffness and geometric properties of the sub-
strate and coating. Prediction of the crack density versus strain evolution is compared with experiments and
simulation results are presented showing the effect of variations in interface and coating toughness as well as
coating thickness.

2. Fragmentation analysis approach

2.1. Basic strategy and assumptions

The fragmentation test is used to evaluate the cohesive strength of a brittle coating and the adhesion of the
coating to a ductile substrate. When strained under a uniaxial load, parallel cracks develop in the coating
spanning the width of the specimen. Upon further straining the number of cracks increases until a saturation
value is reached, see Fig. 1. For a specific coating, the saturation limit is determined by the mechanical prop-
erties of the substrate and the adhesion of the film to the substrate. The higher the adhesion and the stiffer the
substrate, the shorter the distance needed between existing cracks in order to load the coating enough to crack
further. Therefore, the increase in crack density, defined as the inverse of fragment size, is taken as a measure
of the strength or toughness of the coating while the saturation limit is viewed as a measure of the adhesion
between coating and substrate.
Prediction of crack density evolution with strain in the fragmentation test is based on a series of 2-D unit
cell analyses, using a series of simulations with constant crack density, see Fig. 2. In the current case with plas-
ticity in the substrate and growing delamination, this constitutes a strong assumption, as both phenomena are
inherently path dependent. Also inbuilt in this approach is the assumption that the progressive, discrete crack-
ing can be described by a continuously increasing crack density.
If one assumes that crack spacing is equal, it is possible to reduce the problem to a unit cell as depicted in
Fig. 3. In the analysis, symmetry conditions are applied to the vertical edges of the unit cell, i.e., all points on
the edges have the same displacements in the x-direction.

Fig. 1. Sketch of the fragmentation process with progressively denser crack pattern.
 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626 2617

Continuous
σ (ε, cd )

Simulations with different crack densities

Fig. 2. Schematic figure of stress–strain response during straining under continuous cracking and during straining with constant crack
density.

Crack spacing

Free crack
surface
y

Symmetry

Fig. 3. A unit cell of the fragmentation process.

2.2. Derivation of an expression for cohesive toughness

With the 2-D approach outlined above it is possible to obtain the values for force, strain energy density and
dissipation from plasticity and delamination as functions of strain and crack density. A second step in the pro-
cess of predicting the crack density versus strain is to calculate the energy dissipation due to cohesive coating
failure. This is accomplished by a calculation based on energy balance, which somewhat resembles the fracture
mechanics based approach taken to model the fragmentation process for elastic materials without delamination
[13,16]. By deriving an expression for the strain energy as a function of crack density (and strain or stress), the
energy release rate for an increasing crack density is calculated and compared to experimental values for frac-
ture toughness. In the present case where plasticity and delamination are included in the analysis, an analytical
solution would be very hard to obtain, so FE-simulations are used. These are described in a later section.
As a starting point in the derivation the mechanical energy balance is taken as
X¼U þP ð1Þ
where X is the work of external loading, U is the strain energy and P is the energy dissipated due to for in-
stance plasticity or cracking.
By separating the different dissipation contributions and specifying the external loading, Eq. (1) is rewritten
as
Z
X¼ F de ¼ U þ P ¼ U þ P p þ P d þ P c ð2Þ
e

where P with the sub-indices p, d and c stands for plastic, delamination and cracking dissipation and F is
the force applied on the unit cell (of unit length). As mentioned below, the discrete strain functions from
2618 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626

the simulations are converted into continuous functions for each crack density by interpolation. Knowing the
crack density as a function of strain cd(e), the crack density can be eliminated hence resulting in
U ¼ U ðeÞ; P p ¼ P p ðeÞ; P d ¼ P d ðeÞ ð3Þ
By using the definition of fracture toughness of the coating as the energy needed to form the crack surfaces,
the cohesive dissipation as a function of strain (valid for a length and width of unity) is obtained as
P c ¼ Gc hc cdðeÞ ð4Þ
where Gc is the fracture toughness and hc is the thickness of the coating.
Finally, using (2) and (4) the cohesive fracture toughness is solved for to give
Z e Z e 
F de ¼ DU þ DP p þ DP d þ Gc hc DcdðeÞ ) Gc ¼ F de  DU  DP p  DP d hc DcdðeÞ ð5Þ
e0 e0

where D indicates the change of a term, e.g., strain energy, from the strain e0 to e.

2.3. Calculation of crack density

With the values for the force, strain energy and dissipation known as functions of strain for discrete values
of crack-density, Eq. (5) is applied to predict the crack density versus strain. This can be viewed as an opti-
mization problem where the function to minimize is
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
f ðxÞ ¼ ðGc ðxÞ  Gc Þ  ðGc ðxÞ  Gc Þ ð6Þ
In Eq. (6) Gc(x) is the vector of cohesive toughness values calculated at the discrete strain points
x = e1, e2 ,. . . , en where a certain crack density exists. The vector Gc is constant with all elements taking the
value of the desired cohesive toughness. In the current work Matlab was used for the calculations. From
the simulations, the functions were known for unevenly spaced strain points and, using quadratic splines, were
interpolated to obtain an even spacing.
When using Eq. (5) to calculate a set of cohesive fracture values Gc(x) for a test set of strain points x it was
discovered that it was necessary to also interpolate the force data F(x) using quadratic splines in order to bet-
ter represent the force function in-between the simulation data points when the response was non-linear.
Investigations also attempted to interpolate between different crack density simulations in order to enhance
the number of data points. This proved not to present any advantage over the method using data points from
the simulation with a spline interpolation of the force.
As for the actual determination of the best set x, an iterative direct approach was applied. To allow for a
general form of the crack density curve, the discrete values of strain were retained rather than using a function
with fewer parameters, to represent the crack density evolution. These strain values were then determined by
looping through a range of values and selecting the lowest error as determined by (6). It should be noted here
that it was preferable to loop through several strain values as a change in one value slightly changed the spline
interpolation of the force and hence the Gc values calculated earlier.

3. Finite element modeling

In the analyses performed here, the coating is modeled as a linear-elastic solid while an isotropic hardening,
elastic–plastic constitutive law is used for the substrate. The hardening is controlled by stipulating the stress
versus generalized plastic strain obtained from a tensile test, see later for details [25]. A further feature of the
analyses is that the debonding of the coating from substrate is explicitly included by the use of cohesive zone
based interface elements.

3.1. Interface element

Delamination, or debonding, of the coating from the substrate is included in the analysis by interface
elements incorporating a cohesive zone model. With this method, the substrate and coating are meshed as
 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626 2619

(a) (b)
Tn
Film
Jn J
Jt
Substrate

J en J

Fig. 4. (a) Total, normal and transverse opening displacement. (b) Traction–opening displacement relation.

separate entities and then kept together by interface elements, similar to a normal contact interface (but con-
necting nodes with the same coordinates on both surfaces). Delamination is accomplished by introducing a
softening traction–displacement discontinuity relation for the interface, see Fig. 4. Hence, when the load
across the interface is too high, interfacial damage initiates and potentially develops up to fracture. At frac-
ture, the interface element loses all its stiffness and the coating and substrate surfaces act as two free, separate
surfaces.
The interface element formulation developed for mixed-mode crack growth in elastic structures is described
in [26] and only a summary is presented here for completeness. The special assumptions pertinent to the cur-
rent case of small dimensions and elastic–plastic materials are discussed in the material data section. The con-
stitutive formulation is based on an elastic–ideal plastic law coupled to isotropic damage in order to obtain the
form of the traction–opening displacement curve depicted in Fig. 4. The traction is obtained from the elastic
part of the opening displacement Je = J  Jp and the isotropic damage variable a as
 
b ¼ ð1  aÞ 1 Qe  Je with Qe ¼ G 1 þ 1 þ 2m N  N
T ¼ ð1  aÞ T ð7Þ
k0 2 1  2m

where N is the normal of the interface, G is the shear modulus and m is the Poisson’s ratio of the interface
material. Note that the interface stiffness is determined by material properties and by the so-called regulariza-
tion width k0, which basically corresponds to a thickness of the interface. The evolution of the plastic part of
the opening displacement and the damage variable is coupled by a plastic multiplier k_ as

J_ p ¼ kou b
_  =o T; a_ ¼ Bk_ ð8Þ

where u* is the plastic potential to be defined later and B is a function to be determined from and calibrated
against fracture toughness data. Within a strain increment, the plastic multiplier k_ is determined from the con-
dition that the stress should stay on the yield surface at plastic loading
k_ P 0; ^ 6 0;
uðTÞ _ TÞ
kuð ^ ¼0 ð9Þ
When modeling crack- or delamination growth in elastic structures with cohesive zone models, it is generally
observed that the results are little influenced by the chosen value of the maximum stress in the traction–open-
ing displacement relation, as long as this is kept within reasonable limits. Moreover, if one considers that no
stable crack growth occurs, for example in a tensile test where the whole interface stress state is homogenous,
and that enough energy is available to fracture the interface; the maximum interfacial traction corresponds to
the fracture stress. Therefore, the yield potential is taken as a function of the interfacial fracture stresses as
!2 !2 !
Tb t h Tb n i Tb n
u¼r ^f þr
^f ð1  cÞ þr
^f c r
^f
c^
rf r
^f r
^f
b Tb t :¼ j T
with Tb n :¼ N  T; b t j and T
b t :¼ T
b  Tb n N ð10Þ
where r ^f is the interfacial normal fracture stress, c is the ratio shear to normal fracture stress and c P 0 con-
trols the behavior under compressive normal stress. In (10), and throughout the paper, hxi denotes the positive
part of x, i.e. (x + jxj)/2. The plastic potential is chosen similar to the yield function as
2620 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626

!2 !2
 Qe
2 Tb t h Tb n i
^f c ne
u ¼r þr
^f r
^f ð11Þ
Qt c^
rf r
^f

where Qen and Qet are the normal and shear stiffness terms of Qe. From (7) it follows that these components are
related to the shear modulus as
Qen 1
ð1  2mÞQen ¼ G; 2Qet ¼ G ) ¼ ð1  2mÞ ð12Þ
Qet 2
As mentioned above, the maximum traction is assumed to not substantially affect the results for delamina-
tion growth in elastic structures, instead the growth is controlled by the energy dissipated in the interface.
Hence, in the present work, the fracture toughness of the interface is used to calibrate the interfacial model
and more specifically the function Bð TÞ. b In general, fracture toughness varies with crack mode I/II ratio,
so the calibration is done for general mixed mode loadings. Assuming a constant mode I/II ratio (constant
Tb n = Tb t ) during loading, the fracture energies for the mode I and mode II components are obtained from
(8)–(11) as
Z 1
2 h Tb n i 1 h Tb n i
2 2
i
GI ¼ ð1  aÞ da ¼
0 B r ^f B r ^f
Z 1 e b 2
ð13Þ
1 Qen ð Tb t Þ
2
i 2 Qn ð T t Þ
GII ¼ ð1  aÞ da ¼
0 B Qet r ^f B Qet r ^f

where it was used that Tb n and Tb t stay at the same point on the yield surface for fixed mode loading. If one
describes the experimentally determined fracture toughness by a function exp Gic ðbÞ, where b ¼ GiII =ðGiI þ GiII Þ
describes the mode mix, the damage evolution controlling function Bð TÞ b can be determined as
D E 
ð Tb t Þ Qe11
2
1 2 Qe Gi
B ¼ exp i Tb n þ ð Tb t Þ2 ne with b ¼ i II i ¼ ð14Þ
Gc ðbÞ^
rf Qt GI þ GII h Tb n i2 Qe þ ð Tb t Þ2 Qe
22 11

In the implementation the experimental fracture toughness data exp

Gic ðbÞ is input in terms of a polynomial
expression in b.

3.2. Material data

Generally, the fracture toughness data and fracture stress of material interfaces are known from experi-
ments and can be used to calibrate the model as outlined above. However, in case of ultra-thin coatings on
elastic–plastic substrates, no experimental method is able to directly provide such data. Rather, the present
interface properties were back-calculated from simulations of the fragmentation test [27]. As only one exper-
iment was available, it was assumed in the simulations that both the fracture toughness and the failure stress
for the interface are independent of the mode ratio I/II. In other words, the experimental function for the
adhesive fracture toughness is a constant, i.e., exp Gic ðbÞ  exp Gic . Likewise, the choice of equality of the fracture
stress in tension and shear was made, which means that the factor c in Eqs. (10) and (11) takes the value 1.
Practically, the value of the fracture stress was chosen so that the cohesive zone size is in the range of 5–8 ele-
ment lengths. This choice guarantees a numerically stable behavior of the simulations and at the same time a
physically small cohesive zone. It could be argued that for the scale of the current computations, the size of a
physical damage zone could be measured or defined and the cohesive zone size adjusted to suit. However, no
attempts were made to do this in this work. When a cohesive zone model is used in conjunction with an elas-
tic–plastic material, the dissipation in the plastic zone around the crack tip contributes to the apparent tough-
ness in the interface. As the choice of interface fracture stress together with the choice of finite element
discretization might influence the stress state around the crack and hence also the size of the plastic zone,
the calibrated interface toughness, fracture stress and element size should be viewed as one set of properties
describing the interface. In the current case, the cohesive zone is small (25–40 nm) compared with both the
delamination length and the plastic zone size (both several tenths of lm) and it could therefore be expected
 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626 2621

350
Mean value fit of Exp.
Simulation

300

250
Cauchy stress [MPa]

200

150

100

50

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Logarithmic (ln) strain

Fig. 5. Stress–strain behaviour of the substrate under uniaxial tension.

that the results are relatively insensitive to the choice of fracture stress when chosen so as to keep the cohesive
zone size at a certain size. Limited trials in [27] supports this but until extensively investigated the toughness,
stress and element size should be viewed as one cohesive zone property set and not necessarily describe the
correct physical interface toughness and strength. The material properties of the interface were chosen close
to the substrate properties with a Young’s modulus 3.0 GPa and Poisson’s ratio of 0.4. The interface thickness
k0 was arbitrarily selected as 1 nm.
The substrate was modeled as an elastic–plastic material with isotropic hardening where the hardening
behavior is described in the input to Abaqus through a table of Cauchy stress versus generalized logarithmic
plastic strains. This data was obtained from a tensile test of the substrate material, see Fig. 5. The Poisson’s
ratio of the substrate was chosen as 0.4.
Linear elastic behavior was used for the vapor deposited ceramic film with a Young’s modulus 150 GPa and
Poisson’s ratio 0.3. The thickness of the film was taken to be 350 nm in all the examples, 175, 87.5 nm were
additionally used for an investigation into the effect of film thickness.

3.3. Finite element model

The interface element described above was implemented into the Abaqus standard code as a user element
connecting four node, plane strain 2-D elements with full integration (CPE4) [25]. To ensure a good conver-
gence and a stable behavior, a fine discretization with an element length of 5 nm was used in the vicinity of the
coating–substrate interface, see Fig. 6 for an illustration.
As described in the basic strategy and assumptions sub-section, the analysis is based on several simulations
with different, fixed crack densities. With this approach, a unit cell of a size half the fragment length is used, see
Fig. 6. A substrate thickness of 25 lm was included in the analysis and the bottom, horizontal edge was con-
strained in the y-direction. Due to symmetry considerations, the whole vertical edge between the cracks should
remain straight and a controlled displacement U of this was used to load the structure. In addition, the edge
below the coating crack should remain straight and was constrained in the x-direction, note however that the
actual crack coating surface remained free.
2622 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626

Fig. 6. (a) Boundary conditions and (b) meshing for a unit cell with fragment length of 5 lm (cd = 200 mm1).

4. Results and discussion

4.1. Comparison with experimental data

In an earlier paper [27], the cohesive fracture toughness of the coating, Gcc and the adhesive toughness of the
interface, Gic were determined from experimental fragmentation data. To validate the currently proposed cal-
culation approach, the crack density versus strain curve was calculated with the material properties from [27]
and compared with the experimental input data used in the same paper. The properties used were,
Gcc ¼ 43 J=m2 , Gic ¼ 3 J=m2 , rf = 300 MPa and the results are shown in Fig. 7 together with the experimental
data and the fit to experimental data used to determine the properties. It should be noted that only experimen-
tal data up to 4% strain was used to evaluate the properties, but clearly the simulation compares favorably
with the experimental data also above this strain value.

4.2. Effect of different coating and interface fracture toughness

To observe the effects of different fracture toughness of the coating on the fragmentation behavior, simu-
lations were run with a coating fracture toughness of 40, 45 and 50 J/m2, using an interface fracture toughness
of 3 J/m2 and a fracture stress of 300 MPa. See results in Fig. 8. Similarly, simulations were run with the inter-
face fracture toughness taken to be 1.5, 3, 6 J/m2 (fracture stress 230, 300, 300 MPa) with a coating toughness

200

150
Crack density [mm-1]

100

50 Simulation
Fit to experimental data
Experimental data
0
0 0.02 0.04 0.06 0.08 0.1 0.12
Strain

Fig. 7. Back calculated crack density versus strain using material properties determined from fit to experimental data.
 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626 2623

200

150

Crack density [mm-1]

100

50
Gc=50 [J/m 2]
Gc=45 [J/m 2]
Gc=40 [J/m 2]
0
0 0.02 0.04 0.06 0.08 0.1 0.12
Strain

Fig. 8. Crack density as a function of strain for different coating toughness and an interface toughness of 3 J/m2.

200

150
Crack density [mm-1]

100

50
Gi=6 [J/m 2]
Gi=3 [J/m 2]
Gi=1.5 [J/m 2]
0
0 0.02 0.04 0.06 0.08 0.1 0.12
Strain

Fig. 9. Crack density as a function of strain for different interface toughness and a coating toughness of 43 J/m2.

of 43 J/m2. The results are shown in Fig. 9. Interestingly, increasing the coating toughness and decreasing the
interface toughness seem to have a very similar effect. A tendency to a more gradual approach to the satura-
tion level is evident as the interface toughness decreases, nevertheless the resolution of the results does not per-
mit a definite judgment as to whether the saturation density is slightly different for the different interface
toughness results or not. However, in [27] it was shown that for a certain crack density, the strain at which
the maximum strain energy in the coating is reached is dependent on the interface toughness. The lower
the toughness, the more delamination growth takes place and the lower the strain is where the maximum
in strain energy is reached. Crack density increases slower with strain the lower the interface toughness
is, see Fig. 9, therefore, for a low enough toughness the saturation limit must decrease. Furthermore, the
small influence of fracture toughness on saturation level in the current results also indicate that with an exper-
imental scatter as in Fig. 7, the saturation level is not a very clear indication of the interface toughness in the
system.

4.3. Effect of different coating thickness

Simulations were performed to determine the effect of different coating thickness on fragmentation behav-
ior. The coating and interface properties were held constant at Gcc ¼ 45 J=m2 , Gic ¼ 3 J=m2 , rcf ¼ 300 MPa for
all the simulations. As expected, with a thinner coating, the cracking commences later, the crack density
increases more steeply with strain and a higher crack density saturation level is attained, see Fig. 10.
2624 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626

400
350 [nm]
350 175 [nm]
87.5 [nm]

Crack density[mm-1]
300

250

200

150

100

50

0
0 0.02 0.04 0.06 0.08 0.1 0.12
Strain

Fig. 10. Crack density evolution with strain for different film thicknesses.

Several models exist in the literature that predicts the crack initiation strain and its dependence on the
thickness of the coating [1]. The expressions are generally of the form
ec  f ðmaterial propertiesÞ  hc1=n ð15Þ
that describe the predicted dependence of coating thickness. Derived values of n were reported in [1] to be
equal to 2, 3 or 4. In order to see how the simulations performed in this work compare, crack initiation strain
was linearly extrapolated to zero crack density from the first two points on the fragmentation curve, see
Fig. 10. Of course, this can only be viewed as an estimate of the real crack initiation strain, as the assumptions
of a fragmentation unit cell and that all cracks instantaneously span the whole thickness might not be gener-
ally appropriate at crack initiation. Also, if the cracking is unstable at the beginning of the fragmentation pro-
cess, i.e., the crack density jumps from zero to a finite number without straining, then the interpolated value
could be too low. An example of this can be seen in the experimental data in Fig. 7 where crack densities below
10–20 were not recorded.
In order to compare the extrapolated crack initiation values to the predictions mentioned earlier, a function
of the form
 0 1=n
h
ec ¼ b c ð16Þ
hc
was fitted to the data, see Fig. 11. Interestingly, the fitting exponent 1/n was determined to 0.528 (b = 358 with
h0c ¼ 87:5) which corresponds very well with n = 2.
A widely used formula for analyzing the fragmentation data in terms of interfacial shear stress (IFSS), rep-
resenting a perfectly plastic interface, is [28]

0.012
Simulation
0.01 Fit
Crack initiation strain

0.008

0.006

0.004

0.002

0
50 100 150 200 250 300 350 400
Film thickness [nm]

Fig. 11. Crack initiation strain as a function of film thickness together with fitted a function.
 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626 2625

400

Saturated crack density [mm-1]

Simulation
Fit
300

200

100

0
50 100 150 200 250 300 350 400
Film thickness [nm]

Fig. 12. Crack density at saturation for different film thicknesses together with a fitted function.

s ¼ 2hc rcf =lc ð17Þ

where s is the IFSS, rcf
is the fracture stress of the coating and lc is a length close to the fragment length at
saturation. If one assumes that the coating fracture stress and the IFSS remain constant, the dependence of
the saturation limit on the coating thickness is
cd  1=lc ¼ s=2hc rcf ð18Þ
i.e., a linear dependence of coating thickness on crack density saturation level is predicted.
Similarly as before, a function of the form
 0 1=m
h
cdsat ¼ a c ð19Þ
hc
was fitted to the simulation data to evaluate dependence of coating thickness on saturation level, see Fig. 12.
In this case the exponent m is close to 2 or more exactly 2.08 (a = 0.01 with h0c ¼ 87:5). This result, when com-
pared to the corresponding exponent of 1 from Eq. (18), indicates that for the IFSS to remain constant for
different film thicknesses, the coating fracture stress must vary with film thickness as rcf / hc1=2 . Thus, the
stress at which the coating cracks should not be viewed as the fracture stress (by definition size independent);
instead it should be allowed to vary with thickness also if defect controlling size effects are not considered.

5. Conclusions

A finite element based method was developed to predict crack density as a function of strain in the frag-
mentation process of a thin brittle coating on a flexible substrate. The method explicitly includes plasticity
effect in the substrate as well as debonding of the coating from the substrate. Necessary mechanical properties
include the stress–strain behavior of the substrate, the stiffness of the linear elastic coating and the fracture
toughness of the coating and the coating substrate interface.
A comparison with experimental results clearly shows the ability of the present method to predict the whole
fragmentation curve, including the rate of initial cracking and the saturation limit. Numerical experiments
with varying fracture toughness of the coating and interface revealed a similar influence of increased coating
fracture toughness and decreased interface fracture toughness on the evolution of cracking. Surprisingly, no
strong influence of the interface toughness on the saturation limit was found which contradict the common
perception that the saturation level is controlled by the interfacial adhesion level. Also simulations with dif-
ferent coating thickness were performed and, as expected, both the crack initiation strain and the saturation
crack density increased with decreased coating thickness. A fit to the numerical data revealed a close to square-
root dependence of the thickness on both initiation strain and saturation limit. This is in accordance with the-
ories for the initiation strain but is in contradiction to the linear dependence on thickness predicted by shear
lag models that assume the existence of a constant interfacial shear stress (IFSS) and a constant fracture stress
for the coating. Thus, the coating cracking stress used in the IFSS evaluation must be taken as thickness
dependent if defect-controlling size effects are not taken into account.
2626 N.E. Jansson et al. / Engineering Fracture Mechanics 73 (2006) 2614–2626

Acknowledgements

The authors are indebted to the IST program (IST-2001-34215 FLEXled) and the Swiss Federal Office for
Education and Science (OFES) for funding this work. They acknowledge the companies Ferrania Imaging
Technologies and Unaxis Balzers Ltd. (Displays Division) for the supply of the films samples.

References

[1] Leterrier Y. Durability of nanosized oxygen-barrier coatings on polymers. Progr Mater Sci 2003;48(1):1–55.
[2] Chatham H. Oxygen diffusion barrier properties of transparent oxide coatings on polymeric substrates. Surf Coat Technol 1996;
78(1–3):1–9.
[3] Leterrier Y, Medico L, Demarco F, et al. Mechanical integrity of transparent conductive oxide films for flexible polymer-based
displays. Thin Solid Films 2004;460(1–2):156–66.
[4] Lewis JS, Weaver MS. Thin-film permeation-barrier technology for flexible organic light-emitting devices. IEEE J Select Topics
Quantum Electron 2004;10(1):45–57.
[5] BenAmor S, Baud G, Benmalek M, et al. Titania coatings on polyethylene terephthalate: adhesion and XPS studies. J Adhesion
1998;65(1–4):307–29.
[6] BenAmor S, Baud G, Besse JP, et al. Elaboration and characterization of titania coatings. Thin Solid Films 1997;293(1–2):163–9.
[7] Benjamin P, Weaver C. Measurement of adhesion of thin films. Proc Roy Soc Lond A 1960;254(1277):163–76.
[8] Chow TS, Liu CA, Penwell RC. Direct determination of interfacial energy between brittle and polymeric films. J Polymer Sci B
1976;14(7):1305–10.
[9] Leterrier Y, Andersons J, Pitton Y, et al. Adhesion of silicon oxide layers on poly(ethylene terephthalate). 2. Effect of coating
thickness on adhesive and cohesive strengths. J Polymer Sci B 1997;35(9):1463–72.
[10] Kelly A, Tyson WR. Tensile properties of fibre-reinforced metals – copper/tungsten and copper/molybdenum. J Mech Phys Solids
1965;13(6):329–50.
[11] Handge UA, Sokolov IM, Blumen A, et al. Patterns of fragmentation for polymer coatings. J Macromol Sci-Phys B 1999;38(5–6):
971–80.
[12] Handge UA, Leterrier Y, Månson JAE, et al. An analysis of disorder in thin silicon oxide coatings. Europhys Lett 1999;48(3):280–5.
[13] Kim SR, Nairn JA. Fracture mechanics analysis of coating/substrate systems. Part I: analysis of tensile and bending experiments.
Engng Fracture Mech 2000;65(5):573–93.
[14] Lacroix T, Keunings R, Desaeger M, et al. A new data reduction scheme for the fragmentation testing of polymer composites. J
Mater Sci 1995;30(3):683–92.
[15] Mezin A, Sajid B. Undirectional multiple cracking of coatings described through random sequential placement of rods on a line. Thin
Solid Films 2000;358(1–2):46–55.
[16] Nairn JA, Kim SR. A fracture-mechanics analysis of multiple cracking in coatings. Engng Fracture Mech 1992;42(1):195–208.
[17] Hsueh CH, Yanaka M. Multiple film cracking in film/substrate systems with residual stresses and unidirectional loading. J Mater Sci
2003;38(8):1809–17.
[18] McGuigan AP, Briggs GAD, Burlakov V, et al. An elastic–plastic shear lag model for fracture of layered coatings. Thin Solid Films
2003;424(2):219–23.
[19] Yanaka M, Kato Y, Tsukahara Y, et al. Effects of temperature on the multiple cracking progress of sub-micron thick glass films
deposited on a polymer substrate. Thin Solid Films 1999;356:337–42.
[20] Kim SR, Nairn JA. Fracture mechanics analysis of coating/substrate systems. Part II: experiments in bending. Engng Fracture Mech
2000;65(5):595–607.
[21] Yanaka M, Tsukahara Y, Okabe T, et al. Statistical analysis of multiple cracking phenomenon of a SiOx thin film on a polymer
substrate. J Appl Phys 2001;90(2):713–9.
[22] Ogwu AA, Colleran P, Maguire P, et al. Fitting probability distribution functions to saturation crack spacing distribution in
substrate plastic, straining test for thin films and coatings. Surf Engng 2002;18(4):277–82.
[23] Tvergaard V. Effect of fiber debonding in a whisker-reinforced metal. Mater Sci Engng A 1990;125(2):203–13.
[24] Dugdale DS. Yielding of steel sheets containing slits. J Mech Phys Solids 1960;8(2):100–4.
[25] Abaqus/Standard user’s manual version 6.3. Hibbit, Karlsson & Sorensen, Inc.; 2002.
[26] Jansson NE, Larsson R. A damage model for simulation of mixed-mode delamination growth. Comp Struct 2001;53(4):409–17.
[27] Jansson NE, Leterrier Y, Medico L, et al. Modelling of cracking and debonding of a thin brittle coating on a polymeric substrate.
Thin Solid Films, submitted for publication.
[28] Hu MS, Evans AG. The cracking and decohesion of thin-films on ductile substrates. Acta Metall 1989;37(3):917–25.