You are on page 1of 14

Journal of Micromechanics and Microengineering

PAPER

Identification of traction-separation parameters by means of peel testing


and in situ confocal microscopy
To cite this article: Emanuele Cattarinuzzi et al 2019 J. Micromech. Microeng. 29 034001

View the article online for updates and enhancements.

This content was downloaded from IP address 109.171.137.229 on 23/05/2020 at 12:14


Journal of Micromechanics and Microengineering

J. Micromech. Microeng. 29 (2019) 034001 (13pp) https://doi.org/10.1088/1361-6439/aafa2f

Identification of traction-separation
parameters by means of peel testing
and in situ confocal microscopy
Emanuele Cattarinuzzi1,2 , Dario Gastaldi1, Pasquale Vena1,
Fiorella Pozzobon3, Lucrezia Guarino2, Luca Cecchetto2
and Antonio Andreini2
1
  Chemistry, Materials and Chemical Engineering Department, Politecnico di Milano, Milano, Italy
2
  STMicroelectronics srl, Agrate Brianza, Italy
3
  STMicroelectronics srl, Cornaredo, Italy

E-mail: emanuele.cattarinuzzi@st.com

Received 30 October 2018, revised 6 December 2018


Accepted for publication 20 December 2018
Published 21 January 2019

Abstract
Delamination owed to environmental stress is a relevant concern in the field of
microelectronics packaging. This study is focused on the adhesion of a photosensitive
insulator to inorganic passivation layers and metal caps. Peel testing has been combined
with confocal laser scanning microscopy to image the bent shape of the peeled layer in situ.
Elastic plastic beam theory and finite element analysis have been used to identify a cohesive
traction-separation law matching both the work of separation and the peel arm profile. The
results indicate that the actual adhesion energy amounts to ~30% of the external work,
whereas the rest is dissipated by plasticity. Being the latter related to the curvature of the peel
arm during delamination, these results confirm the relevance of in situ imaging in the field of
microelectronics reliability.

Keywords: microelectronics packaging, adhesion, in situ testing, ICPeel, cohesive modelling


(Some figures may appear in colour only in the online journal)

1. Introduction The set of mechanical properties required for a FEA


depends on the failure mechanism of interest. One of the most
1.1.  Motivations and aims of the study popular reliability issue in the field of package reliability is
interface delamination [2], which can be modeled in the finite
Microelectronics packaging shields semiconductor devices
element (FE) framework by means of cohesive traction-sep-
from undesired environmental conditions, such as moisture
aration laws (TSL) [3]. In practice, this means introducing a
and temperature fluctuations. Besides the outer environment, compliant constraint at the boundary between the package and
device operation itself causes thermal cycling owed to Joule the chip. To instance the elastic behavior and the damage evo­
heating. Given these drivers of detriment, a reliable package lution of the TSL, appropriate experimental characterization
must withstand associated swelling and thermal stresses [1]. of the interfaces of interest is due.
Reliability of polymeric packages is routinely screened by According to the scientific literature and industrial stand-
standard testing for package qualification [2], which is a time ards, measuring the adhesion of microelectronics materials
intensive activity. At an earlier stage, finite element analysis can be addressed by peel testing [4–6]. The peel test is an
(FEA) is a relevant resource to predict the impact of design experimental method to measure the interface toughness of
parameters on the mechanical reliability and optimize the flexible laminates. The test consists in pulling one layer (i.e.
device before submitting it to qualification tests. the peel arm) to indirectly stress the interface, which at an

1361-6439/19/034001+13$33.00 1 © 2019 IOP Publishing Ltd  Printed in the UK


J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

Figure 1.  (a) and (b) Sketch of the sub-problem addressed by elastic plastic beam theory. Colored backgrounds in the moment/curvature
relation of the peel arm indicate different energetic contributes computed by the ICPeel algorithm. (c) Sketch of the peel test. Letters
indicate the regions which experience bending as depicted in the moment/curvature relation (a) and (b). The process zone and the root
radius are highlighted. (d) Sketch of the sub-problem addressed by the theory of beams on elastic foundation. (e) Bi-parametric traction-
separation law used to describe the constitutive behaviour of the interface. It is worth stressing that, for bi-parametric TSLs, the opening at
damage initiation (critical opening, δc ) corresponds to the opening at complete debonding (failure opening, δf ). © 2018 IEEE. Reprinted,
with permission, from [15].

early stage acts as a constraint. Once the external work over- adhesives [8–10]: among others, the ICPeel method devel-
comes a critical value, the constraint fails, i.e. delamination is oped by Kinloch and Williams [9] has emerged and it was
initiated. Entering the steady state delamination regime, the accepted by the European Structural Integrity Society (ESIS)
force signal features a plateau. This means that new free sur- as a standard for measuring the fracture toughness of bonded
face is created at a constant rate, corresponding to the clamp laminates [11]. In continue, few basic notions on the ICPeel
displacement rate. The schematic picture is that of a process method are recalled, in the interest of later comparison with
zone of fixed dimensions, which travels along the sample as experimental results (section 2.3) and numerical analyses
the test proceeds. (section 3.2). The energy balance of a peel test can be approxi-
In the simplest model of the peel test, the work of separa- mated by the following equation
tion (WOS) provided by external forces is an estimate of the
Fpeel
fracture toughness with a given mode mixity [7]. The latter WOS =
(1) (1 − cos θ) ∼
= Gi + Gd
b
depends on the relative orientation between the substrate and
the pulling direction, i.e. the peel angle. In the view of cohe- where Fpeel is the force measured during steady state peel, b
sive modeling, it is natural to begin focusing on pure fracture is the width of the sample, θ is the peel angle, Gi is the adhe-
modes: for mode I fracture to be dominant, the peel angle sion energy and Gd represents the plastic dissipation due to
must approach 90°, implying significant bending of the peel peel arm bending. The full analytical treatment (for which
arm at the delamination front. If the peel arm material features the reader is referred to [9, 10]) accounts for further terms
an elastic plastic behavior (e.g. polymers for non-hermetic (e.g. elastic stretching of the peel arm), whereas equation (1)
packaging), the potential risk to overestimate the adhesion is focused on the competition between adhesion energy and
energy is severe: WOS expresses the external energy supplied plastic dissipation, which typically represent the most relevant
to sustain all dissipation processes involved in the peel experi- terms of the balance.
ment, including plastic dissipation of the peel arm owed to To estimate these terms, the peel framework is split into
bending [8]. two separate problems. On the left hand side of figure 1, the
With these premises, this paper focuses on the adhesion of free standing peel arm is modelled as an elastic plastic beam
a polymer for microelectronics packaging to passivation and experiencing bending. The mechanical response is best
metallic interconnects. Specifically, it describes a procedure expressed by the moment/curvature relation (figure 1(a)),
to combine experimental, analytical and numerical methods to which relies on the stress/strain behaviour and the cross sec-
tion dimensions of the peel arm. On the right hand side of
(i) discriminate adhesion energy and plastic dissipation from
the sketch, the peel arm is modelled as a beam on elastic
the total WOS measured in a 90° peel experiment;
foundation (figure 1(d)). La latter represents the interface
(ii) identify the parameters of a cohesive TSL matching
and acts as a deformable constraint with limited energy
experimental observations.
storing capabilities. The constitutive relation is a TSL with
no softening branch (figure 1(e)), identified by an arbitrary
choice of two parameters among toughness (Gi ), stiffness
1.2.  Plastically deforming peel arm and ICPeel method
(Ki ), strength (tc) and failure opening (δf ). This class of
The problem of the peel test with plastically deforming peel TSLs featuring brittle damage can be identified by the fol-
arm has been thoroughly addressed in the field of structural lowing constraint

2
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

Ge tc2
(2)
= =1
Gi 2Ki Gi
where Ge = tc2 /2Ki expresses the recoverable (elastic) part
of the adhesion energy. The ratio Ge /Gi will be from now
on referred to as the aspect ratio of the TSL, ranging from
0 to 1: the lower bound represents an interface which experi-
ences damage initiation at arbitrarily small openings while,
as anticipated, the higher bound represents a brittle interface,
for which damage initiation and complete damage coincide,
leading to abrupt failure.
The moment/curvature relation describes the bending his-
tory experienced by a single cross section, from the unloaded,
bonded state (figure 1(c), point U) to the complete straighten
condition, after debonding (figure 1(c), point C). Adjacent
cross sections  along the peel arm profile (figure 1(c)) cor-
respond to adjacent bending states on the moment/curvature
relations (figure 1(a)). As the delamination proceeds, each
cross section  replaces its neighbour and steps further along
the bending/curvature relation. The plastic dissipation collec-
tively experienced by the cross sections of the peel arm is then
equal to the hysteresis loop of a single cross section  under-
going a full bending/unbending cycle. Hence, the hysteresis of
the moment/curvature diagram provides a measurement of Gd.
Area [UABC] = Gd b
(3)
The complementary energy (Ψ) of the moment/curvature
relation also provides a relevant measurement: using static
equilibrium and large displacement kinematics of beams, Figure 2.  Representative set of ICPeel results for a given value of
the infinitesimal increment of complementary energy can be WOS. The top panel sketches different deformed shapes of the peel
expressed as [9] arm and corresponding process zone lengths. The plot shows ICPeel
estimates in terms of TSL parameters: the green region corresponds
1 to compliant interfaces (Ki ↓), for which the root curvature is so
dΨ = dM = −Fpeel sin (θ − φ) dφ (4)
R smooth that no plastic dissipation occurs, so that Gi equals the

measured WOS. The blue region shows that, for progressively
where φ denotes the local slope of the beam. The comple- stiffer interfaces (Ki ↑), the root radius decreases, implying plastic
mentary energy can be obtained by integrating equation  (4) dissipation during bending (Gd ↑) and drift of Gi from the total
WOS. In the orange region, the root curvature is so sharp ( R0 ↓↓)
over the full range of bending moments experienced by the that the peel arm undergoes plastic dissipation even upon reverse
peel arm. bending, i.e. when straightened along the clamp displacement
´M direction. For case studies falling in this region, plastic dissipation
Ψ = Area [UFABC] = McA R1 dM may easily be the dominant term contributing to the total WOS
(5)
´0 (Gd ↑↑).
= −Fpeel sin (θ − φ) dφ
θ
= Fpeel (1 − cos θ)
∆ = R 0 θ0
(7)
Equation (5) shows that the complementary energy per unit
width matches exactly the definition of WOS. The adhesion where θ0 is the root rotation. For a given WOS value, infi-
energy can be measured as the difference between the com- nite (Gi , Ki ) solution sets exist, corresponding to infinite
plementary energy and the hysteresis loop of the moment/ ways to split the WOS into adhesion and plastic dissipation.
curvature diagram. The only difference between all feasible TSLs is the process
zone length, which results in a different curvature radius at
Area [UFABC] − Area[UABC] the root. Intuitively, stiffer interfaces feature a shorter process
(6) Gi =
b zone, which results in a smaller root radius and implies higher
Measuring the aforementioned integrals requires the esti- plastic dissipation due to peel arm bending (figure 2).
mation of the maximum curvature experienced along the peel Noteworthy, R0 can be measured experimentally by com-
arm, which by definition sits at the root (i.e. the last adhered bining in situ imaging to the testing apparatus. In the field of
point of the peel arm, point A in figure 1) [12]. This is indeed bonded laminates, the interface is a tangible entity and cor-
the connection between the two problems (figure 1(c)), since responds to an adhesive layer with known Young modulus
the root radius, R0 (i.e. a feature of the peel arm) is related to ( Ea ) and measurable post-curing thickness (ha ). This pro-
the length of the process zone, ∆ (i.e. a feature of the inter- vides an estimate of the TSL elastic stiffness (Ki = Ea /ha),
face) [10]. hence the remaining TSL parameter (Gi ) can be computed
3
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

with no ambiguity by the ICPeel method, provided a measure- higher elastic stiffness, Ki ), resulting in a shorter process
ment of the steady state peel force. For this reason, previous zone but higher stresses experienced by the adherend
attempts to match ICPeel results and in situ measurements of above (figures 3(c)–(e)). The latter ultimately experiences
the root radius were mainly aimed at testing the robustness of higher plastic dissipation, which is reflected by a higher
the analytical method [12]. In the field of microelectronics, WOS (figure 3(b));
delamination does not refer to the cohesive failure of an adhe- (ii) when the root radius becomes small enough (i.e.  ⩽4 times
sive layer along the stack, while to the adhesive failure at the the thickness of the peel arm [8]), adherend plasticity in
boundary between two adjacent layers. Whether the interface the vicinity of the crack tip becomes significant and FEA
corresponds to a sharp monolayer of chemical bonds or a very results drift from ICPeel estimates even for brittle TSLs
thin adhesion promoter with measurable thickness [13], such (figures 3(f)–(g)).
material does not exist in bulk form, hence its stiffness cannot
FEA itself can be used to match experimental results by
be characterized in advance. Indeed, the traditional ICPeel
testing different TSLs but, even in a 2D plane strain frame-
approach would model an intangible interface as an adhesive
work, a FE mesh designed to capture through-thickness
layer with vanishing thickness [10], i.e. with an infinite elastic
gradients of strain in a thin adherend requires hours of com-
stiffness (ha → 0 ⇒ Ki = Ea /ha → ∞). This limit case is
putation. Considering the need of an extra axis (e.g. tc) to
not of practical interest in the view of numerical modelling,
account for the aforementioned sensitivity to the TSL aspect
since equation (2) would imply an infinitely high critical trac-
ratio, the full sampling of a (Gi , Ki , tc) parameter space would
tion (tc → ∞) and prevent to observe delamination in a FEA.
not be feasible in practice. In comparison to FEA, the ICPeel
Cohesive zone modelling is a lumped parameter approach
algorithm is dramatically cost effective, with a computation
to describe interface mechanics as a compliant, degradable
time to result in the range of few seconds [10].
constraint. TSL parameters do not necessarily require a cor-
To exploit this advantage, section 3.2 describes a TSL iden-
respondence with physical interface features, hence the elastic
tification procedure where the parameter space (Gi , Ki , tc) is
stiffness Ki can be selected among all eligible ICPeel esti-
sampled by FEA using ICPeel estimates as initial guesses.
mates to best match experimental results.
This work aims at using the in situ measurement of the
root radius as an additional input to identify the best matching 2.  Experimental section
(Gi , Ki ) pair among a set of ICPeel results. In this view, sec-
tion 2.2 presents the use confocal microscopy as a promising 2.1. Materials
in situ imaging technique enabling straightforward image
The coating of interest is an organic photosensitive insulator
processing for root radius estimation. Section  3.1 illustrates
(PI) for microelectronics packaging. Two different substrate
how to match experimental measurements of WOS and R0 to
materials were considered: (i) an inorganic passivation (Pass)
ICPeel estimates, in practice.
and (ii) a metallic cap (Met). All details about chemical com-
position, thickness and fabrication recipes of materials are
1.3.  Influence of TSL aspect ratio and limitations of ICPeel confidential information. The target thickness of the PI layer
in comparison to FEA makes exception since it is a relevant reference to comment
on the plastic dissipation regime owed to peel arm bending
The aforementioned arguments suggest that the ICPeel method (section 3.1).
can be used to instance a cohesive TSL for FEA, with the Rectangular strips (~1  ×  15 mm2) of both substrates were
constraint that damage evolution must be brittle (i.e. without implemented on the same wafer by appropriate layout design.
softening branch). According to [10], this is not necessarily a A homogenous layer of PI was spin coated with a target
limitation: in the ICPeel framework, varying the aspect ratio thickness of 10 µm. Photolithography was used to define the
of a TSL with given toughness and strength does not affect PI coating into rectangular strips aligned with the layout of
the energy balance, since the adherend upon the process zone Pass and Met strips (figure 4). Samples were singled out by
is modelled as an elastic beam on compliant foundation. Its diamond saw cutting: the tip path was set to draw a frame
deformability in the thickness direction is lumped in a linear surrounding the PI strips with a width of ~20 µm, so as to
elastic contribute in series with the cohesive interface [10]. preserve the sharpness of the PI edges achieved after develop-
Neither through-thickness gradient of strain nor associ- ment. As discussed in section  2.2, this choice has a further
ated plasticity are accounted for [14], i.e. the elastic plastic advantage when in situ imaging of the root is pursued by
behaviour used to model the bent peel arm is neglected when means of confocal microscopy and optical sectioning.
modelling the portion of adherend upon the process zone,
which is assumed to deform elastically.
By comparison with FEA, it can be shown that this is not 2.2.  Experimental methods
always the case [15]: adherend plasticity nearby the crack tip
90° peel tests were performed by using a custom micro-tensile
has a twofold implication
stage (figure 5(a)): displacement control was implemented by
(i) it makes WOS sensitive to the TSL aspect ratio (figures a uniaxial DC gear motor (M112.2DG, Physik Instrumente,
3(a) and (b)). For a given toughness and strength, smoother DE), with a 25 mm travel range and 50 nm resolution. Force
softening branches imply steeper elastic branches (i.e. measurement was performed by a canister load cell (Model

4
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

Figure 3.  Examples showing the influence of interface stiffness, Ki , on the WOS value computed by FEA. (Top panel) For a given
toughness and strength, progressively stiffer interfaces (a) result in higher WOS (b). In inset (a), the interface opening has been normalized
with respect to the stretch of the peel arm at first yield in the direction of mode I opening, ∆hy = hpeel arm y. (Mid panel) By focusing on
the process zone (c), the results of 2D FEA are reported in terms of contour of von Mises stresses plotted on the adherend in undeformed
configuration (d) and (e). More details on the numerical analysis can be found in [15]. Coloured dots indicate the correspondence between
the two TSLs indicated in inset (a) and the contour of 2D FEA shown in insets (d) and (e). For both case studies, the left hand side
corresponds to the last adhered point. In inset (d), colour ranges were adjusted to indicate the extents of the region experiencing stresses,
which suggests the length of the underlying process zone. In inset (e), colour ranges were adjusted to highlight the extents of the region
where plastic dissipation has occurred (σMises  σy ). (Bottom panel) (f) ICPeel solution set for a given WOS value. (g) Comparison
between reference WOS value used to obtain ICPeel estimates in inset (f) and the values of WOS computed with FEA with selected (Gi , Ki )
combinations, indicated by coloured markers in both insets [15].

5
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

Figure 4.  Relevant stack and layout of peel samples. (a) Sketch of Pass/PI samples. The red solid line indicates the interface of interest
along the stack. The difference between the width of the PI strip (bPI ) and that of the passivation substrate (bPass ) is highlighted. The
arrangement sketched in inset (a) features half of the real sample, which is reported in inset (b). (c) The opposite side of the sample features
the Met/PI stack, highlighting the interface of interest (red dashed line) and the width difference between PI coating and metal substrate
(bPI < bMet ).

Figure 5.  (a) Micro-tensile stage underneath confocal microscope. The grip arrangement at a 45° angle ((a) top-right inset) ensures a 90°
peel geometry during peel of the sample (b). (c) Sketch of a vertical scan range including the substrate ((c) left inset), which then appears
in the infinite focus image ((c) right inset). (d) Vertical scan range excluding the substrate ((d) left inset). The resulting infinite focus image
features only the PI peel arm ((d) right inset), proving that segmentation between substrate and peel arm can be performed during the raw
acquisition.

31, Honeywell, US), with a measuring range of  ±5 N and a The micro-tensile stage was designed to fit the sample plate
resolution of 1 mN. The grip system was designed to set the of a profilometer for industrial metrology (LEXT-OLS4100,
sample at an angle of 45° with respect to the moving axis Olympus, JP) and in situ imaging of the samples was addressed
(figure 5(a), top-right inset), ensuring a peel angle of 90° with by means of confocal laser scanning microscopy (CLSM).
no need of sliding fixtures (figure 5(b)) [16, 17]. The grip In continue, few basic notions of CLSM will be recalled
system was fabricated by stereolithography, using a commer- to discuss the benefits of this method in the view of image
cial photocurable resin (Tough FLTOTL03, Formlabs, US). processing and peel arm shape segmentation. In CLSM, a

6
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

laser source probes material points individually, i.e. an in-


plane (x, y) scan is performed to image the full field of view. A
dimensional filter confocal to the light source filters the out-
of-focus information, which would otherwise cause blurring.
This effect is referred to as optical sectioning [18]. The sec-
tioning pitch is related to the depth of field of the objective.
By measuring the reflected intensity along an out-plane scan
within a finite vertical range, [zmin , zmax ], the in-focus informa-
tion of every material point in the field of view is progressively
collected and merged into a so called infinite focus image [18].
When pursuing in situ imaging of a peel test, one option is
to image the sample from the side: this choice is more typical
in wide field optical microscopy [4, 12, 16] and scanning elec-
tron microscopy [19], since the whole peel arm profile lies
in-plane and it can be better focused all at once. Segmenting
the peel arm to convert the qualitative profile into a quantitative
coordinate set requires the development of image processing
recipes [12]. To this aim, an appropriate contrast of the peel
arm with respect to the background and the substrate must be
ensured during acquisition.
In this view, the image formation principle of CLSM pro-
vides an inherent advantage: by tuning the bounds of the
vertical scan, optical sectioning can be used to selectively
include/exclude parts of the sample laying on different planes
(figures 5(c) and (d)), provided that these can be resolved by
the sectioning pitch. As mentioned, samples were cut with an
excess of about 20 µm with respect to the PI layout, which is
far beyond the depth of field of typical objectives for surface
metrology [20]. Specifically, a long working distance objec-
tive with 100×  magnification was used, with a field of view
of 125  ×  125 µm2 (LMPLFLN100×, Olympus, JP). In situ
(or better, in-operando) scans were taken during steady-state
delamination at regular intervals of 5 s. Optical sectioning
was exploited to exclude the edge of the substrate from the Figure 6.  (a) Raw infinite focus image ( I ) in grey level contrast
(212 levels). (b) Segmented image (S ) in binary contrast (2 levels,
vertical scan (figure 5(d), left panel). In these conditions, the 1/0), obtained by thresholding I with a threshold intensity of
peel arm is the only surface providing non-negligible reflected 20% with respect to the full grey scale range of I . (c) Plot of peel
intensity, hence this acquisition mode inherently segments arm edges (blue and red solid lines) identified by detecting the
the feature of interest (figure 5(d), right panel). In practice, boundaries between black and white pixels in S .
the image processing recipe to identify the peel arm profile
requires only two steps functions were considered an appropriate option to provide
an experimental reference to be compared with the full ICPeel
(i) binarization with a threshold of 20% of the full grey-level solution set. All image processing was performed in Matlab
range (figures 6(a) and (b)); environment (MathWorks, US). Specifically, the fit was per-
(ii) boundary detection of connected white pixel regions formed using a toolbox freely available in the sharing platform
(figure 6(c)). of Matlab users [22].
The root radius was estimated by best fitting the segmented
peel profiles and computing the peak of the curvature dis- 2.3.  Experimental results
tribution (figures 7(c) and (d)). Best fit was pursued using
constrained least-square splines, with cubic Hermite poly- Quantitative results represent confidential information: in
nomials as shape functions. The latter are commonly used continue, the contents of plots and tables  will be presented
in beam element modelling, when shear deformations are in a normalized fashion. It is worth mentioning that, before
neglected (i.e. Euler beam theory) [21]. The ICPeel method testing a sample, a sharp blade (Solingen, DE) was gently
models the kinematics of the peel arm by using Euler beam driven at the interface of interest to obtain a free standing seg-
theory in the large displacements regime, which accounts for ment of PI (3 to 5 mm in length) for clamping purposes. This
large rotations but not for shear strains (Kinloch et al [9] only manual procedure may introduce defects at the peel root (e.g.
mentions a scalar correction factor to account for shear effects notches in the peel arm) and detriment the rate of success of
when relating root rotation and curvature, not to describe the the test: applying an external force did more often result in
peel arm kinematics). For this reason, Hermite polynomial peel arm failure rather than interface delamination. For this

7
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

Figure 7.  (a) Peel force per unit width (i.e. WOS) measured with the micro-tensile stage during peel of Met/PI (dashed line) and Pass/PI
(solid line) samples. (b) Peel arm profile in the neighbourhood of the root identified by segmenting in situ CLSM images of Met/PI (dashed
line) and Pass/PI (solid line) samples. (c) Example of least square spline (black solid line) fitting the experimental profile (blue dotted line).
The red dot indicates the location of the peel root, corresponding to the peak of curvature distribution (d). The example reported in insets
(c) and (d) refers to the Met/PI sample, but all arguments apply to the Pass/PI sample, as well.

reason, only 1 successful in situ peel test was achieved for Table 1.  Set of input variables for ICPeel algorithm calculations.
each interface of interest. For Pass/PI samples, three further The mechanical properties of the peel arm are confidential and were
reported symbolically. The normalization factor of the Ki range
peel tests without in situ CLSM were successfully performed,
expresses the stiffness of the peel arm in the direction of mode
showing a maximum deviation of 5% in the peel force value. I opening, KPI = 2EPI /hPI [10]. © 2018 IEEE. Reprinted, with
In the configuration illustrated in figure 5, the sensing axis permission, from [15].
of the load cell is aligned with the direction of clamp displace-
Input variable Value/range/symbol
ment, i.e. at 45° with respect to the actual direction of peel.
The measured signal (Flc ) represents the projection of Fpeel Peel arm Young modulus EPI
along the load cell axis, hence the peel force was computed Peel arm yield stress σy
as Fpeel = Flc / cos (π/4), whereas the WOS was measured Peel arm yield strain y
according to (1), with θ = π/2 . Peel arm hardening slope α
At steady-state delamination, the WOS of Pass/PI sam- Peel arm ultimate stress, σu 1.05 σy
ples was found to be 63% higher than that of Met/PI samples Peel arm thickness, hPI [mm] 0.010
(figure 7(a)). Accordingly, the comparison of peel profiles Peel arm width, bPI [mm] 1
revealed a sharper curvature for Pass/PI samples (figure 7(b)). Interface stiffness, Ki 10−6  ⩽  Ki /KPI   ⩽  106
After best fitting the experimental profiles, the curvature dis- Peel angle, θ [rad] π/2
tribution of the least square spline was computed as
d2 y
3.  Identification of traction-separation parameters
dx2
χ= 
(8) Ä ä2  32 3.1.  Matching experiments and ICPeel estimates
dy
1 + dx
When the interface stiffness is unknown, the ICPeel method
and the root location was identified as the x coordinate at the can identify an infinite set of (Gi , Ki ) eligible combinations for
curvature peak (figures 7(c) and (d)) [12]. The root radius mea- the same value of WOS (figure 2). As discussed in section 1.2,
sured for Pass/PI profiles (24.7  ±  0.4 µm) was 14% smaller the ICPeel method was developed to measure the toughness of
than that of Met/PI (28.7  ±  0.3 µm). This indicates that the PI structural adhesives, for which the stiffness is always known
underwent higher plastic dissipation when peeled off from the based on the Young modulus and the thickness of the adhe-
Pass substrate, which is consistent with a higher WOS value. sive ( Ea and ha , respectively). For this reason, the available
To quantify the extent of plastic dissipation and estimate the version of the ICPeel algorithm features Ki in the inputs set
true adhesion energy, experimental data were compared with [10] (indeed, Ea and ha rather than Ki = Ea /ha directly). To
ICPeel estimates and FEA (section 3). compute a full set of (Gi , Ki ) parameters, multiple instances of

8
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

Figure 8.  Schematic representation of the identification procedure pursued by combining mechanical testing, in situ imaging and the
ICPeel method.

the same case study (i.e. a 90° peel experiment with a given have occurred [8]. To cope with this issue, TSL identification
experimental WOS value) were solved, spanning a broad was further addressed by FE modelling.
range of eligible Ki values. The full set of inputs is reported
in table 1. These include the mechanical properties of the peel
arm, which were obtained by configuring the micro-tensile 3.2.  Matching experiments and FEA
stage to perform uniaxial tension and testing free standing PI
strips (1  ×  10 mm2) up to failure. Stress/strain data were fit by A two-dimensional (2D) plane strain finite FE model was
using a bilinear hardening model. developed using a commercial solver (Abaqus, Dassault
® Systemes, FR). The peel arm was modelled as an elastic plastic
  < y material with bilinear work hardening (figure 10), using the
(9) σ = EPI .
y + α ( − y )   y same constitutive parameters presented for ICPeel calcul­
ations (table 1). Quadrilateral plane strain elements were used
The schematic flow of the identification procedure is to implement a structured mesh with more than ten elements
reported in figure 8: as mentioned, tensile testing and 90° peel through the peel arm thickness and a characteristic dimen-
provide, respectively, the stress/strain behaviour of PI and the sion of 200 nm in the direction of crack propagation (figure
WOS needed to compute all (Gi , Ki ) combinations, from pure 10). The free end of the peel arm was constrained by kine-
elastic bending (Gi = WOS) to dominant plastic dissipation matic coupling to the displacement of a reference point. In
(Gi → 0). In situ imaging of the peel profile enables the esti- turns, the latter was constrained by means of simple supports,
mation of the root radius (section 2.3), which can be used to in order to remove the horizontal degree of freedom only.
enter the ICPeel plot of TSL parameters and identify the best In-plane rotations were left free whereas the vertical displace-
matching (Gi , Ki ) pair. ment was controlled to simulate progressive peel. Noteworthy,
It is worth recalling that the root radius range is lower the reaction force of the reference point directly provides
bounded by R0,min, a limit case which essentially indicates the WOS estimated by the FEA, since forces are expressed
that bending the peel arm further would require more plastic by default per unit width in plane strain models. Owed to
dissipation than the total WOS available, which is clearly non the stiffness mismatch between the film and the substrate
feasible. For both samples, the lower bound predicted by the ( EPI /Esubstrates  <  0.05), the mechanics of the latter is expected
ICPeel algorithm was bigger than the experimental values of to play a negligible role: hence, the substrate was modelled
R0 estimated from in situ images (figure 9 and table 2), thus as a linear elastic solid with arbitrarily high stiffness. At its
preventing the identification of a best matching TSL as pro- bottom surface, vertical displacements and in-plane rotations
posed in figure 8. Experimental values of root radii were lower were constrained.
than four times the thickness of peel arm, which suggests that A cohesive surface interaction was used for lumped-
local deformation phenomena neglected by beam theory may parameter modelling of the interface. The built-in bilinear

9
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

TSL model was used, which is a three-parametric model fea-


turing a linear softening branch beyond the interface strength,
tc [23]. Mode I and mode II behaviours were assumed to be
equal. This assumption is a simplification forced by the avail-
able experimental results.
Several combinations of TSL parameters were tested to
match experimental results, both in terms of peel force and
local peel profile. In order to quantify and compare different
TSLs, the following error measurements were introduced
|WOSexp − WOSFEA |
eWOS =
(10)
WOSexp

»
2
i (yexp (xi ) − yFEA (yi ))
e(x,y) =
(11) ´ max(x)
y (x) dx
min(x) exp
Figure 9.  Adhesion energy (black lines) estimated by the ICPeel
algorithm for different interface stiffness values (red lines). The
background colour indicates whether the peel arm bends elastically »
(green), or it experiences plastic dissipation only during bending ē = e2WOS + e2(x,y)
(12)
(blue) or also upon reverse bending (orange). Energy values have
been normalized with respect to the WOS of Met/PI samples. The where the subscripts (·)FEA and (·)exp refer to FEA and exper­
normalization factor of the Ki range expresses the stiffness of the imental results.
peel arm in the direction of mode I opening, KPI = 2EPI /hPI [10]. In comparison to the 2D solution set by ICPeel, (Gi , Ki ),
© 2018 IEEE. Reprinted, with permission, from [15].
the parameter space to sample by FEA is three-dimensional
Table 2.  Comparison between the root radius computed from (3D), (Gi , Ki , tc). Despite ascertaining an inconsistency
in situ profiles and the lower bound estimated using the ICPeel between experimental results and ICPeel estimates, compar-
algorithm. © 2018 IEEE. Reprinted, with permission, from [15]. ison between the two suggests that the peel arm experienced
Interface Pass/PI Met/PI extensive plastic dissipation during delamination. By focusing
on the plastic loading/unloading regime (figure 9, orange
R0 measured in situ [µm] 24.7  ±  0.4 28.7  ±  0.3 region), selected ICPeel solutions were reproduced by 2D
min R0 by ICPeel [µm] 29 37 FEA. Travelling along the ICPeel curve towards smaller root
radii (i.e. higher Ki ) implies increasing strength, according to
constraint (2). For tc in the range of the yield strength of the
peel arm (σy), the latter may undergo necking before devel-
oping sufficient stress to cause delamination. This typically
causes divergence of the analysis. For this reason, sampling
the plastic loading/unloading region can stop as soon as
analysis divergence due to necking is encountered. The con-
verging solutions provide a characterization of the drift from
experimental references when relaxing the ICPeel assumption
of beam kinematics (figure 11(a)).
As anticipated in section 1.3, the aforementioned mismatch
is expected to have a specific trend. Stiffer TSLs (from cold
to hot colours in figure 11(a)) imply increasing plasticity of
the adherend upon the process zone. Hence, TSLs matching
well the WOS do underestimate the sharpness of the peel arm
curvature and, vice versa, stiffer TSLs approaching well the
peel profile largely overestimate WOS.
For convenience, two specific actions were devised to
sample the (Gi , Ki , tc ) parameter space:
(b.1) relaxing constraint (2) to explore different TSL aspect
ratios. As anticipated in figures 3(f)–(g), decreasing
Ge /Gi leads to sharper profiles, which implies higher
Figure 10.  Sketch of the FE model and boundary conditions.
WOS owed to higher plasticity in the adherend upon
The red line indicates the pre-cracked region, whereas the green
line indicates the cohesive interaction. Constitutive modelling of the process zone (figure 11(b.1));
the peel arm and the interface has been sketched to indicate the (b.2) scaling the TSL for a given aspect ratio, i.e. scaling
parameters required to perform an analysis. toughness and strength by keeping Ge /Gi constant,

10
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

Figure 11.  Schematic representation of the identification procedure pursued by combining ICPeel and FEA.

which ultimately means keeping a constant stiffness.


For instance, down scaling Gi by a given factor (and
tc by its square root to keep Ki constant) helps com-
pensating for excessive WOS, smoothing in turns the
peel arm curvature (figure 11(b.2)).
Given the specific mismatch trend of the ICPeel initial
guesses (figure 11(a)), either action decreasing one error type
would increase the other one. Satisfying match with exper­
imental results requires a proper iteration of actions (b.1) and
(b.2), which was pursued minimizing the effective error (ē ) by
trial and error. The identified TSLs and their parameters are
reported in normalized fashion in figure 12 and table 3. Figure 12.  Plot of best matching TSLs identified by FEA.
For both interfaces, the actual contribute of adhesion to
the total WOS (Gi /WOS) was in the range of 30%, thus con- stiffness relies on the availability of the peel arm profile data,
firming the dominant role of plastic dissipation. Noteworthy, which supports the relevance of in situ imaging. Table 3 shows
the comparison between Pass/PI and Met/PI in terms of adhe- that the values of interface strength approach the yield stress
sion energy (+67%) differs (though not dramatically) from and hence, according to table 1, the ultimate stress of PI meas-
comparing their WOS values directly (+63%). In addition, ured during uniaxial testing (σu = 1.05σy ). This may explain
although tougher, Pass/PI was estimated to be significantly the competition between interface delamination and peel arm
softer than Met/PI (−50%). Selecting an appropriate interface failure observed during the experiments.

11
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

Table 3.  Parameters of best matching TSLs identified by FEA. Acknowledgments


TSL parameter (normalized) Pass/PI Met/PI
Matteo Laganà (Gemma Prototipi, IT) is greatly acknowl-
Interface toughness, Gi /WOS 0.32 0.33 edged for the stereolithography service used to manufac-
Interface stiffness, Ki /KPI 4.75 9.50 ture the grippers of the peel stage. John D’Errico is greatly
Interface strength, tc /σy 0.89 0.99 acknowledged for the development of the Matlab tool box
Critical opening, δc /∆hy 0.09 0.05 used for constrained least-square spline fitting [22].
Failure opening, δf /∆hy 1.00 1.85
WOS error, eWOS <6% <6%
Profile error, e(x,y) <6% <7% ORCID iDs
Effective error, ē <8% <8%
Emanuele Cattarinuzzi https://orcid.org/0000-0002-3066-
4. Conclusions 0064

The adhesion of PI to inorganic passivation and metal caps


References
has been measured using small scale peel testing. Pass/PI
samples were found to be over 60% tougher than Met/PI. In
situ confocal microscopy has been successfully used to image [1] Frear D R 1999 Materials issues in area-array microelectronic
packaging JOM J. Miner. Met. Mater. Soc. 51 22–7
the deformed shape of the peel arm upon steady state delami- [2] IPC Plastic Chip Carrier Cracking Task Group
nation. The use of optical sectioning has shown promise in (B-10a) 2014 Joint IPC/JEDEC standard for
the view of practical peel profile segmentation, as the signal moisture/reflow sensitivity classification for
coming from outside the region of interest was inherently fil- nonhermetic surface-mount devices (www.jedec.org/
tered out during confocal acquisition (figure 6). document_search?search_api_views_fulltext=j-std-020E)
It is worth noting, in situ imaging of the peel profile pro- [3] Elices M, Guinea G V, Gmez J and Planas J 2002 The cohesive
zone model: advantages, limitations and challenges Eng.
vided an extra reference for the identification of a bi-linear Fract. Mech. 69 137–63
cohesive law. For case studies compliant to the assumption of [4] Omiya M, Kishimoto K, Inoue H and Amagai M 2002
beam kinematics, estimating the curvature at the peel root pro- Measurement of copper thin film adhesion by multi-
vides the closure to the analytical ICPeel method in the case stages peel test Proc. of the 4th Int. Symp. on Electronic
of interfaces with negligible thickness (figure 8). In the event Materials and Packaging (https://doi.org/10.1109/
EMAP.2002.1188885)
of excessive bending at the root ( R0  4hpeel arm ), the solution
[5] Ostrowicki G T and Sitaraman S K 2012 Magnetically
falls outside the framework of beam theory [8], but ICPeel actuated peel test for thin films Thin Solid Films
estimates can still serve as initial guesses to sample the TSL 520 3987–93
parameter space by FE modeling (figure 11). Experimental [6] IPC, Flexible Circuits Test Methods Subcommittee (D-15)
WOS and peel profile shape were matched with an effective 2014 Peel Strength, Flexible Dielectric Materials
error below 8% (table 3) by scaling the TSL and tuning its (Association Connecting Electronics Industries)
IPC-TM-650
aspect ratio. Although the identification was pursued by trial [7] Kendall K 1975 Thin-film peeling—the elastic term J. Phys.
and error, a future perspective consists in the development of D: Appl. Phys. 8 1449–52
an optimization scheme based on the minimization of the mis- [8] Kim K S and Aravas N 1988 Elastoplastic analysis of the peel
match measurements (10–12). test Int. J. Solids Struct. 24 417–35
Another relevant perspective regards sample preparation: [9] Kinloch A J, Lau C C and Williams J G 1994 The peeling of
as mentioned in section 2.3, pre-cracking the sample by hand flexible laminates Int. J. Fract. 66 45–70
[10] Georgiou I, Hadavinia H, Ivankovic A, Kinloch A J, Tropsa V
detriments the rate of success and, in practice, prevents to
and Williams J G 2003 Cohesive zone models and the
collect a statistically significant batch of measurements. The plastically deforming peel test J. Adhes. 79 239–65
relief of a free standing PI segment to be used for clamping [11] European Structural Integrity Society 2010 A Protocol For
purposes may be addressed by introducing sacrificial release Determination Of The Adhesive Fracture Toughness Of
layers during microfabrication steps, as proposed in [24] in a Flexible Laminates By Peel Testing: Fixed Arm And T-Peel
different experimental framework. Methods (London: ICPeel)
[12] Kawashita L F, Moore D R and Williams J G 2005 Analysis
Besides instancing representative cohesive laws (figure 12), of peel arm curvature for the determination of fracture
identification results allow to discriminate how much external toughness in metal-polymer laminates J. Mater. Sci.
energy was spent for delamination and plastic dissipation. For 40 4541–8
both interfaces, the true adhesion energy was found to be only [13] Béfahy S, Lipnik P, Pardoen T, Nascimento C, Patris B,
1/3 of the measured WOS. This confirms that results based on Bertrand P and Yunus S 2010 Thickness and elastic
the sole measurement of the peel force are strongly affected modulus of plasma treated PDMS silica-like surface layer
Langmuir 26 3372–5
by plastic dissipation of the peel arm. The peel profile avail- [14] Wei Y and Hutchinson J W 2003 Peel test and interfacial
able from in situ imaging is an essential reference to identify toughness Comprehensive Structural Integrity ed I Milne,
such competing contribute. R O Ritchie and B Karihaloo (Oxford: Elsevier) pp 181–217

12
J. Micromech. Microeng. 29 (2019) 034001 E Cattarinuzzi et al

[15] Cattarinuzzi E, Gastaldi D, Vena P, Pozzobon F, Guarino L, [19] Sargent J P 1998 Microextensometry, the peel test, and
Cecchetto L and Andreini A 2018 Mechanical the influence of adherend thickness on the measurement
reliability of microelectronics packaging: small scale of adhesive fracture energy Int. J. Adhes. Adhes.
adhesion measurements and in situ imaging 19th Int. 18 215–24
Conf. on Thermal, Mechanical and Multi-Physics [20] Olympus 2014 Bringing answers to the Surface: Lext
Simulation and Experiments in Microelectronics and OLS4100 Brochure (M1775E-072014) (Tokyo: Olympus
Microsystems (EuroSimE) (https://doi.org/10.1109/ Corporation)
EuroSimE.2018.8369889) [21] Abaqus 6.14 Documentation 2014 3.5.3 Euler-Bernoulli
[16] van der Sluis O, Hsu Y Y, Timmermans P H M, Gonzalez M beam elements Theory Manual (Providence, RI: Dassault
and Hoefnagels J P M 2011 Stretching-induced interconnect Systemes)
delamination in stretchable electronic circuits J. Phys. D: [22] D’Errico J 2009 SLM–Shape Language Modeling
Appl. Phys. 44 034008 (Natick, MA: MATLAB Central File Exchange)
[17] Cattarinuzzi E, Lucchini R, Gastaldi D, Vena P, Lorenzelli L (https://it.mathworks.com/matlabcentral/
and Hoefnagels J P M 2015 In situ experimental fileexchange/24443-slm-shape-language-modeling)
characterization of interfacial toughness of aluminum thin [23] Abaqus 6.14 Documentation 2014 36.1.10
films on polyimide substrates Proc. of the 20th Int. Conf. on Surface-based cohesive behavior Analysis User’s
Composite Materials (ICCM20) Manual (Providence, RI: Dassault Systemes)
[18] Leach R 2011 Optical Measurement of Surface Topography [24] Bagchi A and Evans A G 1996 The mechanics and physics
(Berlin: Springer) (https://doi.org/10.1007/978-3-642- of thin film decohesion and its measurement Interface Sci.
12012-1) 3 169–93

13

You might also like