You are on page 1of 7

In-depth Raman Spectroscopy Analysis of Various Parameters Affecting the Mechanical Stress

near the Surface and Bulk of Cu-TSVs


Ingrid De Wolf*, Veerle Simons, Vladimir Cherman, Riet Labie, Bart Vandevelde and Eric Beyne
IMEC
Kapeldreef 75, B-3001 Leuven
*
Also at dept. MTM, KU Leuven
Abstract
This paper discusses mechanical stress measured with
micro-Raman spectroscopy in the silicon substrate near CuThrough Silicon Vias (TSV). A discussion of the relation
between the observed Raman shift and the various stress
tensor components is given, showing that this relation is often
wrongly applied, and that in many cases the compressive
stress along the vertical axis of the TSV, dominates the Raman
results and hides the tensile axial component which is of most
relevance for its impact on CMOS devices. The effect of
measurement depth, TSV depth and density, and an oxide cap
is shown. Both surface and cross-sectional results are
discussed. Also a direct correlation between results from
Raman measurements and electrical results from FET-arrays
near a TSV is given.
Introduction
3D technology requires several non-standard processing
steps such as the fabrication and filling of high aspect-ratio
through silicon vias (TSVs). The Cu inside the TSVs is known
to be under tensile stress. This stress not only can affect the
performance of nearby devices, it can be expected to cause
additional reliability problems in the BEOL layers (ex. Cu
pumping) or affect the isolation and barriers surrounding the
TSV. Even stress in the Si, near the bottom of the TSV, might
affect the TSV behavior after the Si wafers are thinned and
that part of the TSV is released for bonding.
Local mechanical stress in Si can be studied using microRaman spectroscopy (RS). [1] Several recent publications
correlate RS results measured in the Si near TSVs, with
results from models and electrical measurements. [2-7] These
studies however all have one assumption in common: They
neglect the stress component along the length axis of the TSV
in their calculation of stress from the Raman frequency shift,
and assume unixial or biaxial stress in the Si surface plane.
This assumption is made because it results in a linear relation
between stress and Raman frequency shift, but, as will be
shown in this paper, it is incorrect and can lead to erroneous
conclusions. A better approach is to use an analytical model or
finite element modeling (FEM) to predict the strain tensor
elements in the silicon, and then calculate the expected Raman
shift. This approach, which was developed for Si isolation
structures [8,9], was successfully applied to Cu vias [10,11]
but is time-consuming and requires FEM, it does not offer fast
insight in the stresses.
In this paper, a triaxial stress state is proposed to describe
the relation between stress induced by TSVs and the measured
Raman frequency shift in silicon. Using Raman measurement
results obtained at different depths in the silicon, and results

978-1-4673-1965-2/12/$31.00 2012 IEEE

from TSVs with different lengths, it is shown that the vertical


stress component (in depth direction) has a very important
influence on the Raman results and cannot be neglected, as is
commonly done. We not only show measurements performed,
as is conventionally done, from the top surface of the Si near
the TSV, but also from 2-dimensional X-sectional analysis.
This provides information of stresses near TSVs at deeper
depths in the Si and even in the Si underneath a TSV (before
wafer thinning).
Finally, we discuss RS-measurements performed on the
X-section at a transistor (FET) array located near a TSV on a
fully processed wafer. It is shown that this analysis is not
possible from the top Si surface, because the poly-Si on top of
the gate of the devices masks the signal from the Si substrate.
The X-sectional measurements are directly correlated with the
electrical results obtained on the same devices and confirm
that the TSV induced stress extends several microns away
from the TSVs and affects the devices.
Stress near Cu TSVs
It is well known that there is tensile mechanical stress in
the Cu inside TSVs at room temperature, at the end of the
processing cycle. The magnitude of this stress depends on
many factors: Cu chemistry, deposition and anneal
temperatures, TSV dimensions etc. This stress in the Cu TSV
will introduce stress in the surrounding silicon substrate. Fig.
1 shows a simplified model indicating the various stress
components in cylindrical (left) and Cartesian (right)
coordinates.

Fig. 1: Stress components introduced by a Cu-TSV in the


surrounding silicon. Left: Cylindrical coordinates. Right: In a
Cartesian axis system with x taken along [110], y along [001]
and z along [-110].
Because Cu is under tensile stress, it will introduce tensile
stress in the silicon along the radial direction (r). The
circumferential stress c is expected to be compressive. Along
the depth, also the axial component, a, is expected to be
compressive. If we look at the stress in a point P, located in
the x-y plane in the xyz-axes system, this translates into a
tensile stress xx and compressive stress yy and zz. For
simplicity shear stress components are neglected here. This is
valid only far away from the TSV top and bottom edges. The

331

axis system is chosen such that the silicon surface is in the XZ plane and the X-section in the X-Y plane. This choice is
made to simplify the calculations of the Raman selection rules
and shifts for measurements on the cross-section XY-plane.

assumption for the RS result interpretation. In that case, the


relation between the Raman shift and the three stress
components as shown in Fig. 1 is given by:

Raman spectroscopy as stress sensor


The Raman frequency of silicon () changes due to strain.
The relation between ( = (strain) o (no strain)) and
the strain is given by the eigenvalues from the following
secular equation [8,9]:

Triaxial:
(cm-1) = -1.93x10-9 (xx + zz) - 0.75 10-9yy (Pa)

(1)
This equation is given for an axis system x=[100], y=[010],
z=[001]. In the general case when all the strain tensor
elements are non-zero, this equation cannot be solved and
modeling is required to obtain information on the stress
components. If there is uniaxial or biaxial stress, the solution
is straightforward and the relation between Raman peak
frequency and stress becomes simply linear. For the Cartesian
system of Fig. 1, this is:
Uniaxial:

Biaxial:

(cm-1) = -1.93x10-9 xx (Pa)


(2)
xx (MPa)= 518 (cm-1)

(cm-1) = -1.93x10-9 (xx + zz) (Pa)


xx + zz (MPa)= 518 (cm-1)
(3)

The exact number (518) depends on the material property


values used for p, q, and r in Eq. 1. Depending on the values
used, it can range between 430 and 520.
Eq. 2 (uniaxial stress) is often used to correlate Raman
measurements near TSVs with stress.[2-7] However, the
assumption of uniaxial stress is not correct for TSVs and can
lead to erroneous conclusions. Indeed, lets assume that a
Raman shift of -0.038 cm-1 is measured. Eq. 2, which is often
applied for RS studies, would translate this into xx = 20
MPa, tensile. Now assume that there is not uniaxial but biaxial
stress and that xx is 100MPa (tensile) and zz is -80MPa
(compressive). According to Eq. 3 that would also give a
Raman shift of -0.038 cm-1. Using Eq. 2 would thus give a
large underestimation of the tensile radial stress component,
xx. This is the stress component that is often correlated with
electrical measurements showing impact of TSV induced
stress.
Actually, also assuming biaxial stress is incorrect in the
case of TSVs, because it neglects the third stress component,
the axial component, yy. This component has to be taken into
account and in reality often will dominate the Raman results.
Indeed, analytical and finite element models indicate that xx
and zz have a similar magnitude but opposite sign, i.e.
according to Eq. 3 no Raman shift would be measured
although the stresses can be very high. In reality a Raman shift
is measured, but it is dominated by the axial stress component,
yy. For the further discussions, we use a triaxial stress

(4)

Although the assumption that the shear strain is zero is not


valid, certainly not close to edges of the TSV, this relation is
much more reliable to determine TSV induced stress than Eq.
2 and 3, and will provide more realistic values and insight in
TSV induced stresses.
Experimental
RS experiments were performed with either the 458 nm
or 633 nm laser line as excitation wavelength. For the first, the
penetration depth in the silicon is about 300 nm, so only the
stress near the top surface is measured. For the latter, the
penetration depth is about 3200 nm, and with this laser line
stresses at a larger depth will dominate the results [12]. The
spot size on the sample is about 1 m. All the results were
calibrated using the plasma lines of the laser. [1] Experiments
were performed on Cu TSVs with 5 m diameter and a depth
of 22 or 40 m. Experiments were done on two kinds of Cu
chemistry (called A and B).
Raman spectroscopy results: top-surface measurements
In this first results part we first recall a previously
published experiment but discuss it in view of Eq. 4. In [12],
we showed the stress measured near two TSVs and nearby
shallow trench isolation (STI) islands. Here we focus on the
results near the TSVs. A RS-experiment was done scanning
the top surface of the silicon across two Cu nails, as indicated
in the inset (top-right) of the figure.

T
S
V

T
S
V

Fig. 2: Stress induced Raman frequency shift () measured


on (100) Si surface, scanning across two 5 m diameter CuTSVs (orange bars) (Chemistry A) using a 458 nm and a 633
nm laser line. At the position of the TSVs no signal is
obtained.
The measurement was done using the 458 nm (blue) line
of the laser, which has a penetration depth in Si of about 300
nm and thus probes the stress in the top layers. The
measurement was repeated at the same position using the 633
nm laser light (red line), which has a much larger penetration
depth (about 3200 nm) and thus deeper stress fields will

332

contribute to the signal. The results are clearly different. Near


the surface, remains overall negative close to the TSV,
indicating that a tensile stress component dominates. When
probing deeper, becomes positive near the TSVs,
indicating that compressive stress dominates the Raman signal
close to the TSV, and only a small tensile component is
measured further away. Okoro et al. [13] performed
simulations of the TSV induced stress. Some of the results are
shown in Fig. 3. Far away from the top surface of the TSV, the
shear stress becomes zero and the radial (xx) and
circumferential (zz) stress become equal but with opposite
sign. From Eq. 4 follows that in this case the Raman shift will
be dominated by the compressive axial stress. We indeed
observe in nearly all experiments done with the 633 nm line a
dominating positive shift of the Raman peak. This experiment
shows clearly the pitfalls of Raman spectroscopy. The tensile
axial stress can be of the order of 100 MPa or more but remain
undetected in a RS-experiment because its effect on is
suppressed by the other stress components.

for the shorter one (~ 0.15 cm-1). One can assume that this
effect is entirely due to changes of the axial stress component,
zz, because the diameter of the TSV did not change.

T
S
V

Fig.4: Stress induced Raman frequency shift () measured


on (100) Si surface, scanning across a 5 m diameter and 22
m (red) and 40 m (black) deep Cu-TSV (orange bar)
(Chemistry B) using a 633 nm laser line.
Fig. 5: Finite element
model of the triaxial stress
components introduced by
a 5 m diameter, 40 m
deep TSV in the
surrounding Si at a depth
of 2 m below the Si
surface.

Fig. 3 Plot of the stress profile in Si, starting from the edge of
a 5 mm wide Cu-TSV at room temperature. (a) At the center
region, i.e. deep in the silicon far from the TSV top and
bottom; (b) at the top of the Si. From [13].
Near the surface, the axial stress is highly compressive
close to the TSV, but falls fast back to zero and the picture is
dominated by the tensile shear and radial stress, and the
compressive circumferential stress (Fig. 3). From the Raman
results, it seems that the tensile components dominate. It is
more difficult to make conclusions, because we do not know
the effect of the shear stress on the Raman shift, and in
addition the stress near the top surface of the sample is also
influenced by STI layers.
Fig. 4 shows the effect of TSV length on the Raman shift
for 22 and 40 m deep TSVs, diameter 5 m. The
measurements were done using the 633 nm laser line, so
results are from a relative large depth where we can assume
that the shear stress can be neglected. In this experiment the
Cu chemistry inside the TSVs differs from the one in Fig. 2.
No tensile stress component is detected. Again, this does not
mean that it is not present, yy dominates the result and xx
and zz might be large but only their difference affects .
And, as is shown in the simulations, this difference is expected
to be small. Clearly visible is that the positive shift of
close to the TSV is larger for the longer TSV (~ 0.4 cm-1) than

40 m

22 m

xx
=r

xx
=a
zz
=c

2 m below Si surface
77 MPa
Tensile

-96 MPa
Compressive
16

-230 MPa
Compressive

To further analyze these results, finite element simulations


were done to calculate the stress at a depth of 2 m below the
Si surface for a 40 m deep, 5 um diameter TSV (Fig. 5). The
depth of 2 m was chosen because the penetration depth of the
633 nm laser light, with an exponentially decaying intensity, is
3.2 m. So, the stress at that position will give a relative good
match. If we take the values of the 3 stress components at the
same point close to the TSV (indicated by arrow in Fig. 5),
and use Eq. 4 to calculate , we obtain = 0.37 cm-1. This
FEM prediction is very close to the measured value 0.4 cm-1.
This experiment again clearly shows how important it is to use
the correct relation for the Raman data. Although only a
positive shift of is measured, there might be a high tensile
stress component present in the silicon near the TSVs.

333

10 m

5 m

Fig. 6. Stress induced Raman frequency shift () across an


array of TSVs (633 nm light, chemistry B)

0.20
0.15

x (m)

10
20
30

-1

40

0.10

0.05
0.00
-0.05
-0.10

50
0

10

15

20

25

0.20

10

0.15

20
30

-1

x (m)

y (m)

(cm )

with cap

T
S
V

bottom of the TSV, also a very small patch with a negative


is observed. To study this in more detail, a second scan was
made imaging the Raman shift near the 2nd TSV from the
right, as indicated in Fig. 8, bottom. The combination of the
two results clearly shows a dominating positive Raman shift,
the overlapping stress fields in between TSVs, and indeed the
negative Raman shift underneath the TSV. Can this be
explained using Eq. 4?

(cm )

Fig. 6 shows a measurement, using 633 nm light, across an


array of TSVs. Again we can assume that the measured Raman
shift is dominated by the compressive yy component at this
penetration depth. One clearly sees that the stress fields from
the various TSVs overlap, resulting in a larger stress at the
center.
In another experiment, the influence of a 500 nm SiO2
layer on top of the TSV was studied (Fig. 7). A sample with
and without oxide cap on top was sintered at 420 oC, to mimic
the back-end-of-line (BEOL) processing temperature load.
This sintering step is expected to bring the Cu in the TSV,
because of thermal expansion, into a compressive stress state
at higher temperatures. This will introduce expansion of the
Cu and possibly plastic deformation. When cooling down to
room temperature, this plastic deformation results in the so
called Cu pumping [14,15]. The sample with oxide cap
present clearly shows a larger positive at room
temperature. This can again very likely be attributed to an
increase of the compressive axial stress component, yy. A
possible explanation is that the compressive stress build-up
during the sintering step is larger than for the case with oxide
cap, because the Cu is not free to expand in the vertical
direction, at the top of the TSV. A larger part of this stress
might remain present at room temperature. But further
experiments are required to investigate these results in more
detail.

40

0.10

0.05
0.00
-0.05

50

-0.10
0

10

15

20

25

y (m)

Fig. 8: Stress induced Raman frequency shift () (right)


measured on (110) X-section surface in the area indicated by
the yellow rectangle in the microscope image (left). 633 nm
laser line, chemistry B.
Actually, Eq. 4 is only valid for measurements in
backscattering from the (100) surface of a Si wafer. It is not
valid for measurements on the X-section surface. Solving the
secular equation (after appropriate rotation of all relevant
tensors to the correct axes system) shows that for
measurements from the X-section surface (X-Y surface in Fig.
1), assuming triaxial stress, the relation for the Raman peak
measured in our configuration (this depends on the
polarization of incident and scattered light) is given by:

without cap

Fig. 7. Stress induced Raman frequency shift ()


measured on (100) Si surface, scanning across a 5 m
diameter Cu-TSV (orange bar) with (black) and without (red)
oxide cap on top. 633 nm laser line, chemistry B.

Triaxial:
(cm-1) = -2.31x10-9 xx 1.93 10-9 yy 0.37 10-9zz (Pa)

Raman spectroscopy results: X-section measurements


In general Raman spectra are measured in backscattering
from the top surface of the silicon wafer, because this is the
easiest way and in general does not require sample
preparation. But experiments can as well be done from the
cross-sectional surface. A X-section was made through an
array with Cu-TSVs by careful polishing. Two 2D-scans were
performed to measure the Raman shift in the Si along the
width and depth of the TSVs. In a 1th scan the outer right TSV
of an array was measured. The yellow rectangle in Fig. 8, topleft, shows the analyzed region. We clearly observe a positive
Raman shift next to the TSV. The red band along the top
surface is a measurement artifact. But at the right side near the

In between the TSVs, at large depth, it can be assumed that


xx and zz are equal (Fig. 3). However, for a X-section, one
could expect that the out-of plane component, zz, is reduced a
lot, because the free silicon surface can partly relax. In
addition, about half of the TSV is gone because of Xsectioning, so, a reduced stress can be expected. Also a
reduction of xx can be expected for the same reason, but not
that much of yy. yy is compressive and probably dominates
also in X-section experiments the Raman image. Underneath
the TSV, we indeed expect a tensile value for zz because the
TSV is under tensile stress and wants to shrink. This
demonstrates the complexity of Raman spectroscopy for the
analysis of stress due to TSVs. Probably the combination of
top and X-sectional measurements, with the correct settings of

334

the Raman experiment, can provide more detail information


on the different stress components.
Correlation with electrical measurements
The main reason for the increasing interest in the analysis
of stress induced by TSVs, in addition to possible reliability
issues, is the impact of this stress on the electrical
performance of nearby electronic devices. It was shown that
this impact indeed cannot be neglected for certain
technologies [16,17]. The following experiment was to
investigate whether there is indeed a direct correlation
between drain current of FE-transistors and the stress
measured using RS. The drain current of 4 large FETs
(5mx10m), with increasing distance from a TSV (Fig. 9,
top), was measured. There is a clear variation in drain current
with distance from the TSV (Fig. 9, bottom).

FET 4

FET 3

FET 2

FET 1

TSV

The result of the Raman measurement is shown in Fig.10,


bottom. A scan was done along the line visible in the figure.
At the position of the TSV, no signal was obtained because
metal of the bond pad on top of the TSV was still present.
Left from the TSV, a wavy pattern is visible in the Raman
shift near the FETs. This is probably caused by polycrystalline
Si which is on top. The only indication of a varying stress can
be seen by observing the minima of the wavy pattern, as
indicated by the dotted line. This seems to indicate that there
is a TSV-induced stress variation in this sample. Because
these measurements are not conclusive, a X-sectional sample
was prepared. The X-sectional picture is shown in Fig. 11. For
clarity, the top view image is shown at the top to indicate the
position of the FETs. A 2D-scan as in Fig. 8 was done near the
surface below the FETs, but for clarity we only show slices of
the measurement results obtained at depths close to the Si
surface. Only these stresses are relevant for their impact on the
FETs. The measurement that is taken exactly at the surface
also shows the wavy pattern that is caused by the
polycrystalline Si. But about 0.1 m under the surface, where
the focused laser beam spot does not probe the poly-Si
anymore, this waviness is gone and a clear variation of the
Raman shift with position is seen.

Top view
FET 4

Fig. 9: Top: Microscope picture of FETs and their location


with respect to a TSV. Bottom: Drain current of the FETs.

FET 3

X-section view

FET 2

FET 1

TSV

After electrically measuring the FET arrays, all top metal


layers were etched away to allow performing Raman
measurements. These measurements were not possible on the
fresh sample because of the presence of passivation layers
giving luminescence signals and of metal layers blocking the
Raman light. A top view of the sample after etch, showing
clearly the position of the FETs, is given in Fig. 10, top.
FET 3

FET 2

FET 1
TSV

T
S
V

Fig. 10: Top: Microscope picture of the sample after etch.


Bottom: Stress induced Raman frequency shift () measured
on the (100) Si surface across the FETs and TSV area. The
TSV position is indicated by the yellow bar.

Fig. 11. Top and X-section picture of the sample. Stress


induced Raman frequency shift () measured on the (110) Si
X-section across the FETs and TSV area at the surface (top)
and about 0.1 m below the surface (bottom).
Overall is positive, indicating that a positive stress
component is dominating the picture. The origin of this
component is not clear, and it does not reduce to zero far away
from the TSV. If we neglect this component, and take a zero
reference far from the TSV (indicated by the dotted line), we
see a clear variation in the Raman shift going from more
negative close to the TSV to zero further away. If we calculate
this shift with regard to this defined zero-line at the center
position of the 4 TSVs, and plot it together in Fig. 12 with the

335

previously measured drain current of the TSVs (Fig. 9), we


see a very good correlation. They follow the same decay,
away from the TSV. This confirms that the stress extends
minimally 10 microns away from the TSVs, for this Cu
chemistry and technology. Very close to the TSV,
increases again, probably due to the axial component which
dominates the picture. One has to keep in mind that close to
the TSV and the Si surface, it is difficult to deduce the values
of the different stress components from the Raman shift.
Further experiments, both on X-section and top surface, are
required for this purpose.

TSV

FET 1

FET 2

FET 3

FET 4

Id

Fig. 12: Comparison of drain current of FETS and Raman


frequency shift deduced from Fig. 11 at the same position.
Conclusions
The paper demonstrates the strength but also the pitfalls of
RS for the study of local mechanical stress in Si. The main
conclusion is that a uniaxial or biaxial stress model cannot be
used to correlate the measured Raman frequency shift with
stress near TSVs. A triaxial model is proposed that provides
more realistic results. The measured Raman frequency shift
depends on all three axial stress components. Actually, to be
more precise also shear stress components should be included
in the calculations, making them more complex. But at large
depths below the Si surface, these components can be
neglected and the triaxial model is valid. For conventional
measurements from the top (100) Si surface, the compressive
axial component dominates the Raman frequency shift. The
tensile axial stress and compressive circumferential stress for a
large part cancel their effect on out.
Measurements can also be done from the cross-section
surface of silicon. In that case, a different relation between the
Raman shift and the stress tensor components has to be used.
This relation depends on the experimental set-up, i.e. on the
polarization of incident and scattered light and on the crystal
directions of the cross-section plane. In addition, stress
relaxation effects due to the free cross-sectional surface and
the reduced Cu nail have to be taken into account.
The presented results provide new insights in stress
induced by TSVs and the effect of various TSV parameters on
this stress.
Acknowledgments
Without the great contribution of the whole 3D-team of
imec and their industrial partners, this work could not have

been done. Part of this work was done in the frame of the EC
project eBrains.
References
1. De Wolf, I., Micro-Raman spectroscopy to study local
mechanical stress in silicon integrated circuits,
Semicond. Sci. Technol., Vol. 11 (1996), pp. 139-154.
2. Murugesam, M. et al., Wafer thinning, bonding and
interconnects induced local strain/stress in 3D-LSIs with
fine-pitch high-density microbumps and through-Si vias,
Proc IEEE electron Devices Meeting (IEDM), 2010, pp.
10-30 10-33.
3. Kwon, W.S. et al., Stress evolution in surrounding
silicon of Cu-filled through-silicon via undergoing
thermal annealing by multiwavelength micro-Raman
spectroscopy, Applied Physics Letters, Vol. 98 (2011),
pp. 232106-1 - 232106-3.
4. Le Texier, F. et al., Investigation of local stress around
TSVs by micro-Raman spectroscopy and finite element
simulation, Proc Interconnect Technology Conference
and 2011 Materials for Advanced Metallization
(IITC/MAM), 2011.
5. McDonough, C. et al., Thermal and spatial dependence
of TSV-induced stress in Si, Proc. Interconnect
Technology Conference and 2011 Materials for
Advanced Metallization (IITC/MAM), 2011.
6. Dao, T. et al., Through silicon via stress
characterization, Proc IEEE IC Design and technology
(ICICDT), 2009, pp. 39-41.
7. Tanaka, T. et al., Development of self-assembled 3-D
integration technology and study of microbump and TSV
induced stress in thinned chip/wafer, Proc IEEE Intern.
SOI Conference, 2010, pp.1-4.
8. De Wolf, I et al., Stress measurements in silicon devices
through Raman spectroscopy: Bridging the gap between
theory and experiment, J. Appl. Phys. Vol. 79, No. 9
(1996), pp. 7148-7156.
9. De Wolf, I. et al., Addendum: Stress measurements
Stress measurements in silicon devices through Raman
spectroscopy: Bridging the gap between theory and
experiment [J. Appl. Phys. 79, 7148 (1996)], J. Appl.
Phys. Vol. 85, No. 10 (1999), pp. 7484-7485.
10. Okoro, C. et al., Extraction of the appropriate material
property for realistic modeling of through-silicon-vias
using -Raman spectroscopy,
Proc Interconnect
Technology Conference (IITC), 2008, pp. 16-18.
11. Hsieh, C.C. et al., Orthotropic stress field induced by
TSV and its impact on device performance, Proc
Interconnect Technology Conference and 2011 Materials
for Advanced Metallization (IITC/MAM), 2011.
12. De Wolf, I., Raman spectroscopy analysis of mechanical
stress near Cu-TSVs. In: Stress management for 3D ICs
through silicon vias, Proc International workshop on
stress management for 3D ICs using through silicon vias.
AIP conference proceedings, Vol. 1378, 2011, pp138149.

336

13. Okoro, C., Thermo-mechanical characterization of


copper through-silicon-via interconnect for 3D chip
stacking, PhD dissertation, KU Leuven, 2010
14. Okoro, C. et al., Elimination of the axial deformation
problem of Cu-TSV in 3D integration, Proc Stressinduced phenomena in Metallization:11 th International
Workshop, AIP conference proceedings, 2010, Vol.
1300, pp.214-220.
15. De Wolf, I. et al. Cu pumping in TSVs: Effect of preCMP thermal budget, J. Microelectronics Reliability,
Vol. 51, No. 9-11 (2011), pp. 1856-1859.
16. Mercha, A. et al. Impact of thinning and through silicon
via proximity on high-k/metal gate first CMOS
performance, Proc IEEE Symposium on VLSI
technology, Honolulu, HI, June 2010, pp. 109-110
17. Minas, N. et al. 3D integration: Circuit design, test and
reliability challenges, Proc. 16th IEEE International online testing symposium (IOLTS), Corfu, Greece, July
2010, pp. 217

337