ETH Library
Experimental investigation into
vortex structure and pressure
drop across microcavities in 3D
integrated electronics
Journal Article
Author(s):
Renfer, Adrian; Tiwari, Manish K.; Brunschwiler, Thomas; Michel, Bruno; Poulikakos, Dimos
Publication date:
2011-09
Permanent link:
https://doi.org/10.3929/ethz-b-000039358
Rights / license:
In Copyright - Non-Commercial Use Permitted
Originally published in:
Experiments in Fluids 51(3), https://doi.org/10.1007/s00348-011-1091-5
This page was generated automatically upon download from the ETH Zurich Research Collection.
For more information, please consult the Terms of use.
RESEARCH ARTICLE
Experimental investigation into vortex structure and pressure
drop across microcavities in 3D integrated electronics
Adrian Renfer
•
Manish K. Tiwari
•
Thomas Brunschwiler
•
Bruno Michel
•
Dimos Poulikakos
Received: 24 November 2010 / Revised: 12 February 2011 / Accepted: 28 March 2011 / Published online: 12 April 2011
Ó Springer-Verlag 2011
Abstract Hydrodynamics in microcavities with cylindri-
cal micropin fin arrays simulating a single layer of a water-
cooled electronic chip stack is investigated experimentally.
Both inline and staggered pin arrangements are investigated
using pressure drop and microparticle image velocimetry
(lPIV) measurements. The pressure drop across the cavity
shows a flow transition at pin diameter–based Reynolds
numbers (Re
d
) *200. Instantaneous lPIV, performed using
a pH-controlled high seeding density of tracer microspheres,
helps visualize vortex structure unreported till date in
microscale geometries. The post-transition flow field shows
vortex shedding and flow impingement onto the pins
explaining the pressure drop increase. The flow fluctuations
start at the chip outlet and shift upstream with increasing Re
d
.
No fluctuations are observed for a cavity with pin height-to-
diameter ratio h/d = 1uptoRe
d
*330; however, its pres-
sure drop was higher than for a cavity with h/d = 2 due to
pronounced influence of cavity walls.
1 Introduction
High-performance next-generation multi-core processors
require new architectures and advanced packaging. Verti-
cal integration improves the bandwidth from core to cache
memory, by reducing wiring length. Furthermore, it helps
in obtaining a larger processor area, which effectively
becomes the number of dies in the stack times the maxi-
mum projected lithography size for two-dimensional chips
(Ulrich and Brown 2006). Therefore, the semiconductor
industry is investing heavily into the development of
through-silicon vias (TSV), which are the most important
building blocks toward three-dimensional (3D) integration.
However, in multi-layered, 3D packages, both heat flux and
thermal resistances accumulate, resulting in junction tem-
peratures above the reliability threshold. This constrains
the electrical design and is the reason why thermal man-
agement is one of the key challenges for future micropro-
cessor development. Interlayer water cooling is far superior
to traditional air cooling techniques since it can remove the
dissipated heat directly between the individual chip layers,
promising performance increases for several device gen-
erations. The superior heat transfer capability of water
notwithstanding the pumping power associated with the
pressure drop across the chip is a criterion contributing to
the overall energy costs for cooling 3D electronic chip
stacks. To seal the electrical interconnects from water, the
TSVs are embedded into silicon pins and the overall
arrangement consists of a micropin fin array inside a mi-
crocavity. Therefore, the hydrodynamic investigation into
Electronic supplementary material The online version of this
article (doi:10.1007/s00348-011-1091-5) contains supplementary
material, which is available to authorized users.
A. Renfer M. K. Tiwari D. Poulikakos (&)
Department of Mechanical and Process Engineering,
Laboratory of Thermodynamics in Emerging Technologies,
ETH Zurich, ML J 36, 8092 Zurich, Switzerland
e-mail: dimos.poulikakos@ethz.ch;
dimos.poulikakos@sl.ethz.ch
A. Renfer
e-mail: arenfer@ethz.ch
M. K. Tiwari
e-mail: mtiwari@ethz.ch
T. Brunschwiler B. Michel
Advanced Thermal Packaging, IBM Research Laboratory,
8803 Rueschlikon, Switzerland
e-mail: tbr@zurich.ibm.com
B. Michel
e-mail: bmi@zurich.ibm.com
123
Exp Fluids (2011) 51:731–741
DOI 10.1007/s00348-011-1091-5
microfluidic chips with micropin fin arrays inside micro-
channels is of paramount importance in developing the
next-generation integrated liquid cooling of 3D chip stacks.
Classically, for macroscale flows across a single cylin-
der, a critical Reynolds number marks the inception of
vortex shedding. For an array of cylinders on the other
hand, the flow field characteristics are more complicated,
mainly due to interacting wakes (Ziada and Oengo
¨
ren
1993). In low aspect ratio (h/d) pin fin array cavities, the
effect of confinement is expected to play an additional role,
resulting in different flow regimes than in flows across
cylinder arrays and cannot be described solely through a
Reynolds number based on the cylinder diameter.
Flow across large-scale tube bundles with various con-
figurations and shapes has been extensively investigated
experimentally by several researchers using hot-wire or
laser Doppler anemometry, PIV as well as pressure mea-
surements. An overview of selected studies can be found in
Paul et al. (2007). Iwaki et al. (2004) conducted PIV
experiments using inline and staggered tube bundles
(d = 15 mm) at Reynolds numbers of 5,400 and higher.
They could identify three vortex structures behind the
tubes; however, the vector fields were obtained by
ensemble cross-correlation over 200 images. These studies
focus on macroscale tubes, but the detailed hydrodynamics
in micropin fin arrays confined in microchannels remains
unexplored. Flows with low aspect ratio pin fins are likely
to be different due to pronounced effects of cavity wall
damping (Brunschwiler et al. 2009). Kos¸ar et al. (2005)
performed an experimental study on 100-lm-long micropin
fin bundles with various configurations and reported on an
increasing pressure slope at higher flow rate because of
cylinder–wake interaction. These authors neither provided
a detailed explanation for the pressure transition nor any
flow visualization to support this observation. Prasher et al.
(2007) observed a pressure gradient transition in the
hydraulic performance of low aspect ratio micropin fins but
did not provide a physical explanation. Brunschwiler et al.
(2009) investigated various pin fin arrangements with dif-
ferent pin diameter, pitch and cavity heights for 3D inte-
grated chip cooling. For inline and staggered distorted
designs with a pitch and cavity height of 200 lm, they
observed a clear flow regime transition through pressure
measurements. No flow measurements were performed to
investigate this transition. Alfieri et al. (2010) developed a
hydrodynamics and conjugate heat transfer model of a
single row of inline micropins in order to obtain correla-
tions for modeling a 3D chip stacks simulator as a
non-equilibrium porous media. No flow transition was
considered. Liang et al. (2009) presented a numerical
approach to simulate laminar flow across a row of six
cylinders for different longitudinal pitches. They found a
vortex shedding activity moving upstream with increasing
pitch between the cylinders at a fixed Reynolds number. In
the study of Nishimura et al. (1993), an upstream devel-
opment of the transition to vortex shedding with increasing
Reynolds number was observed in an array of inline
and staggered large diameter tubes (d = 15 mm, Re
d
=
60–3,000). No experimental visualization and measure-
ment of the transition flow in micropin fin arrays confined
in cavities has been reported to this date.
Herein, we report velocity and pressure drop measure-
ments in micropin fin chips in inline and staggered dis-
torted arrangement for flows with pin diameter–based
Reynolds number (Re
d
) up to 290. The velocity measure-
ments are performed using microparticle image velocime-
try (lPIV) for flow visualization at the microscale
(Santiago et al. 1998; Wereley and Meinhart 2010).
Instantaneous velocity field measurements required high
microparticle seeding, which are prone to undesirable
coagulation and sticking to solid boundaries of the flow
domain (Lindken et al. 2009). In spite of these limitations,
unsteady and/or turbulent flow dynamics in simple geom-
etries such as micron size channels (Li and Olsen 2006a, b;
Angele et al. 2006; Blonski et al. 2007; Natrajan and
Christensen 2010), capillaries (Natrajan and Christensen
2007), obstacle-type valveless micropumps (Sheen et al.
2008), and inkjet print heads (Meinhart and Zhang 2000)
have been previously reported by measuring instantaneous
velocity fields. However, a complex geometry such as the
microcavity with the micropin fin array considered here
requires special care and techniques to avoid unwanted
particle sticking. A novel pH-controlled, higher seeding of
microparticles is employed to enhance signal-to-noise ratio
in lPIV and to obtain instantaneous velocity measure-
ments. The instantaneous lPIV measurements enabled us
to perform unsteady flow characterization in such a com-
plex microscale geometry as opposed to more common
ensemble averaged lPIV measurements, which are only
suited for steady state flows (Raffel et al. 2007). Detailed
velocity data were obtained at different locations in micr-
opin fin arrays, and vortex structures were analyzed in
order to explain the pressure drop trend along the flow
cavity. Furthermore, we show a position dependence of the
pressure transition along the flow direction and the absence
of vortex shedding for a cavity height equals pin diameter,
in the Reynolds number domain investigated. The geo-
metric sizes and flow rates selected to be studied are spe-
cifically suitable for integrated chip cooling. The
corresponding pressure drop in such microcavities is
essentially a necessary penalty of the chip cooling process.
Our measurements show that the vortex shedding leads to a
sharp rise in the pressure drop across the chip. However,
vortices will also lead to enhanced mixing and heat trans-
fer, thereby requiring a trade-off in the design of such
chips. Through our work, we have identified these
732 Exp Fluids (2011) 51:731–741
123
competing aspects in designing microcavities with micr-
opin fin arrays for integrated cooling of electronic chips.
2 Experimental setup
2.1 Microfluidic chip fabrication
The microfluidic chips were prepared using standard
microfabrication techniques. Figure 1a, b shows pictures
and schematics of the chips used in the study, respectively.
The cylindrical micropin fins as well as the fluid ports were
etched into silicon with deep reactive ion etching (DRIE)
followed by anodic bonding of a 500-lm-thick glass plate
to the top of the chip in order to form the cavity and still
allowing optical access for lPIV. The resulting microflu-
idic chip cavity was square shaped in top-view
(10 9 10 mm
2
) with a height of 200 ± 10 lm. The pres-
sure drop along the fluid path was probed using four
pressure ports located at 0, 1, 5, and 10 mm along the flow
direction (see Fig. 1a). The pressure ports at 0 and 10 mm
are located within the cavity, 1 mm from the inlet/outlet as
shown in the scanning electron microscope (SEM) image
in the inset of Fig. 1a. The chip cavity was connected to a
water loop using two rectangular 1-mm-wide inlet and
outlet slots.
Two different cavity designs with 50 9 50 pins in the
flow area of the chip were used as shown in Fig. 1b. The
first is a pin fin inline (PFI) and the second a staggered
distorted cell (PFS-dc) arrangement of pins; both with a pin
pitch p of 200 ± 10 lm and a pin diameter d of
100 ± 5 lm(p/d = 2). Note that for the staggered dis-
torted cell, every second column of pins is shifted by half a
pitch in the flow direction.
2.2 Measurement method and experimental conditions
The flow was driven with an impeller pump (Conrad
Electronic SE, Germany) to minimize pulsation, which
could be the source of transient effects, at a maximum flow
rate of 190 ml/min. The variation of the flow rate above
50 ml/min was less than 2%. The flow was measured with
a laminar pressure gradient flow sensor (range: 0–500 ml/
min; Omega, USA) with a full scale accuracy of 2%. A
differential pressure gauge (range: 0–2.0 bar, accuracy:
0.2%; Omega, USA) was used to measure the pressure drop
between the ports. All the pressure drop measurements
were conducted at room temperature with de-ionized (DI)
water as the working fluid. However, the pH of the water
was adjusted using sodium hydroxide for lPIV experi-
ments for the reasons described below. The pressure
measurements were recorded with a delay of a few minutes
after setting the flow rate, to ensure stabilization of flow
through the flow loop. Figure 2 shows the flow loop and
connected lPIV system used for flow visualization. The
lPIV setup consisted of an epi-fluorescent microscope
system (FlowMaster Mitas, LaVision, Goettingen, Ger-
many) as shown in the figure. The test chips were placed on
Fig. 1 a Photograph and SEM
image of the microfluidic chip
with the pressure ports indicted
(top) and a SEM image of the
micropin fins in the flow cavity
(bottom). b Design of the inline
(PFI) and staggered distorted
cell (PFS-dc) chips used for the
experimental study
Exp Fluids (2011) 51:731–741 733
123
the 3D stage of the epi-fluorescent microscope in order to
change the location of the observed (focused) area. The
flow was seeded with fluorescent buoyant particles (1 lm
in diameter, Nile red FluoSpheres; Invitrogen, Carlsbad,
CA). The tracer particles were excited with a double-pulsed
532 nm Nd:YAG laser, and the emitted light (575 nm) was
recorded with a 2,048 9 2,048 pixel CCD camera. The
energy per pulse was adjusted to approximately 10 mJ
since the fluorescent signal saturated at higher pulse ener-
gies. Achieving instantaneous velocity fields from double-
shot images requires a high seed particle density so that
sufficient information is available for two-frame (double-
shot) cross-correlation. In our measurements, in the
absence of any further medication to DI water, particles
stuck heavily onto the walls of the pins and the chip cavity.
This is a common challenge faced in lPIV experiments
(Santiago et al. 1998) and result from Derjaguin and
Landau, Verwey and Overbeek (DLVO)-type interaction,
which accounts for both electric double-layer interaction
and van der Waals attraction, between the particles flow
boundaries (in our case pin surface and cavity walls) (Perry
and Kandlikar 2008). The microspheres were charge sta-
bilized by grafting proprietary polymers with pendant
carboxylic acid groups. It is known that a convenient
means to alter DLVO interaction is to alter the pH of the
solution, which alters the double-layer interaction, thus
stabilizing the charge-stabilized particles. We found that
adjusting the water pH to *9 using sodium hydroxide
minimized the accumulation of anionic microspheres on
micropins and cavity walls. The increased content of
hydroxide ions (due to high pH) deprotonate the carboxylic
groups on the microsphere surface such that the resulting
negative charge on the particle prevents agglomerations.
This step was crucial to obtaining instantaneous velocity
measurements by cross-correlating the image pairs at any
instant.
The timing of the laser illumination and frame acquisi-
tion were controlled using the LaVision software. The PIV
image pairs (i.e., the double-shot images) were acquired
with a frequency of 4 Hz. The time difference between the
images of any pair was adjusted between 2 and 10 lsto
yield a maximum particle displacement of approximately
1/4th of the initial interrogation window size of 256 9 256
pixels. This corresponds to a particle displacement of
18.8 lm. The intensity patterns from the fluorescent par-
ticles associated with the laser pulse was recorded on
individual frames. To compute the velocity vectors, an
adaptive multi-pass cross-correlation technique was used.
Starting from an interrogation window of 256 9 256 pix-
els, the vector fields computed at every iterative pass were
used to as to adjust the window shift for the following
passes. The final interrogation windows of size 64 9 64
(18.8 9 18.8 lm) or 32 9 32 pixels (9.4 9 9.4 lm) with a
50% overlap, in both cases, were used. The resulting vector
spacing was 9.4 lm (for 64 pixel windows) or 4.7 lm (for
32 pixel windows).The size of interrogation window is
specified in the caption of all figures in Results and dis-
cussion section showing velocimetry measurements. For
steady flows, ensemble averaged cross-correlation was
performed with 68–100 image pairs depending on the flow
situation and compared with instantaneous velocity fields
from double-shot images. To reduce image random noise,
the particle images were pre-processed with a 3 9 3 low-
pass filter, and a spatial sliding minimum subtraction was
used to remove background noise.
In this study, a 109 microscope objective with a
numerical aperture of 0.3 was used for imaging providing a
depth of correlation 2z
corr
= 27.6 lm (Olsen and Adrian
2000). The number of particles within an interrogation
window of 64 9 64 pixels was chosen to be n = 10.
Therefore, the volumetric particle concentration of the
seeding particle was computed as
c
vol
¼ 100cV
particle
; ð1Þ
where V
particle
denotes the particle volume and c the
number of particles in an interrogation depth, which was
computed as c ¼ n=ð2z
corr
A
int
Þ with A
int
designating the
interrogation area. Substituting the numerical values, we
obtain c
vol
= 0.054%.
Although the particle seeding was chosen to achieve a
final interrogation window size of 64 9 64 pixels, some
flow conditions allowed an additional refinement with a
Fig. 2 Schematic of the lPIV system with the microfluidic chip
connected to a water loop
734 Exp Fluids (2011) 51:731–741
123