Microchannel Heat Transfer: Early History, Commercial Applications, and Emerging Opportunities

Abstract

In 1980, high-performance computing was becoming limited by the heat dissipated in semiconductor chips. IBM was introducing a new chip packaging technology that featured a specific thermal conductance of about 5000 W/m2 ·°C and occupied approximately 1 liter of space in order to cool 300 W. IBM was also developing a superconducting computer technology to circumvent the thermal problem posed by continued scaling of semiconductor chips. The following year, two of us (DBT and RFWP) showed theoretically and experimentally that by scaling down the dimensions of a conventional plate-fin liquid-cooled heat sink to a channel width of ∼50 μm, operating in the laminar flow regime, and integrated within the silicon chip, we could achieve in a laboratory demonstration at least a 20-fold improvement in specific thermal conductance, and more than 1000-fold greater volumetric heat removal. The reception of this advance was mixed, but what really stalled its adoption was the emergence of high-speed low-power CMOS semiconductor circuitry. Two decades later even scaled CMOS circuitry was getting too hot, and various commercialization attempts were then undertaken; some were successful, others not. New commercialization opportunities are now appearing including ones that enable society’s more efficient use of energy. A specific example of one such opportunity will be described, i.e., the use of microchannels in a novel, highly efficient regenerative heat-exchanger configuration, intended for heat-treating low-viscosity liquids for purposes such as pasteurization. Water was successfully heat-treated in continuous-flow tests of an experimental scaled-down prototype ultrahigh-temperature (UHT) pasteurizer incorporating a linear counterflow microchannel (50 μm parallel-plate channel separation) heat exchanger having an integrated electric heater at the hot end. The use of an integral electric heater permitted a unique manifold-less arrangement for reversing the flow directions at the hot end, wherein perfect local mass balance was enforced locally (i.e., between every pair of adjacent counter-flowing microchannels), eliminating a major potential source of flow maldistribution that would have otherwise reduced heat-exchanger effectiveness. Water entered the device at room temperature, steadily heated to 135°C in about 2.5 s, was maintained at 135°C for ∼2.5 s, and then cooled in ∼2.5 s, exiting at no more than 2°C above its original temperature, indicative of high heat-exchanger effectiveness. Heat leaks to ambient air required an excess of heater power, but those could be mostly eliminated in a scaled-up design and with proper attention to exterior insulation. Subsequent tests with milk flowing in heated microchannels revealed that fouling can be a severe problem (perhaps exacerbated by the long-tailed residence-time distribution characteristic of laminar flow), limiting continuous use to less than 2 hours for UHT pasteurization conditions. Conventional high-temperature short-time (HTST) milk pasteurization employs much lower peak temperatures and it is more likely that a practical microchannel system could be constructed for that application.

Full text

1 Copyright © 2011 by ASME
MICROCHANNEL HEAT TRANSFER: EARLY HISTORY, COMMERCIAL APPLICATIONS,
AND EMERGING OPPORTUNITIES
David B. Tuckerman
Tuckerman & Associates, Inc.
Lafayette, California, USA
R. Fabian W. Pease
Stanford University
Stanford, California, USA
Zihong Guo, Jenny E. Hu, Ozgur Yildirim, Geoff Deane, and Lowell Wood
Intellectual Ventures Laboratory
Bellevue, Washington, USA
ABSTRACT
In 1980, high-performance computing was becoming
limited by the heat dissipated in semiconductor chips. IBM
was introducing a new chip packaging technology that featured
a specific thermal conductance of about 5000 W/m2∙°C and
occupied approximately 1 liter of space in order to cool 300 W.
IBM was also developing a superconducting computer
technology to circumvent the thermal problem posed by
continued scaling of semiconductor chips. The following year,
two of us (DBT and RFWP) showed theoretically and
experimentally that by scaling down the dimensions of a
conventional plate-fin liquid-cooled heat sink to a channel
width of ~50 µm, operating in the laminar flow regime, and
integrated within the silicon chip, we could achieve in a
laboratory demonstration at least a 20-fold improvement in
specific thermal conductance, and more than 1000-fold greater
volumetric heat removal. The reception of this advance was
mixed, but what really stalled its adoption was the emergence
of high-speed low-power CMOS semiconductor circuitry. Two
decades later even scaled CMOS circuitry was getting too hot,
and various commercialization attempts were then undertaken;
some were successful, others not.
New commercialization opportunities are now appearing
including ones that enable society's more efficient use of
energy. A specific example of one such opportunity will be
described, i.e., the use of microchannels in a novel, highly
efficient regenerative heat-exchanger configuration, intended
for heat-treating low-viscosity liquids for purposes such as
pasteurization. Water was successfully heat-treated in
continuous-flow tests of an experimental scaled-down
prototype ultrahigh-temperature (UHT) pasteurizer
incorporating a linear counterflow microchannel (50 µm
parallel-plate channel separation) heat exchanger having an
integrated electric heater at the hot end. The use of an integral
electric heater permitted a unique manifold-less arrangement
for reversing the flow directions at the hot end, wherein perfect
local mass balance was enforced locally (i.e., between every
pair of adjacent counter-flowing microchannels), eliminating a
major potential source of flow maldistribution that would have
otherwise reduced heat-exchanger effectiveness. Water entered
the device at room temperature, steadily heated to 135°C in
about 2.5 s, was maintained at 135°C for ~2.5 s, and then
cooled in ~2.5 s, exiting at no more than 2°C above its original
temperature, indicative of high heat-exchanger effectiveness.
Heat leaks to ambient air required an excess of heater power,
but those could be mostly eliminated in a scaled-up design and
with proper attention to exterior insulation. Subsequent tests
with milk flowing in heated microchannels revealed that
fouling can be a severe problem (perhaps exacerbated by the
long-tailed residence-time distribution characteristic of laminar
flow), limiting continuous use to less than 2 hours for UHT
pasteurization conditions. Conventional high-temperature
short-time (HTST) milk pasteurization employs much lower
peak temperatures and it is more likely that a practical
microchannel system could be constructed for that application.
Proceedings of the ASME 2011 9th International Conference on
Nanochannels, Microchannels, and Minichannels
ICNMM2011
June 19-22, 2011, Edmonton, Alberta, CANADA
ICNMM2011-58308

2 Copyright © 2011 by ASME
HISTORICAL BACKGROUND
In the late 1970‟s, computer technology was well
established on a seemingly endless progression of
exponentially increasing levels of performance and
functionality for a given price point. The driver for this
phenomenon was continual advances in silicon semiconductor
technology. Specifically, “Moore‟s Law” – a heuristic that the
number of transistors on an integrated circuit (IC) chip doubled
every 18-24 months, was enabled by steady improvements in
photolithography technology (permitting the creation of ever
finer etched patterns on increasingly large silicon wafers) and
concomitant advances in thin film materials technologies.
A byproduct of Moore‟s law was a steadily increasing
consumption of electric power per unit area by the IC chips. At
the time, the prevailing circuit technologies for enterprise-class
computers (such as were being manufactured by IBM, the
industry leader) employed bipolar transistors, which are
controlled by electric currents, and which always draw current
in their „on‟ state. Thus, all other things being equal (switching
voltages V and transmission line impedances Z not varying
greatly, for fundamental physics and geometrical reasons),
power consumption per circuit was relatively stable (P~V2/Z)
and so tended to increase proportionately with the number of
transistors on a chip. To make matters worse, the desire to
maximize computer speed (clock frequency) pushed computer
designers to pack the thousands of chips comprising a 1970‟s-
era „mainframe‟ computer into ever-closer proximity, both
horizontally and vertically. Thus, not only were silicon chips
increasing their individual „areal‟ power density (Watts/m2), the
volumetric power density (Watts/m3) of the entire computer
system was on an exponentially increasing trend. As a
computer performs no „work‟ in a physicist‟s sense of the
word, and uses thermodynamically irreversible switching
circuits, its electrical energy consumption is almost
immediately converted entirely into thermal energy. For those
design engineers responsible for the interconnections and
physical configuration of the computer (collectively referred to
as „electronic packaging‟), computers were rapidly evolving
into highly concentrated heat sources. The challenge was an
unfamiliar one to the electronics industry: In contrast with the
predecessor technology of vacuum tubes, ICs were then, and
still are today, highly intolerant of elevated temperatures. This
is partly a performance concern, but most importantly a
reliability concern. Numerous performance degradation and
outright failure modes follow exponential-like (i.e., Arrhenius)
dependences on temperature. A particularly notorious and
spectacular failure mode at that time was electromigration in
thin-film aluminum metal conductors, a phenomenon driven by
high current densities (A/m2) within the ICs that were
enormous compared with traditional electrical power devices
and cables. Other reliability concerns included interdiffusion
reactions between dissimilar materials, solder fatigue, and
leakage currents in diodes and dielectrics (the latter continues
to be a concern today). A 10°C temperature rise above
nominal, even if relatively localized, could substantially reduce
the median-time-to-failure of the entire computer, in some
cases by a factor of 2. Thus, the design of comprehensive
„thermal management‟ systems to extract the heat from the
computer became a major technical concern to the packaging
engineer. Typically these systems involved a combination of
conductive and convective heat transfer methods (radiative heat
transfer being of negligible utility given the combined
constraint of low temperatures and high heat fluxes). The
working fluid for convective heat transfer was normally chilled
water for the central processing unit (CPU), although the less
demanding portions of the computer (e.g., main memory) could
frequently be air-cooled.
A peak of sophistication and complexity in thermal
management at the end of the 1970s was embodied in a
packaging scheme called the “Thermal Conduction Module”
(TCM), first introduced by IBM on its model 3081 [1]. The
TCM had the unusual challenge that a collection of logic chips
were each mounted face-down on an array of solder balls
(circuit side down, or “flip-chip”) on a highly complex
multilayer ceramic interconnection substrate (essentially a
glass-ceramic version of a printed-circuit board, in the form of
a 90-mm square tile). Flip-chip packaging technology was
almost unique to IBM at that time (it is commonplace today),
and posed special reliability concerns (low-cycle solder fatigue)
requiring careful thermal expansion matching of the ceramic
substrate to minimize post-soldering stresses. To prevent
stresses induced by power cycling of the chips, as well as the
need to minimize peak temperatures in general for reliability, a
low thermal resistance heat path from the back sides of the
chips was required (the flip-chip attachment and glass-ceramic
substrate severely limited heat transfer from the front faces of
the chips). This back-side heat path had an additional design
constraint to not impose any significant shear or tensile stresses
on the relatively soft flip-chip solder balls; this appeared to rule
out the use of any type of adhesive or metallurgical bonding
(the traditional methods for achieving good heat transfer from
ICs). IBM‟s solution was to use a structure loosely reminiscent
of an automobile engine block. A square array of cylindrical
aluminum pistons with slightly convex crowns were inserted
into close-fitting cylinders bored into a monolithic block of
aluminum. Each piston was spring-loaded against the back of a
chip. Heat would conduct from the chip to the piston crown,
spread to the piston sidewalls, and then conduct to the
aluminum block, which was in turn bonded to a copper block
containing several large water-cooled cavities. The TCM had
to be filled with helium gas to improve the chip-to-piston and
piston-to-cylinder thermal resistance, thereby permitting
maximum design heat fluxes of ~0.2 W/mm2 at the back of
each chip. The complete TCM module handled a 300 W total
heat load and occupied roughly 1 liter of space; in contrast, the
array of silicon chips generating the 300 W occupied, in total,
about 1/1000th of that volume. The heat flux of 0.2 W/mm2
was widely believed to be approaching the practical limit of
what could be managed in a high-performance computer, given
all the other packaging constraints (electrical performance,
solder ball stress levels, etc.). Phrases like „physical limits‟
were used in a highly cited paper of that era [2]. As a result,

3 Copyright © 2011 by ASME
there was serious concern that it might become impossible to
continue exploiting Moore‟s law to enable computers with
performance much beyond IBM‟s Model 3081.
Contemporaneously with the development of the 3081, and
motivated in part by thermal concerns, IBM had a major
research program to develop „Josephson junction‟ technology,
which leveraged the unique quantum-mechanical properties of
superconducting tunnel junctions at ultralow cryogenic
temperatures (4 K) to create a family of ultralow-power digital
logic gates, with the hope that a supercomputer could
ultimately be constructed. This was an extraordinarily
ambitious research program at IBM, lasting more than a
decade, which one of us (DBT) had the privilege to participate
from 1977 to 1980. At the time, laboratory Josephson circuits
were 1-2 orders of magnitude faster than silicon transistors
were capable of (30 years later, this gap has been more than
closed, owing to Moore‟s Law), and they consumed far less
power. A 1979 article by an IBM author championing
Josephson technology asserted that “heat sinks are, by
necessity, bulky and do not allow close packaging of chips.” [3]
A 1980 article by another author from the same group
described the design difficulties that a computer with a 1-ns
cycle time (i.e., a 1 GHz clock) would face: “If the millions of
high-speed transistors that make up such a machine were
packed into a few cubic inches, the heat evolved could not be
removed fast enough and the computer might well melt.” [4]
Today‟s 3-GHz laptop computers belie those assertions,
but such was the prevailing view ca. 1980. Underlying that
belief was the presumption that bipolar transistors were the
only appropriate semiconductor logic technology for high-
performance computers, owing to their high speed (a
consequence in part of their ability to drive high currents,
which of course causes the high power dissipation that was
causing such concern). Few people appreciated that an
emerging device, the insulated-gate field-effect transistor (and
in particular a specific version known as the MOS transistor),
would ultimately revolutionize computing. MOS transistors are
voltage-controlled devices, rather than current-controlled, and
their insulated gate can hold the transistor‟s state with
essentially zero current (hence zero power). Equally important,
it was possible to fabricate two „complementary‟ versions of
the MOS transistor (called nMOS and pMOS, reflecting the
fact that the transistor controlled the flow of negative or
positive electrical charges in the silicon, respectively). Logic
circuits that are built with such complementary pairs of MOS
transistors, known as CMOS circuits, can have the unique
property that they consume virtually no power in their resting
states, only drawing power when they are changing states.
Moreover this transient power consumption scales favorably as
circuitry is miniaturized: a classic paper by IBM‟s Robert
Dennard on MOSFET scaling showed that, to a first
approximation, there is no change in total power consumption
as the number of MOS transistors on a given chip area are
increased as a result of Moore‟s law miniaturization [5].
However in 1980, the speed of CMOS circuitry was much
lower than bipolar circuitry, and CMOS was thought mainly
useful for low-performance applications that required ultralow
power, such as digital watches or pocket calculators.
Dennard‟s paper did predict that MOS circuit speeds would
increase as the transistors were shrunk, but in order to rival
bipolar technology the scaling needed to be quite substantial,
more than most people thought would be feasible. Today, after
30 years of innovation and shrinking transistor dimensions,
CMOS chips power essentially all modern computers, from
laptops to supercomputers, and they perform at speeds much
greater than the bipolar technology of 1980.
HOW THE MICROCHANNEL HEAT SINK CAME TO BE
We have been asked to review how and why the first
microchannel heat sinks came to be developed in 1980. We
(DBT and RFWP, at least) did not foresee the computer
industry‟s complete transition to CMOS any better than the
IBM engineers, so we too believed the problem of heat
dissipation in silicon-based computers to be a pressing concern.
But DBT‟s personal experience in the Josephson project was
troubling: although ultralow power, it appeared implausible
that Josephson technology could attain the level of complexity
and reliability required to build commercial supercomputers,
for a variety of reasons. (IBM ultimately cancelled the program
in 1983). In the spring of 1980, DBT had the good fortune to
join RFWP‟s nascent research group at Stanford University as a
doctoral candidate under a Fannie and John Hertz Foundation
Fellowship. The Hertz Fellowship turned out to be a critical
catalyst to our subsequent work, for two reasons. First, DBT‟s
stipend was not tied to a particular funded research project. So
we were free to brainstorm and explore different ideas for good
PhD thesis topics, at least for theoretical work, without being
worried that we might be directing government or industrial
R&D monies for a purpose other than that which the funding
entity had intended. Second, DBT had been introduced to
Lowell L. Wood (LLW is a co-author on this paper), through
LLW‟s role as a Hertz Foundation interviewer and Director.
LLW was also head of the „Special Studies Division‟ of
Lawrence Livermore National Laboratory (LLNL). In the late
1970s, he ran a project known as the „S-1 Computer‟, wherein a
very small coterie of brilliant engineers, using internally
developed design automation tools and entirely off-the-shelf
components, designed and built (in record time) a
supercomputer for the U.S. Navy that was comparable in
performance to IBM‟s supercomputers, on a budget that was
orders of magnitude less than what anyone in the computer
industry thought possible. LLW keenly desired to push the S-1
to even higher levels of performance by designing custom
silicon integrated circuits. DBT thought this was an exciting
prospect and was inclined to help, but knew that this would
lead to the thermal problems that IBM had been warning about
in its justification for the Josephson program.
The Stanford EE department in 1980 was (and still is) a
place of great creativity and innovation in semiconductor
technology, as one might expect from its location in Palo Alto,
the birthplace of “Silicon Valley”. Professors such as James
Plummer (now Dean of Engineering) and Richard Swanson

4 Copyright © 2011 by ASME
(who later founded Sunpower, which eventually grew to
become a major manufacturer of high-performance solar cells)
were advancing the technology of semiconductor devices in a
direction that would ultimately eclipse the performance of
Josephson devices. Prof. James Meindl (now director of the
Georgia Tech Microelectronics Research Center) was directing
the Stanford Electronics Laboratories, which provided access to
world-class semiconductor R&D facilities. Prof. James Angell
(sadly, now deceased) led a research group pioneering what is
now known as MEMS (Micro Electromechanical Systems)
technology. RFWP, an early pioneer in electron-beam
lithography, had recently been recruited from AT&T Bell
Laboratories to build a research group that would advance the
state of the art in microstructure fabrication. “I‟m interested in
making things small”, was his guidance to DBT at the start of
the summer. “Figure out a thesis topic that is in some way
related to that”. DBT, having no prior knowledge of heat
transfer beyond Fourier‟s law, responded: “Maybe we can use
microstructures to improve upon IBM‟s TCM and thereby help
Lowell Wood build the world‟s fastest supercomputer?”
Things proceeded very quickly from that point onward. It
was clear that thermal interfaces and long heat conductive paths
had been major contributors to thermal resistance in the TCM;
what if we could eliminate all such interfaces and long paths?
Ideally this would mean direct contact between the silicon chip
and the coolant fluid. IBM had in fact experimented with direct
immersion of flip-chip ICs in an inert fluid, with cooling
enhanced by „nucleate boiling‟, however the maximum heat
flux before thermal runaway was approximately 0.2 W/mm2,
similar to the TCM; a forced convection approach seemed more
promising to us. While the front (device) face of the silicon
chip is physically closest to the heat-generating transistors
(typically within 10 µm), we found that the various insulating
and protective dielectric layers atop the transistors (silicon
dioxide and polyimide, most commonly) were far poorer
thermal conductors than the silicon substrate itself, by 2-3
orders of magnitude. A mere 1-µm layer of protective polymer,
for example, could pose as much thermal resistance as a full
700 µm thickness of silicon chip! The obvious conclusion was
that a high-performance cooling scheme should access the back
side of the chip, rather than the front. Back-side cooling has
the additional benefit that one could consider having water in
direct contact with the chip, without any risk of corrosion,
electrical conduction, or contamination (silicon is inert and
impermeable to water at all operating temperatures of interest).
In contrast, any front-side contact or direct immersion scheme
would require costly, thermally inferior (lower thermal
conductivity and lower heat capacity) coolants such as
fluorocarbons. For a nonmetal, crystalline silicon‟s thermal
conductivity of ~140 W/m∙K, about 1/3 that of pure copper, is
quite good. A 0.2 W/mm2 heat flux conducting through a
standard 0.7 mm wafer thickness would induce a temperature
rise of only 1 K. Clearly, heat conduction to the back of the
silicon was not going to be a serious concern. This led us to
focus our attention on achieving the most efficient heat transfer
from the back of the silicon to a water coolant. We easily
calculated that a modest flow of only 10 ml/s could carry away
over 40 Watts of heat from a chip with only a 1 K average
coolant temperature rise. So the ability to supply enough water
seemed not by itself a major limitation.
It was now clear that the real heat-transfer bottleneck was
between the silicon and the coolant. Even with no education in
convective heat transfer, it seemed obvious that enhancing the
surface area of the silicon should be helpful, e.g., by creating
finned structures. Parallel-plate fins and channels seemed
simplest to analyze, and we were aware that Prof. Angell‟s
MEMS laboratory had been working with new “orientation-
dependent” chemical etchants that Ken Bean [6] and Don
Kendall [7] had shown could enable fabrication of rectangular
microchannels in silicon wafers of a specific crystal orientation
(<110>). Aspect ratios as high as 600:1 had been created with
this technique, so we would have almost complete freedom to
choose the fin dimensions for optimal performance.
It seemed intuitive that using very narrow channels could
provide superior heat transfer, subject to the limitation that one
needed sufficient coolant flow to limit the temperature rise. To
proceed further, we began educating ourselves about
convective heat transfer. We quickly learned that three giants
of the field, A. Louis London, William M. Kays, and Robert J.
Moffat were right next door in Stanford‟s Mechanical
Engineering department! (Bill Kays was also Dean of
Engineering at that time). Kays‟ classic text “Convective Heat
and Mass Transfer”, along with Kays and London‟s classic
reference “Compact Heat Exchangers”, became our bibles for
the next few weeks. It was clear that a plate-fin microchannel
structure would drive us into what is traditionally thought of as
a worst-case heat-transfer scenario: laminar flow, possibly
with a fully-developed thermal boundary layer. However, even
with worst-case assumptions, the inverse relationship between
heat transfer coefficient and channel width was striking and
suggested that high performance could still be achieved with
sufficiently narrow channels (e.g., ~50 µm). Optimizing for a
practical water pressure led to a design that, using
microchannels etched entirely within the confines of a standard
chip thickness, was predicted to outperform the TCM by 1-2
orders of magnitude in heat flux, and 3 orders of magnitude
volumetrically (Figure 1). By the end of the summer we had a
plan to fabricate and test our optimized silicon heat sink.
Figure 1 – The original silicon microchannel heat sink
structure, as published by DBT and RFWP in 1981 [8].

5 Copyright © 2011 by ASME
Kays, London, and Moffat reviewed our ideas and became
enthusiastic supporters and advisors throughout DBT‟s doctoral
program. We were well aware that our heat-transfer
calculations would seem crude and elementary from their
perspective, yet they always treated us as serious contributors
to the state of the art. They were particularly intrigued and
excited to learn that such precision microfabrication techniques
existed, offering the prospect of scaling heat-exchanger
technology down to ultra-compact dimensions. Nevertheless,
securing funding to build the prototype was challenging, as it
did not fit the goals of any existing funded research programs.
We were able to borrow equipment and use some laboratory
facilities gratis through the courtesy of various colleagues, but
certain items such as photomasks required funding. We did
secure some modest initial funding from the military and also a
small grant from Honeywell, but the experimental program was
otherwise unfunded until 1983.
By mid-October we had fabricated satisfactory silicon
finned structures, and integrated a heat source (a vapor-
deposited thin-film resistor) to simulate a functioning IC. But
we had not converged on a workable technique for delivering
and confining the coolant to the microchannels. We finally
adopted a simple electrostatic glass-metal bonding technique
(„Mallory bonding‟) in use at Prof. Angell‟s MEMS laboratory,
as this process gave an inspectable bond that could be reliably
hermetic (Figure 2). On November 11, 1980, we made our first
successful demonstration, feeding almost 200 Watts of
electrical power to a 10-mm square area, with a maximum
temperature rise of only 20°C. We spent the next several
months refining our fabrication processes and experimental
techniques, testing the devices at successively higher power
levels, and collecting publication-quality data. By that time,
word of our results had spread, and we were encouraged by
George E. Smith of Bell Laboratories (now a Nobel Laureate
for his invention of the charge-coupled device) to submit them
to his new fast-publication journal, IEEE Electron Device
Letters. In view of the likely audience for this journal
(primarily electronic engineers in the semiconductor industry),
we wrote what we hoped would be a very „reader friendly‟
article that assumed no prior background in heat transfer.
Figure 2 – Cross section (from [12]) of the first
microchannel heat sink. The silicon fins (dark areas)
have a 100µm pitch and are hermetically bonded to a
glass cover plate. The ~50-µm wide microchannels were
formed by anisotropic etching of <110> silicon.
Our paper “High-Performance Heat Sinking for VLSI” [8]
appeared in May 1981 and attracted a gratifyingly strong
amount of interest from the academic community and from
industrial research laboratories such as IBM. Not long after,
George Smith told us that the IEEE had created a new annual
award (the „IEEE Paul Rappaport Award‟) for the best paper in
an Electron Device Society journal, and that we had been
selected as the first recipients. The paper has since received
over 1000 direct citations in the technical literature. After years
of being told by reputable sources that thermal concerns would
soon pose a „fundamental limit‟ to Moore‟s law, semiconductor
and computer professionals now understood that this wasn‟t
true. Given the overwhelmingly positive reception that our
work had received, we felt that there would be no problem
getting funding to continue building on this work. We soon
discovered how wrong we were.
We had originally received some support from the US
military‟s „Joint Services Electronics Program‟, which by
coincidence conducted an annual review of its various program
at Stanford only two weeks after our first experimental success.
Although there had previously been a few grumbles that this
kind of work sounded „too applied‟, we felt that the spectacular
results we had just achieved would more than justify their
modest investment. Instead we were expelled from the
program because our work was „not science‟. A few weeks
later RFWP recounted this to the President of the Royal Society
(surely „Mr. Science‟ himself) who put his hand to his brow
and expressed how sorry he was to hear that “that kind of
thinking is pervasive in the United States as well as in the
United Kingdom”. Even after our paper received the IEEE
Paul Rappaport Award, the military reviewers of our proposals
were vitriolic (and irrational) in their criticism. However in
1982 Larry Sumney, who had directed the US Department of
Defense‟s Very High Speed Integrated Circuits Program,
became president of the Semiconductor Research Corporation,
funded by the US semiconductor industry. Having heard of our
work, he encouraged us to submit a proposal which led to a
succession of contracts on thermal management.
We have speculated as to why the reaction of the military
reviewers (and a few civilian ones) was so hostile; we believe
some perceived a threat to the justification for their pet
programs (notably the use of exotic semiconductors which, as
with IBM‟s superconducting computer program, promised a
lower power/performance ratio than silicon). In the book
„Longitude‟ Dava Sobel describes how some senior scientists
(notably the ageing Isaac Newton, but not the Royal Society
itself) stalled awarding the British Admiralty‟s prize money for
the chronometer on the grounds that the solution should have
been an astronomical one. So it‟s not a new phenomenon;
maybe every significant advance poses a threat to some in
powerful places and this should be taken into account before
any public announcement.
The remainder of our work, during 1982-1983, put priority
on demonstrating variations of microchannel cooling that we
hoped would have more commercial utility. This work was
documented in several subsequent conference papers [9-11],

6 Copyright © 2011 by ASME
and more thoroughly in DBT‟s PhD thesis [12]. It was clear to
us that the single greatest impediment to commercial adoption
of microchannel cooling techniques was our configuration in
which the coolant structures were integrated directly into the
silicon IC substrate. Eliminating all non-essential thermal
interfaces had enabled some impressive laboratory
demonstrations, as high as 1309 W from a 10-mm square area
with a maximum temperature rise of 108 K. But integrating the
microchannels was deemed commercially impractical for a
number of compelling reasons. It would be far better to
construct liquid-cooled microchannel heat sinks as separate
units (so-called „cold plates‟), and then attach the IC chips
(thinned, if necessary) to them. Ideally one would construct the
cold plates from a low-cost material which has higher thermal
conductivity than silicon, and much greater fracture toughness,
as fracture of the silicon heat sinks was a major worry in our
original work. Copper, for example, clearly meets these
requirements, and subsequent work by others has exploited
copper microchannels very successfully, as discussed later.
The use of microchannel cold plates necessarily directs
one‟s attention to the creation and maintenance of a reliable,
low-stress, ultrahigh performance thermal interface
between the heat source and the heat sink. This is a highly
nontrivial technological challenge, one that has until recently
not received as much attention from the technical community
as it deserves. We believe there is great opportunity for
innovative breakthroughs in this area. In some cases, a high
quality solder or brazing “die attach” process will do the job.
But there are many situations where it will not. For example,
the bonding method might impose undue stresses which impact
device performance or reliability (a particularly important issue
with laser diodes). It might induce fatigue failures as a result of
thermal or power cycling (particularly if there is a mismatch in
thermal expansion properties between the heat source and sink).
It might have internal voids. Often the design requires that the
interface be easily separable and reworkable. In some cases it
is difficult to achieve or maintain the necessary coplanarity
between the source and sink, particularly for larger ICs. These
concerns have increased the importance of developing better
“thermal interface materials” (often abbreviated as TIMs). A
good TIM provides a high thermal conductance path between
the heat source and heat sink surfaces, requires low (preferably
zero) external compressive loading, can tolerate some non-
planarity of each surface without developing voids or inducing
hot spots, and exerts low (again, preferably zero) shear forces
on the source and sink.
In our work from 1981 to 1983, we made an attempt to
design an ideal TIM using what we dubbed a “void-free micro-
capillary thermal interface” (Figure 3). This involved the use
of a thin liquid interfacial layer (a high-vacuum pump oil, Dow
Corning 704), wherein re-entrant micron-scale grooves were
formed in one of the mating surfaces. These microstructures
had several roles: they imparted a permanent hydrostatic
tension in the liquid (owing to capillary forces) which held the
mating surfaces together (effectively acting like a vacuum
chuck, but without requiring an external source of vacuum),
they collectively acted as a reservoir for the liquid, and their re-
entrant shape ensured that the interface was always and
everywhere wetted (i.e., an inherently void-free joint). We
successfully demonstrated very low (i.e., lower than the
microchannel heat sink itself) thermal resistance with this
technique. However, there were some limitations to our
specific implementation, e.g., fabrication challenges associated
with achieving the correct micro-capillary shapes, hot spots
from small entrapped particles, tight coplanarity requirements,
and compatibility concerns with commercial „flip-chip‟
packaging methods. These limitations might have been
overcome had we taken the work further than we did (and been
better funded). But by late 1983 our industry sponsors had
begun the transition to low-power CMOS circuitry and wanted
us to work on other problems. Nonetheless, if we could have
found a metallurgically stable non-hazardous liquid-metal alloy
with good wetting characteristics that would have worked in
place of silicone oil, the microcapillary thermal interface would
have been a much more compelling technology, owing to the
combination of much greater surface tension and much higher
thermal conductivity. Liquid-metal alloys (often applied as a
solid foil at room temperature, but which liquefy at the device
operating temperature, so-called “phase-change” TIMs) are in
fact now being used in as thermal interface materials in
commercial practice [13], and it would be interesting to revisit
the feasibility of the microcapillary concept in conjunction with
the variety of metallic TIM alloys that are now available.
Figure 3 – Microcapillary thermal interface concept (from
[11]) for providing a low-stress high-conductance liquid
thermal interface between planar surfaces. Micron-scale
capillaries create a negative pressure (hydrostatic
tension) in the liquid which holds the surfaces together.
Alternatively, there have been recent attempts to create
TIMs using forests of vertically aligned single-wall carbon
nanotubes, which are known to have outstanding axial thermal
conductivity. The general idea is that, provided the forest is not
too dense, the nanotubes can individually buckle as needed to
compensate for coplanarity variations between the surfaces.
Moreover the forest will not transfer shear stresses, permitting

7 Copyright © 2011 by ASME
unrestricted differential expansion of the opposing surfaces.
Some very encouraging results have recently been reported [14]
although they still require some amount of applied compressive
force (order of magnitude ~1 atmosphere) to achieve good
results, and the thermal resistance is still 1-2 orders of
magnitude greater than what is thought to be theoretically
possible. This discrepancy is believed to be due to thermal
contact resistance between the ends of the nanotube and the
contacting surface. It may be possible to reduce this contact
resistance, as well as eliminate the need to apply a continuous
compressive force, using techniques such as developed by Prof.
Liming Dai‟s group at U. Dayton [15], wherein the nanotubes
were modified to effectively make „lap joint‟ contacts
(providing strong stiction to the surface via van der Waals
forces, which is now believed to be the mechanism by which
geckos can climb smooth vertical surfaces), rather than „butt
joint‟ contacts. Several other research groups are pursuing
similar ideas (e.g., Ken Goodson‟s Microscale Heat Transfer
Group at Stanford [16], Arun Majumdar‟s group at UC
Berkeley [17], and Pulickel Ajayan‟s group at Rice Univ. [18]).
Prof. Goodson‟s group recently announced a “CNT nanotape”
having thermal conductance near 2 mm2∙°C/W and mechanical
compliance, measured using a MEMS resonator, below 20 MPa
[19]. These developments are very encouraging, and we look
forward to the possibility that they may lead to commercially
viable products.
PRACTICAL IMPLEMENTATIONS
As discussed previously, the advent of high-performance
CMOS integrated circuits ultimately obviated any near-term
need for extreme cooling measures such as we had proposed.
The earliest adopters of microchannel technology actually came
from the emerging field of semiconductor lasers. These
devices generate high heat fluxes, and their performance,
lifetime, and wavelength stability were extremely sensitive to,
and adversely affected by, high temperatures. They were often
grouped into closely packed arrays, exacerbating the cooling
challenge. In the mid-1980‟s, one of us (DBT) collaborated
with a group at Lawrence Livermore National Laboratory
(LLNL) to apply microchannel cooling to laser diode arrays
[20]. That group made significant design and performance
improvements, including developing metallization schemes for
soldering the laser diode bars to the silicon heat sinks, leading
to some highly successful implementations for internal LLNL
programs. The group later helped a newly formed company
named SiMMtec (an acronym for “Silicon Monolithic
Microchannel”) commercialize the technology. SiMMtec sold
high-power laser diode arrays into various defense-related
programs, and was in business during most of the last decade
(2001-2010) [21].
Nowadays most microchannel coolers for lasers are
constructed out of copper, as it leads to a more robust design
and higher performance. In some cases, a lower expansion
alloy such as Cu-W or Cu-Mo is used, to achieve a much closer
thermal expansion match to the laser diode, which is itself a III-
V semiconductor material such as GaAs. For example,
Spectra-Mat, Inc., of Watsonville CA manufactures a line of
expansion-controlled microchannel coolers for the laser diode
market. The High Power Diode Laser Technology Conference
at Photonics West has, for many years, focused on
technological improvements in high power diode lasers in
conjunction with microchannel coolers. There are reportedly
many commercially successful laser product lines that use
microchannel coolers, e.g., Nuvonyx, DILAS, LaserLine, Rofin
Sinar, Spectra Physics, Cutting Edge Optronics (CEO), and
Jena Optik. Perhaps the largest arrays ever built with
microchannel cooling were made by Nuvonyx for Textron
under the JHPSL program (a 50 kW laser system, generating
~72 kW of heat) and HELLADS program (75 kW optical
power, ~108 kW of heat) [22].
While the laser industry has been the primary commercial
user to date, our original vision that microchannel cooling
would penetrate the computer industry has been realized to
some degree, especially in recent years. Cooligy, a company
founded by three Stanford professors (Ken Goodson, Juan
Santiago and Tom Kenny) developed a liquid cooling loop
containing a 3-dimensional micro-fluidic network that was used
by Apple to cool the CPU in its Power Mac G5 desktop system.
Cooligy was subsequently acquired by Emerson Corp. in 2005
[23]. More recently, Dell Computer and Acer both sell gaming
PCs that use a cooling loop containing a copper microchannel
cold plate (with integral pump) manufactured by Asetek.
Asetek has reported shipping over 500,000 such systems.
IBM, originally a user of water cooling beginning in 1964
for all its high-end mainframe computers, abandoned water
cooling in 1992 upon transitioning to CMOS semiconductor
technology. However, the power consumption of CMOS has
been steadily increasing over the last two decades, with the
result that multichip module heat fluxes are now exceeding the
highest values previously achieved in IBM‟s most advanced
bipolar mainframe (model ES9000, launched in the late
1980‟s). In 2008 IBM announced the resumption of water
cooling in its high-performance computing platform, the Power
575 Supercomputing node/system [24]. Each 158-Watt
processor module is in spring-loaded contact (~1kN force),
through a pair of thermal interface layers, with its own water-
cooled copper cold plate, containing 42 high-aspect ratio
channels (0.5 mm x 8 mm, on a 1.15 mm pitch) that were cut
using a precision mechanical circular saw. While not
technically microchannels (per Kandlikar‟s criterion [25] of
hydraulic diameters less than ~200 µm), the flow was solidly in
the laminar regime (Re=223). The non-CPU parts of the
system (about half the heat load) remained air-cooled. In the
more recent Power 775 system, however, almost the entire
system is cooled using similar cold-plate technology, designed
to handle a 270 kW total heat load [26]. The ambient room air
plays no role in the cooling of the machine, hence there is no
requirement for room air conditioning, a significant energy
savings (suggesting that microchannel cooling can play a role
as a “green” technology). The use of water cooling enabled a
34% increase in processor speed and a 20°C reduction in
temperature compared with what could be accomplished with

8 Copyright © 2011 by ASME
air cooling. The recently announced Blue Gene®/Q
supercomputers also use water-cooled cold plates [27].
Unlike the transition from bipolar ICs to CMOS, there
does not appear to be any disruptive technology on the horizon
that will again reduce computing power requirements by an
order of magnitude (though it is admittedly dangerous to make
such statements, as recent research in superconducting devices
and „quantum computing‟ offer at least a theoretical hope of
greatly improved performance at very low power levels [28]).
We expect that heat fluxes in the highest performance
computers will continue their upward trend for the foreseeable
future. As a result, we expect increasing adoption of
microchannel-type cooling methods in high-end machines,
most likely of the copper cold plate variety that is already in
use by IBM. In view of the now-extensive literature on
microchannel cooling (see for example Kandlikar‟s excellent
review paper [25]), which covers many improvements in
performance, modeling, and implementation, we have no
worries that cold-plate performance will pose a limitation to
computer performance. (To illustrate how large a heat-transfer
coefficient can be achieved, Goodson and Pease described a
microchannel configuration employing diamond walls [29].)
Instead, the thermal bottleneck will be the thermal interface to
the cold plate. We believe that there is tremendous opportunity
for improvement in TIMs, and we suggest that such
improvements would likely offer more near-term value to the
computer industry than further enhancements and refinements
in microchannel cooling.
Microchannel heat transfer structures have other interesting
applications beyond simple cold-plate convective cooling of
semiconductor devices. There are many industrial situations
where one would desire ultra-compact heat exchangers to
transfer heat between different fluids, sometimes integrated
with other micro-fluidic functions to achieve chemical
reactions. The very short thermal time constants of
microchannel structures (resulting from the combination of
very high heat transfer coefficient and very low heat capacity),
coupled with very short diffusion times (high mass transfer
coefficients) can allow acceleration of various physical and
chemical process flows.
One example of exploiting the short thermal time constant
(about 10 ms) of the microchannel heat sink coupled with the
high thermal conductance compliant interface is now being
researched in RFWP‟s laboratory. That is to use locally
controlled expansion to assure accurate overlay in large area
(e.g., 300 mm diameter) nano-printing. In this technique, a
quartz template with a relief image of the required pattern is
pressed into a liquid polymer on top of the wafer, which sits on
a microchannel heat sink chuck. The pattern on the template is
accurately positioned over the existing structure on the wafer,
then a laser pulse crosslinks the polymer into a solid resist film.
With appropriate chemistry, the quartz template can be
removed without damaging the polymeric resist pattern.
Features finer than 10 nm can be patterned in this manner.
Achieving accurate overlay over the entire pattern requires that
local distortion of template and wafer match. By using an array
of light microscopes to monitor local overlay errors and feeding
back to local heating elements within the microchannel heat
sink chuck, we can rapidly correct these errors [30].
Companies such as Velocys, Inc. have demonstrated
microchannel heat exchangers for applications such as
liquefaction of natural gas (LNG), steam methane reforming for
production of hydrogen (e.g., in fuel cells), Fischer-Tropsch
synthesis of hydrocarbons, and distillation columns. Brandner
et al [31] at the Karlsruhe Institute of Technology has
constructed a water-water crossflow microchannel heat
exchanger having 1 MW of heat duty, in a structure about the
size of a drinking glass. Microchannel heat transfer structures
are incorporated into commercial air conditioning and
refrigeration systems by corporations such as Delphi and
Danfoss. The remainder of this paper will describe our recent
explorations at Intellectual Ventures Laboratory of one new
potential application for microchannels, to perform thermal
cycling of liquids (in particular, for pasteurization) with very
high recuperation of the heat.
DESIGN CONCEPT FOR A NOVEL MICROCHANNEL
ULTRAHIGH TEMPERATURE PASTEURIZER
In late 2008, discussions with the Bill and Melinda Gates
Foundation encouraged us to explore low-cost, small-scale
pasteurization solutions for economically disadvantaged
populations, e.g., subsistence farmers without access to
refrigeration. One idea was to create an inexpensive, highly
energy-efficient “personal ultra-pasteurizer” which might
enable an individual family to render fresh cow‟s milk aseptic,
giving it a long non-refrigerated shelf life if properly packaged.
Conventional high-temperature/short-time (HTST) milk
pasteurization requires heating milk to at least 72°C for at least
15-20 seconds [32]. Ultrahigh temperature (UHT)
pasteurization requires a much higher temperature (typically
135°C), but the exposure time can be much shorter (1-2
seconds), and the process kills all spoilage-inducing and
disease-causing organisms and their spores, some of which can
survive conventional HTST pasteurization.
Pasteurization of liquids (water, milk, beer, juices, etc.)
can, in theory, be performed as a continuous-flow process that
consumes almost no net energy in steady-state operation,
provided one uses a sufficiently efficient heat exchanger (HX)
to recover heat from the outgoing hot liquid and transfer it to
the incoming cold liquid. We have seen claims as high as 95%
heat recovery for commercial HTST “flash” pasteurizers [33,
34] and 85% recovery for UHT pasteurizers [35].
There has been some prior work [36-38] on small-scale
solar-heated HTST-type pasteurizers (intended for treating
drinking water, rather than milk), however many of them do not
recover any of the heat, and those that do have not
demonstrated high HX effectiveness (below 80%). Achieving
a very high effectiveness is desirable because, for a fixed
supply of heat (whether from the sun, a battery, or a flame), the
maximum possible throughput of the system will be inversely
proportional to the HX ineffectiveness; for example, if a
pasteurizer could recover 99% of the heat, its throughput

9 Copyright © 2011 by ASME
potential is 20 times greater than one which recovers only 80%.
The solar pasteurizers also have the obvious limitation that they
only function when direct sunlight is accessible.
Our challenge was to create a miniature (ideally hand-held)
UHT pasteurizer that could process 1 ml/s (~1 gallon/hour) of
milk with at least 99% heat recovery. This would imply a
power consumption of less than 5 W, which could easily be
supplied by a modern cell phone battery. (Presuming the
unavailability of electric power distribution, we anticipated the
battery would later be recharged either from a small solar panel
or by a human-powered generator). Specialized counterflow
heat exchangers have been reported in the literature with
effectiveness as high as 99.8% [39], so there is no fundamental
reason to think that our goal is impossible; the issue is how to
achieve the goal with a low-cost, potentially mass-producible
configuration that meets the various design constraints. As far
as we are aware, existing commercial pasteurization systems
use the heat-recovery HX in a conventional manner, i.e., as a
discrete 4-port component, physically separated from the
source of heat. As we shall explain, integration of the heat
source with the HX offers means to create a 2-port device
having higher effectiveness.
(a)
(b)
(c)
Figure 4 – Initial UHT pasteurizer design concept,
showing (a) overall configuration of the proposed heat
exchanger stack, (b) flow detail of a single balanced
counterflow microchannel pair (one of 100 “µchannel
pairs” within the HX stack, and (c) approximate shape of
steady-state temperature profiles within a µchannel pair.
The basic idea is simple (Figure 4) [40]: construct a high-
effectiveness linear (i.e., not folded) parallel-plate (actually a
high-aspect ratio rectangular cross section, with an array of
internal supports to control channel dimensions and prevent
collapse) counterflow microchannel HX having an integrated
electric heater near one end (the “hot end”). The heater is
coupled via high-thermal-conductivity paths along each
microchannel wall to create a nearly isothermal region (the “hot
zone”) that, using a temperature sensor and feedback loop, can
maintain the milk at 135°C over an axial distance corresponding
to an average of 2.5 seconds (i.e., 50 mm @ 20 mm/s mean
velocity). The microchannels provide maximal heat transfer in
a minimal volume. The stacked parallel-plate construction,
wherein the axial flow direction of each layer is opposite to that
of its nearest neighbors above and below, offers the prospect of
a simple, low-cost assembly process that leverages the mass-
production techniques of the printed circuit board industry
(e.g., simultaneous lamination of multiple layers). It also
simplifies the problem of creating inlet and outlet manifolds at
the ends of the microchannel bundles, and makes scaling of the
design to higher flow rates straightforward (i.e., by adding
layers to make the stack taller, and/or increasing the width of
the parallel-plate channels (i.e., making the stack wider).
A key feature of our pasteurizer is that there are no
manifolds at the hot end of the HX. Instead, every
microchannel containing the incoming (heating) flow has a
single internal connection at the hot end to the microchannel
directly below it (e.g., a slot in the channel wall between the
layers), which provides the return path for the outgoing
(cooling, pasteurized) flow (Figure 4b). In other words, we
have a stack of what we will call „µchannel pairs‟, where each
µchannel pair functions as an individual counterflow
„subexchanger‟, having perfect mass balance (and hence perfect
heat capacity rate balance, since we are dealing with a single
fluid) between the two directions. It is well known that a
nonuniform distribution („maldistribution‟) of flow amongst the
individual channels of a HX can create significant
inefficiencies [41], even when the two flows are balanced in
aggregate. Maldistributions can result from the design of the
end manifolds, and/or from manufacturing tolerances that
produce variations in the hydraulic diameters of different
channels. For example, in his PhD thesis describing a novel
polymer HX intended for solar pasteurization, Denkenberger
[38] discussed in detail how maldistributions in fluid flow had a
significant adverse impact on the experimentally measured
effectiveness of his prototype devices.
With microchannel geometries, it can be especially
difficult to hold the very tight percentage tolerances that would
normally be required to achieve a very high HX effectiveness.
However, we believe that, to first order, the presence of perfect
mass balance within each µchannel pair, along with the global
mass balance, eliminates the main sources of maldistribution-
induced ineffectiveness. To see this (Figure 4c), consider that
each µchannel pair would, if thermally isolated from the pairs
above and below it, exhibit an axial temperature profile that
rises linearly, starting at the cold end (nominally 20°C) and

10 Copyright © 2011 by ASME
leveling off upon reaching the start of the hot zone (nominally
held at 135°C by the electric heater). There would, however, be
a slight near-constant offset ∆T between the two flow
directions, wherein the outbound (pasteurized) mixed-mean
fluid temperature is hotter than the inbound (raw) fluid
temperature at each axial location along the ramp by
approximately ∆T=(1-ε)(135-20), where ε is the effectiveness
of the counterflow µchannel pair. A typical value for ε in our
architecture could be 99.5%, provided the exterior channel
walls are perfectly insulated from the ambient, and axial
conduction is negligible; in this case the outbound flow
temperature profile would be consistently ~0.6°C greater than
the inbound flow. If we now bring two such µchannel pairs
into thermal contact (i.e., fuse the bottom wall of the upper
µchannel pair to the top wall of the lower µchannel pair), some
heat transfer will occur through that fused wall (i.e., from the
outbound flow of one µchannel pair to the inbound flow of the
adjacent µchannel pair). However, because the two µchannel
pairs had identical internal temperature profiles in isolation,
their cross-sectional heat content would also be the same at
corresponding axial locations, so there should be no net heat
transfer from one µchannel pair to the other, hence no driver for
additional entropy generation. On the contrary, the rate of heat
transfer from the hotter flows to the cooler flows will be the
same as before, but the additional pathway for heat flow
actually diminishes total entropy generation by reducing
internal temperature gradients. HX effectiveness for the
conjoined µchannel pairs will therefore be greater than for the
individual pairs (perhaps increasing to 99.6% from the original
99.5% in our example). Extending this idea to a large vertical
stack of µchannel pairs (our nominal design had 100 such
pairs), we asymptotically approach periodic boundary
conditions, in which case the fluid temperature profile within
each channel will have a cross-sectional profile somewhat (but
not exactly) resembling a parabola, and we can use the
published value of the Nusselt number for parallel plates with
fully developed velocity and temperature profiles and constant-
heat-rate boundary condition, i.e., Nu=8.235, [42] as an initial
approximation for estimating the internal heat transfer of our
system. It is interesting to compare this with the case where
one of the parallel plates is insulated (adiabatic), wherein
Nu=5.385; this is the relevant figure to use when analyzing an
isolated µchannel pair. So, the ratio of 8.235/5.385=1.53 gives
us an approximate idea of the increase (~50%) in Number of
Transfer Units (NTUs) that is achieved by laminating many
µchannel pairs in thermal contact. Taking our previous
example of ε=99.5% for a single µchannel pair, we would
estimate ε=99.7% effectiveness for a large (e.g., 100-layer)
stack of pairs.
In practice, notwithstanding the perfect mass balance
locally (i.e., within each individual counterflow pair) and
globally (i.e., total input = total output), there will inevitably be
variations in the round-trip mass flow from one µchannel pair
to another. We believe this does not, within reason, seriously
impair the HX effectiveness. That this is the case may not be
immediately apparent, but it results from a similar argument to
that used in the previous paragraph. Consider two µchannel
pairs having matched temperatures at the cold end and hot
ends, but with two different mass flows, due to maldistribution.
Each µchannel pair is individually balanced, and will have
virtually identical heat transfer coefficients (fully developed
laminar flow; thermal entrance lengths are <1% of channel
lengths in our design regime and can be neglected) and so
would, if thermally isolated, equilibrate to the same linear axial
temperature profile, except for a difference in the offset
temperature ∆T between the inbound and outbound flows.
Recall our example of a nominally 99.5% effective (i.e., 0.5%
ineffective) µchannel pair; using the terminology of the NTU
method, this is a balanced counterflow HX with NTU~200. If a
particular µchannel pair had 50% more flow than nominal, and
an adjacent µchannel pair had 50% less flow than nominal (a
particularly extreme maldistribution), the µchannel pairs would,
if thermally isolated, each equilibrate to similar linear
temperature ramps, with the same start and endpoints (20°C and
135°C respectively), excepting that the faster pair will have
NTU~200/1.5, hence 0.75% ineffectiveness, and the slower
pair will have NTU~200/0.5, hence 0.25% ineffectiveness. The
faster pair will therefore have a ~0.9°C temperature offset
(compared with a nominal 0.6°C) between the hotter outbound
flow and the cooler inbound flow; the slower pair will
conversely have a ~0.3°C offset. The difference between the
cross-sectional average temperature within one µchannel pair
and that of the other µchannel pair will initially be about 0.3°C.
If the two µchannel pairs are now stacked in close thermal
proximity (i.e., sharing a common wall), there will be some net
heat transfer from the faster (hotter) µchannel pair to the cooler
(slower) µchannel pair, but given the small average temperature
differential, the added entropy generation should be of roughly
the same magnitude as was already occurring within the slow
layer, and certainly less than was originally occurring in the
fast layer. So we think it likely, but have not rigorously
proved, that there is a general theorem wherein the
effectiveness of the entire HX could never be worse than the
least effective individual µchannel pair (which would generally
be the one having the highest flow rate); this is obviously true if
the µchannel pairs are thermally isolated from each other, but is
less obvious when they are laminated together in intimate
thermal contact, where there could be large flow rate
mismatches between neighboring pairs. Such mismatches
would seriously degrade HX effectiveness in a conventional
manifolded counterflow HX, but our pair-wise balance
provides protection from this degradation.
The preceding argument is based on perfect pair-wise
balance, and so could fail in our parallel-plate HX if there is a
significant flow nonuniformity across the wide width of a
channel, which was 15 mm in our design (but effectively only
about 10 mm wide, owing to the numerous standoffs used to
maintain the narrow microchannel height near its desired value,
and to prevent channel collapse from the large differential
pressure that exists during operation between adjacent counter
flowing channels). In hindsight it may have been better to
partition the parallel plate into a set of rectangular channels,

11 Copyright © 2011 by ASME
where each has its own return slot at the hot end, rather than
relying on a common return slot. One reason we did not do this
is that our particular fabrication method used discrete copper
standoffs. If we had constructed continuous sidewalls out of
the same copper material, we would have created a large axial
heat leak that could have ruined the energy efficiency of our
pasteurizer. Another reason was that the manifolding scheme
at the cold end would have required greater complexity. For
future designs, we recommend consideration of a different
fabrication method that would permit thin polymer sidewalls
(ideally no thicker than the existing parallel-plate channel
walls) to be constructed. The structure would then resemble a
traditional tube-bundle HX. It would still require that every
channel be connected to a nearest neighbor at the hot end, but it
does not necessarily need to be a vertical feedthrough as in our
prototype. For example, if one used thin-walled microchannels
having a square cross section, one could connect the hot end of
each channel to its nearest neighbor on the left or right, rather
than the neighbor above or below. This would provide
identical performance in either case, but one method might
prove easier to fabricate than the other.
While the HX design concept described herein is highly
effective in recovering the heat used in pasteurization as a
continuous-flow process, it should be noted that there will
always be an initial non-recoverable energy expenditure as the
pasteurizer develops its steady-state temperature profile. The
theoretical minimum would obviously be determined by the
heat capacity of the pasteurizer and its initial contents, but it
would require unusual measures to approach that minimum
(e.g., installing preheaters along the length of HX, with
temperature sensing devices to set the desired initial linear
profile). As a practical matter, the simplest approach to
accelerating the start-up time might be to initialize the HX with
pure water at a very slow flow rate, perhaps 1/20 the design
flow rate; the system will equilibrate to the desired asymptotic
linear temperature profile much more quickly with a slow flow.
As our experiments will show, the cost in wasted energy, time,
and fluid utilization can be quite significant if the device is
operated at its full design flow rate during startup.
ULTRA-PASTEURIZER DESIGN CONSIDERATIONS
AND PROTOTYPE CONSTRUCTION DETAILS
Raw cow‟s milk is mostly water, typically 87% by weight.
It also contains milkfat in the range of 2.4%-5.5%, and various
solids (primarily the sugar lactose, minerals containing
calcium, phosphorus, potassium and sodium, and the protein
casein; there are also various other trace minerals, acids,
enzymes, vitamins) that total in the range of 7.9-10.0% [32].
Thermal conductivity and heat capacity are substantially the
same as water, however viscosity is higher than for water, and
can vary substantially from different sources. We elected to
design based on what is likely a worst-case (as far as pumping
power is concerned) viscosity assumption of .006 Pa∙s (one of
the higher published figures for milk at 25°C, and over 6x that
of 25°C water), independent of temperature. Raw milk also
contains particles, predominantly casein-coated fat globules,
which are typically in the 3-6 µm range, but can be as large as
22 µm. This was not thought to be a limiting factor in our
design, as the microchannels we experimented with were 2-3x
that size, and moreover we observed from filtration
experiments that fat globules are highly deformable, readily
passing intact through screens as fine as 2 µm.
Our nominal design goal was to ultrapasteurize 1 ml/sec of
milk with an overall steady-state power consumption that was
only 1/100 of what would be required to heat the milk with no
heat recuperation; this corresponds to a power budget of 4.8
Watts. The vapor pressure of water at 135°C is 313 kPa
(absolute); to prevent internal boiling we chose to incorporate a
back-pressure regulator on the output side of the pasteurizer set
at 240 kPa (gauge). This implies a minimum pumping power
of 0.24 Watts; given the likely inefficiency of small pumps and
the additional pressure drops that would be added by the
microchannels, we felt it prudent to budget 1 Watt for pumping
power, leaving 3.8 Watts for heat leaks and HX ineffectiveness.
The selection of material was a critical design decision.
We assumed the HX core would be constructed from a single
material having isotropic thermal conductivity k. Holding all
physical dimensions constant (thickness, length, etc.), and
varying k, there is a tradeoff between losses caused by
restricted heat conduction through the microchannel walls
(which reduces effective heat transfer coefficient between
adjacent channels), and losses due to excessive axial heat
conduction (Figure 5).
Figure 5 – Plot showing unrecovered heat per channel
pair as a function of channel wall thermal conductivity,
illustrating the tradeoff between axial heat conduction and
through-wall heat conduction for a particular HX
configuration. (For illustrative purposes only; our actual
design used different HX dimensions and flow rate).
It was clear that a HX made from most metals would likely
have an axial heat flow that exceeded our entire power budget.
The one exception was stainless steel (the material which most
commercial pasteurizers are constructed from), which has

12 Copyright © 2011 by ASME
unusually low thermal conductivity; for AISI 316, k~16
W/m∙K, which would have produced a viable design (in fact
somewhat better than polyimide). We nevertheless elected to
design a polymer film HX, believing that a polymeric
construction could enable the lowest cost, most manufacturable
design. Polymer heat exchangers are used commercially and
can give good performance if the channel walls are sufficiently
thin to overcome their low thermal conductivity [43].
Polyimide film (k~0.16 W/m∙K, 12-µm thick) was chosen in
order to leverage capabilities from the flexible printed-circuit
board industry.
Figure 6 – Exploded view of a single balanced
counterflow µchannel pair prior to lamination. The milk
enters the upper adhesive layer from one side of the cold
end, travels to the hot end, passes vertically through a
slot in the mid layer (which is populated with copper
standoff bumps on both sides to maintain channel
separations), and exits the lower adhesive layer at the
cold end through the side opposite to the entrance.
Our proposed pasteurizer design was a stack of 100 pairs
of microchannels (with nominal 50 µm separation between the
12-µm thick parallel channel walls), to be laminated using
flexible printed-circuit board technology. Figure 6 shows an
exploded view of the components of one µchannel pair. The
channel separation was defined and maintained by a 2-
dimensional array of photolithographically defined 100-µm
square 40-µm tall copper standoffs on each side of the „middle
layer‟ (the channel wall separating the two counterflowing
flows in the µchannel pair). The in-plane spacing of the
standoffs was chosen based on our requirement that the
polyimide films (channel walls) not elastically deflect more
than 5 µm under the influence of differential pressure between
adjacent layers. The standoffs were staggered in a diagonal
array; simulations showed that this configuration minimized the
number of standoffs required to maintain channel separation
without imposing penalties in flow friction or heat transfer
(Figure 7). The worst-case pressure differential of 250 kPa
occurs at the cold end of the HX, where the inbound flow is
entering at its maximum pressure, and the outbound flow is
exiting at its lowest pressure. Consequently the standoffs had
to be more closely spaced in the cold region (Figure 8). On
average, the standoffs occupied approximately 1/3 of the
channel cross section, so our parallel-plate approximation was
not very precise; finite-element modeling was used to get a
more precise estimate of the actual friction factor.
Figure 7 – Detail of copper standoffs on both sides of
each mid-layer. These prevent channel collapse due to
differential pressure between adjacent counterflows. The
densest pattern (shown here) is near the cold end, where
pressure differentials are greatest.
Figure 8 – Diagram of a single mid-layer, showing
loosening of the standoff spacing from cold end to hot
end. After lamination of the multilayer stack, nichrome
ribbon heaters were inserted into the two slots on either
side of the hot zone.
The average flow length of each microchannel was 75 mm,
with the first 50 mm being the region of linear temperature
ramp. The last 25 mm contained the hot zone, where in
addition to the standoffs, each polyimide film had a continuous
coated film of 3-µm thick electrodeposited copper on one side,
which simulations (Figure 9) showed had sufficient thermal
conductance to impose a uniform 135°C channel wall
temperature with less than 0.2°C variance across the width of
the hot zone if heated from both edges of the pasteurizer (as
channel widths were 15 mm, the maximum conduction distance
from either heater to mid-channel was only 7.5 mm). A 200-
µm wide laser-punched slot in every 2nd polyimide layer creates
the return flow path for each µchannel pair. For a nominal 1
ml/s flow, the nominal flow velocity is 20 mm/s, so on average
the milk heats linearly for 2.5 s, then dwells for 2.5 s in the hot
zone (reversing direction halfway through), then cools down
for another 2.5 s.

13 Copyright © 2011 by ASME
Figure 9 – Simulation showing the beneficial effect of
plating 1 µm of copper on the polyimide channel walls in
the hot zone. There would be less than a 0.2°C
temperature difference from the edges of the HX (where
the nichrome heaters are located) to the middle. This
permits very precise and stable control of pasteurization
temperatures throughout the entire hot zone. In the
actual prototype, the copper was 3 µm thick, which would
function even more effectively than 1 µm as an isothermal
surface.
The Reynolds number of the flow for our worst-case
viscosity assumption would be Re=0.3, leading to a worst-case
pressure drop of 250 kPa from inlet to outlet, which when
added to the back pressure at the outlet requires a pump capable
of 490 kPa. Using the parallel-plate constant-heat-flux double-
wall value of Nu=8.235 (not strictly accurate, as the standoffs
interrupt the parallel plate geometry), we would expect each
channel‟s convective heat transfer coefficient to be hconv
=kwaterNu/D=(0.61 W/m∙K)(8.2)/(100 µm)=50,000 W/m2∙K.
The polyimide films themselves have a significantly lower
conductive heat transfer coefficient, hcond= (0.16 W/m∙K)/(12.5
µm)=12,800 W/m2∙K. However, to calculate total heat transfer
from the outbound flow to inbound flow within each µchannel
pair of our periodic structure, one must reduce hconv by a factor
of 2, and increase hcond by a factor of 2, with the result that the
combined heat transfer coefficient is heff=1/(2/hconv +0.5/hcond
)=12,600 W/m2∙K. With a mass flow rate per µchannel pair of
0.01 ml/s, multiplied by water‟s heat capacity of 4.2 MJ/m3∙K,
we have a heat capacity rate of .042 W/K per pair, which is
fully transferred over an effective area of 0.001 m2 (50 mm
long ramp ×10 mm effective width × 2 heat transfer surfaces),
hence a heat transfer rate of 42 W/m2∙K. The ratio of this to heff
gives NTU=12,600/42=300, which gives us an expected HX
effectiveness of 300/301=99.7%, i.e., only 0.3% of the applied
heat should be lost. Calculations and modeling show that axial
heat flow does not significantly affect this result, increasing
total heat loss by less than 5%. However the use of stainless
steel instead of polyimide would have approximately doubled
the heat loss, reducing HX effectiveness to 99.3%.
Owing to the parabolic laminar-flow velocity profile
between the parallel plates, we expect a residence time
distribution (RTD) wherein the milk at the center of the channel
would dwell for as little as 1.7 s, but milk near the channel
walls could reside for much longer than the average 2.5 s. We
had some concern that milk constituents immediately adjacent
to the wall could experience such long dwell times as to
exacerbate fouling or adversely affect the taste of the milk.
Brownian motion will mitigate this to some degree; for
example a 100 nm sphere has a diffusion coefficient of 4.4
µm2/s [44], so after 2.5 s its RMS expected displacement is
~3.3 µm, placing it in a slightly different streamline. Smaller
particles (e.g., protein molecules) will diffuse substantially
farther, so the main concern would be for larger particles. If
the long tail of the RTD is later deemed to be a problem, it may
be necessary to modify the microchannels with structures to
enhance mixing [45].
ULTRA-PASTEURIZER EXPERIMENTAL RESULTS
USING WATER
We constructed an experimental prototype of our design
using an outside contractor that specialized in the manufacture
of both laminated printed circuit (PC) boards and flexible
copper circuits on polyimide film. Conventional PC board
adhesive materials were used to laminate the layers together
into a sealed unit. To simplify the process flow, we laminated
only 20 layers of microchannels (i.e., 10 balanced pairs) which
is 1/10 the number of layers intended for our nominal design.
Flow rates were therefore 1/10 of their nominal values. Input
and output ports for each microchannel were created on
opposite sides of the cold end by leaving 10-mm wide openings
in the sidewalls of the laminate. The laminate was
instrumented with a series of thermocouples on its upper and
lower faces, measuring temperatures along the centerline of the
topmost and bottom-most channel wall at several axial
locations (the top series measuring an inbound flow, and the
bottom series an outbound flow, with close but not perfect axial
co-location). Two gasketed Ultem® manifolds, one over the
input ports and one over the output ports, were held in
compression against the sidewalls of the HX using four
through-bolts. To prevent rupture of the topmost and bottom-
most channel walls and provide insulation, the layer stack was
held under compression by sandwiching it between two 5.15-
mm thick rectangular blocks of Ultem® polymer which were in
turn sandwiched between two 8.2-mm stainless steel blocks, all
held together by a set of through-bolts around the perimeter.
Figure 10 is a photo of the assembled prototype. A
nichrome ribbon heater was inserted into a slot on one edge of
the hot zone and held in compression against the edges of the 3-
µm thick copper heat spreaders whose function is to distribute
heat isothermally throughout the hot zone. An identical heater
was situated on the side directly opposite to the first heater.
The entire assembly was enclosed in a polystyrene foam block
in an attempt to minimize heat leaks to the ambient.
Most experiments were conducted with deionized water, in
order to collect performance data using a well-understood
liquid that posed no risk of fouling. Pressure-vs-flow rate tests
at room temperature with no applied heat showed the linear

14 Copyright © 2011 by ASME
relationship characteristic of laminar flow (Figure 11), having a
slope consistent with our expectations, which confirmed that
the flow was well distributed among the 10 µchannel pairs, and
that our internal standoffs were fulfilling their function of
holding microchannel dimensions near their nominal values.
Applying a fixed amount of power to the heaters while
maintaining flow rate constant at its nominal design value (0.1
ml/s rather than 1 ml/s, as this had 1/10 the number of layers
called for in a full system) resulted in the internal temperatures
evolving as shown in Figure 12.
Figure 10 – Photograph of the assembled prototype UHT
pasteurizer, including Ultem input and output manifolds,
fittings, and back-pressure regulator. This prototype had
only 10 µchannel pairs (each pair being a microchannel
counterflow HX), rather than 100 in the nominal design.
The entire stack is 1.25 mm tall (its edge barely visible in
this photo), sandwiched between two 5.15-mm thick
Ultem blocks, and then two 8.2-mm thick stainless steel
blocks, all held together with through-bolts.
The temperature in the hot zone rises slowly, approaching
an asymptotic value with a traditional exponential decay. The
long time constant (~120 s) initially surprised us, until we
realized that the very high effectiveness of the HX was working
against us during the heat-up (something that could be partly
mitigated by operating at lower flow rate until the desired
internal temperature profile has been approximated), as was the
heat capacity and thermal diffusivity of the pair of Ultem
blocks (aerogel or even vacuum would have been preferable
methods of insulation). We noted that the ribbon heaters were
significantly hotter (typically by 40 to 60°C) than the hot zone,
and the differences were not identical nor stable. We attribute
this to a poor thermal interface between the heater and the thin
copper films (on polyimide) in the hot zone (yet another reason
why better TIM materials are needed!). In the future it may be
much better to integrate thin-film heater resistors directly on or
within the hot zone channel walls, rather than attempting to
inject the heat from the sides of the HX.
Figure 11 – Flow test of the pasteurizer using room-
temperature water. The pressure drop was ~50 kPa (7
psi) at the nominal design flow of 0.1 ml/s for this 1/10-
scale device. The linearity implies that no significant
channel wall collapse between the parallel-plate laminar-
flow layers occurred, despite differential pressures as
high as 100 kPa across the 12-µm thick polyimide
channel walls, thus confirming adequacy of the standoffs.
Figure 12 – Pasteurizer test using constant water flow of
0.1 ml/s and a constant 19-W total power input from the
two heaters. T1 and B1 are thermocouples located in the
hot zone (top-most and bottom-most layers), and reached
120°C when the test was terminated (later tests
exceeded 130°C). The long time constant (still increasing
at 200 s) is largely due to the heat capacity of the two
large Ultem blocks sandwiching the pasteurizer. B2 and
T2 are located at an axial location in between the cold
end and the start of the hot zone, and as expected exhibit
a linear relationship with B1 and T1. The difference
between outlet and inlet manifold fluid temperatures, a
measure of net heat transferred to the liquid, cannot be
resolved on this graph but was 1.9-2.0°C at the end of
this test, suggesting an HX effectiveness of ~98%. The
two nichrome heaters are substantially hotter than the hot
zone (and not equal), implying poor and inconsistent
thermal contact with the edges of the copper heat
spreaders (which maintain the hot zone as isothermal).

15 Copyright © 2011 by ASME
With a hot zone temperature of 120°C, the temperature of
the water after it exited the manifold was elevated 2.0°C above
the inlet temperature, implying 98% heat recovery. However,
at a 0.1 ml/s flow rate this implies a net transfer of slightly less
than 1 W of heat into the water. Yet, a much higher heater
power, 19 W, was being applied. Quick calculations show that
a heat flow of approximately this magnitude would be expected
during the time period tested, based on the heat capacity and
thermal conductivity of the Ultem „insulating‟ blocks. Even
without improving the insulation technique, the relative impact
on pasteurizer efficiency of this external heat path would have
been much less had we built our nominal 200-layer design
instead of a scaled-down 20-layer version: the full-scale design
would have had similar magnitude external heat leaks, but
would process 10 times as much liquid, making the external
and internal losses within a factor of 2 of each other. Scaling
our design further to much larger flows, such as are processed
in commercial plants, would further reduce the effects of
external heat leaks, as the ratio of exterior surface area to HX
heat duty would be vastly reduced. While one might be
concerned that better exterior insulation would have redirected
heat flow into the HX and thereby reduced its effectiveness
below the measured 98%, a little thought regarding the 2-3
order of magnitude difference in heat-transfer coefficients
present at the outer surface of the HX stack between heat
flowing radially inward (into the stack of water-cooled
microchannels) vs. heat flowing outward (through polymer
and stainless steel blocks and finally ambient air) convinced us
that this effect would be insignificant.
RESULTS USING MILK
Our ultimate goal was to operate the ultrapasteurizer with
milk rather than water, but were concerned about the possibility
of fouling, which is a well known, highly complex, extensively
studied (but still not fully understood) problem in commercial
pasteurization systems. Bansal and Chen [46] have written
good review papers on the subject. To give an idea of the
magnitude of the problem, we quote two sentences here from
their paper: “Generally, milk fouling is so rapid that heat
exchangers need to be cleaned every day to maintain
production capability and efficiency and meet strict hygiene
standards.” “About 80% of the total production costs in the
dairy industry can be attributed to fouling and cleaning of the
process equipment”. In touring a commercial UHT milk
pasteurization plant, we were informed that the daily
“Cleaning-in-Place” protocols consisted of a 3-4 hour 17-step
procedure that included a sequence of high-temperature rinses
and flushes such as chlorinated water, NaOH, detergents, and a
nitric-phosphoric acid mixture.
Fouling is known to occur in two general forms: a soft
spongy buildup of denatured proteins that deposits primarily in
the range 75-110°C, and a much harder scale-like coating
(predominantly calcium phosphate) that precipitates out at
higher temperatures. There is often an induction period of very
slow fouling that lasts approximately 20 minutes (though it can
range from 1 to 60 minutes), followed by a much more rapid
buildup rate. Numerous researchers have studied various
potential anti-fouling coatings (e.g., titanium nitride, carbon
nanotubes, low-surface energy fluoropolymers such as PTFE)
that can be deposited on stainless steel. The general idea is to
either prevent the fouling products from nucleating on the walls
in the first place, and/or to suppress adhesion of those products
after they have formed. Some of the most promising published
results (studied at conventional rather than UHT temperatures)
used proprietary coatings which appear to be composite
materials having significant fluoropolymer content. One such
proprietary material, AMC148 [47], showed significant
improvements, ranging from a 76% to a 92% reduction in
fouling rate [48]. Puri and Jun [49] experimented with carbon
nanotubes (CNTs), achieving excellent anti-fouling
performance (over 6 hours of operation with no measurable
build-up) but there were concerns about high cost, potential
toxicity, and durability of the CNTs that have not been
sufficiently investigated. Balasubramanian and Puri [50, 51]
tried TM-117P (a graded Ni-P-PTFE coating), AMC148-18 and
also LECTROFLUOR® 641 [52] (another proprietary coating);
all showed reduced fouling rates, but TM-117P had some side
effects (discoloration, fluorine leaching), and AMC148 is
incompatible with high-pH cleaning solutions used in the dairy
industry. LECTROFLUOR 641 was the most promising of
their coatings, reducing the incidence of fouling by 94-95% vs.
control, with no measurable leaching (hence deemed
compatible with food processing equipment). They went on to
do a pilot trial of LECTROFLUOR on a plant that pasteurized
milk, and another that processed tomato juice, demonstrating a
reduction in both the thermal energy budget and the downtime
of the systems.
Unfortunately our operating regime was so different from
the published literature (i.e., microchannels 50-100x finer than
is conventional; laminar flow instead of turbulent; polyimide
walls rather than stainless) that no prediction could be made
about our microchannel system. We decided to create test
chambers containing a single parallel-plate microchannel with
50-µm separation. The chamber was constructed from two
slabs of copper (a later version was partially stainless steel,
Figure 13), one of which had a shallow cavity machined into its
surface to create the microchannel zone. The inner surfaces
could be coated with thin films of polymer or other materials to
simulate alternative channel materials, and in particular to
experiment with various anti-fouling coatings. A pair of
threaded holes (NPT pipe threads) in opposite ends of one of
the slabs provided an input and output for the milk. Raw milk
was pumped at a nominal rate of 0.1 ml/s, creating an average
velocity of 20 mm/s in the microchannel. A 285 kPa back-
pressure regulator on the output prevented boiling. The
chamber could be either wholly or partially immersed in an
isothermal oil bath to create different temperature profiles
along the length of the microchannel, in an attempt to
approximate the actual temperature profile that existed in the
pasteurizer. Each test was run until the device was nearly
completely clogged; the chamber was then disassembled,
inspected, and cleaned for the next test.

16 Copyright © 2011 by ASME
Figure 13 – Design showing one half of the chamber used
to study milk fouling in a single 50-µm microchannel. The
chamber was a copper/stainless hybrid design wherein
the copper portion was immersed in a 135+°C oil bath.
The stainless steel (input) end of the chamber permitted
the milk temperature to ramp somewhat gradually to the
maximum temperature before exiting in the copper end;
this was thought to mimic our UHT pasteurizer conditions
better than an earlier all-copper design, which thermally
shocked the milk with a very rapid heat-up due to
copper’s good thermal conductivity.
Figure 14 – Typical results from milk fouling experiments
varying milk composition and pasteurization temperature.
The test microchannel chamber did not remain viable for
a significantly longer time with skim milk than whole milk,
but reducing the temperature from 130°C to 110°C almost
doubled the operational lifetime.
Through an extensive series of experiments, we found that
the microchannel test chamber, when tested at peak
temperatures of 135°C, would clog too quickly to be of
practical use, in all circumstances that we tested. Figure 14
shows results of a representative experiment. In this and other
experiments, the longest time that we could sustain flow in the
50-µm chamber was 2 hours (or 3 hours in the 65-µm
chamber), and 30-60 minute lifetimes were much more
common. This is not all that surprising; if the fouling builds up
at a certain rate, one would expect narrower channels to clog
more quickly. Laminar flow may also exacerbate fouling by
creating long residence times very near the walls. Other than
the shorter life, in most other respects our observations were
consistent with the literature. We observed both the „soft‟ and
the „hard‟ types of fouling in their expected temperature ranges.
We tested a LECTROFLUOR coating in our copper test
chamber but found that, unlike the published work, it did not
reduce the time to clogging in our microchannels. It did
however greatly reduce the adhesion of the deposits, making
them easily removed with light abrasion. This suggests an
explanation for LECTROFLUOR‟s anti-fouling action in
conventional pasteurizers: with macroscopic channels and
higher Reynolds numbers, fouling may still occur on the walls,
but the adhesion is poor enough that turbulence can stimulate
its detachment.
Based on the fouling chamber experiments, we did not
expect our prototype UHT pasteurizer unit to last very long
when operated with milk (rather than water). We first flow-
tested the pasteurizer with room-temperature raw milk at the
nominal 0.1 ml/s flow rate; no fouling (as measured by
differential pressure in the HX) was observed after 20 minutes.
We then applied power to the heater to maintain the hot zone at
70°C, simulating HTST pasteurization conditions. After 5
minutes we again saw no evidence of fouling. We then ramped
up the power with the intent to reach the full 135°C UHT
process temperature. Upon passing 110°C, a small pressure rise
was noted. Passing 120°C and approaching 130°C, the pressure
began to rise very rapidly, with complete blockage occurring
within 10 seconds. The prototype was unusable thereafter; by
allowing the channels to completely clog, it was then
impossible to flush the system with any cleaning fluids. At this
time it is not understood why the prototype‟s lifetime was so
much shorter than the fouling chamber test data led us to
expect.
In summary, our UHT microchannel pasteurization
architecture does appear to enable very high energy efficiency,
which should only improve with scaling to larger capacity
systems. It showed promising results in experiments with
deionized water, with no evident lifetime limitation. However,
in order to apply the architecture to milk, or any other liquid
that generates fouling when heated, design modifications may
be required to extend the useful life of the system. These may
include the use of somewhat larger channels (50 µm was a very
aggressive design choice), identifying suitable low-adhesion
coatings for the UHT environment, developing periodic
cleaning processes, or simply choosing an application requiring
less aggressive thermal treatment (e.g., a conventional HTST
pasteurizer, which would be expected to foul much more
slowly and be much easier to clean). There may be other
applications in the emerging field of microreactors where the
energy efficiency enabled by the high degree of heat recovery
may be useful.

17 Copyright © 2011 by ASME
ACKNOWLEDGMENTS
The authors would like to thank David Nash, Nathan
Pegram, Mike Vinton and Ted Ellis for their support in the
production of our prototype devices, and the laboratory and
facility staff at Intellectual Ventures Lab. We would also like
to thank John Magerlein, Ray Beach, Ken Goodson, and Mark
Zediker for helpful discussions regarding the history and
commercial status of microchannel cooling. The authors thank
Bill and Melinda Gates for their active support of this work and
their sponsorship through the Global Good Fund.
REFERENCES
[1] Blodgett, A. J., and Barbour, D. R., 1982, "Thermal
Conduction Module: A High-Performance Multilayer Ceramic
Package," IBM Journal of Research and Development , 26(1),
pp.30-36. For a photo of the TCM, see http://www-
03.ibm.com/ibm/history/exhibits/vintage/vintage_4506VV2137
.html)
[2] Keyes, R. W., 1975, “Physical Limits in Digital
Electronics”, Proceedings of the IEEE, 63, pp. 740-767.
[3] Anacker, W., 1979, “Computing at 4 degrees Kelvin”, IEEE
Spectrum, May issue, p. 26.
[4] Matisoo, J., 1980, “The Superconducting Computer”,
Scientific American, 242, pp. 50-65.)
[5] Dennard, R. H., Gaensslen, F. H., Rideout, V. L., Bassous,
E., LeBlanc, A.R., 1974, “Design of Ion-Implanted MOSFET's
with Very Small Physical Dimensions”, IEEE Journal of Solid
State Circuits, 9(5), pp. 256-268.
[6] Bean, K. E., 1978, “Anisotropic Etching of Silicon”, IEEE
Transactions on Electron Devices, ED-25(10), pp. 1185-1193.
[7] Kendall, D. L., 1975, “On Etching Very Narrow Grooves in
Silicon”, Applied Physics Letters, 26(4), pp.195-198.
[8] Tuckerman, D. B., and Pease, R. F. W., 1981, “High-
Performance Heat Sinking for VLSI”, IEEE Electron Device
Letters, EDL-2(5), pp. 126-129.
[9] Tuckerman, D. B., and Pease, R. F. W., 1982, “Optimized
Convective Cooling using Micromachined Structures”,
Electrochemical Society Extended Abstracts, pp. 197-19,
Abstract #125.
[10] Tuckerman, D. B., and Pease, R. F. W., 1982, “Ultrahigh
Thermal Conductance Microstructures for Cooling Integrated
Circuits”, 32nd Electronics Components Conference
Proceedings, pp. 145-149.
[11] Tuckerman, D. B., and Pease, R. F. W., 1983,
“Microcapillary Thermal Interface Technology for VLSI
Packaging”, Symposium on VLSI Technology, pp. 60-61.
[12] Tuckerman, D. B., 1984, “Heat-Transfer Microstructures
for Integrated Circuits”, Stanford University PhD Thesis; also
distributed as Lawrence Livermore National Laboratory report
UCRL-53515.
[13] Leyerle, C., 2005, “The Ten Myths of Liquid Metal
Thermal Interfaces”, presented at the Next Generation Thermal
Management Materials and Systems Conference, available at
http://www.enerdynesolutions.com/downloads/rti_present_061
605.pdf
[14] Cola, B. A., 2010, “Carbon Nanotubes as High
Performance Thermal Interface Materials”, Electronics
Cooling, 16(1), pp. 10-15.
[15] Qu, L., and Dai, L., 2007, “Gecko-Foot-Mimetic Aligned
Single-Walled Carbon Nanotube Dry Adhesives with Unique
Electrical and Thermal Properties”, Advanced Materials,
19(22), pp. 3844-3849.
[16] Gao, Y., Marconnet, A., Panzer, M., LeBlanc, S., Dogbe,
S. ,Ezzahri, Y., Shakouri, A., and Goodson, K., 2010,
"Nanostructured Interfaces for Thermoelectrics", Journal of
Electronic Materials, 39, pp. 1456-1462. See also
http://microheat.stanford.edu/Journals.html for other papers
from Prof. Goodson‟s Microscale Heat Transfer Laboratory.
[17] Tong, T., Zhao, Y., Delzeit, L., Kashani, A., Meyyappan,
M., and Majumdar, A., 2007,“Dense Vertically Aligned
Multiwalled Carbon Nanotube Arrays as Thermal Interface
Materials”, IEEE Transactions on Components and Packaging
Technologies, 30, pp. 92-100.
[18] Ge, L., Sethi, S., Ci, L., Ajayan, P. M., and Dhinojwala,
A., 2007, “Carbon Nanotube-based Synthetic Gecko-tapes”,
Proc. Nat. Acad. of Sci., 104, p. 10792
[19] Ken Goodson, private communication; see also
http://www.eetimes.com/electronics-news/4212401/Nanotape-
could-make-solder-pads-obsolete
[20] Mundinger, D., Beach, R., Bennett, W., Solarz, R.,
Krupke, W., Staver, R., and Tuckerman, D., 1988,
"Demonstration of High-Performance Silicon Microchannel
Heat Exchangers for Laser Diode Array Cooling, Applied
Physics Letters, 53(12), pp. 1030-1032.
[21] Ray Beach, private communication.
[22] Mark Zediker, private communication.
[23] Ken Goodson, private communication.
[24] Campbell, L. A., Ellsworth Jr., M. J., and Sinha, A. K.,
2009, “Analysis and Design of the IBM Power 575
Supercomputing Node Cold Plate Assembly”, Proc.
IPACK2009, paper #89244.
[25] Kandlikar, S. G., 2010, “History, Advances, and
Challenges in Liquid Flow and Flow Boiling Heat Transfer in
Microchannels: A Critical Review”, Proc. Intl. Heat Transfer
Conference, IHTC14-23353.
[26] Ellsworth Jr., M. J., Goth, G. F., Zoodsma, R. J., Arvelo,
A., Campbell, L. A., and Anderl, W. J., 2011, “An Overview of
the Power 775 Supercomputer Water Cooling System”, Proc.
IPACK2011, paper #52130.
[27] http://www.theregister.co.uk/2010/11/22/ibm_blue_gene
_q_super/page2.html
[28] Yablonovitch et al, Center for Energy Efficient Electronics
Science, http://www.e3s-center.org/
[29] Goodson, K. E., Kurabayashi, K., and Pease, R. F. W.,
1997, "Improved Heat Sinking for Laser-Diode Arrays using
Microchannels in CVD Diamond," IEEE Transactions on
Components, Packaging, and Manufacturing Technology, B20,
pp. 104-109.
[30] Chandokar, S., and Pease, R. F. W., 2009, “Theoretical
Analysis of Thermal Actuator Based Nanoimprint
Lithography”, Proc. NNT2009 (San Jose CA).

18 Copyright © 2011 by ASME
[31] Brandner, J. J., Bohn, L., Henning, T., Schygulla, U.
and Schubert, K., 2007, “Microstructure Heat Exchanger
Applications in Laboratory and Industry”, Heat Transfer
Engineering, 28(8), pp. 761 – 771.
[32] Bylund, G., 1995, “Dairy Processing Handbook”, Tetra
Pak Processing Systems.
[33] FleetwoodGoldcoWyard, Inc. product literature.
[34] Lawrence Berkeley National Laboratory report LBNL-
50934
[35] R. H. Feldmeier, U.S. Patent 7,435,440
[36] Andreatta, D., Yegian, D., Connelly, L, and Metcalf, R.,
1994, “Recent Advances in Devices for the Heat Pasteurization
of Drinking Water in the Developing World”, Proc. 29th
Intersociety Energy Conversion Engineering Conference,
AIAA, Inc.
[37] Denkenberger, D.C., and Pearce, J. M., 2006, “Compound
Parabolic Concentrators for Solar Water Heat Pasteurization:
Numerical Simulation”, Proceedings of the Solar Cookers
International Conference, Granada, Spain.
[38] Denkenberger, D. C., 2010, “Low-cost high-effectiveness
microchannel heat exchanger: Fabrication, Modeling and
Validation”, PhD thesis, University of Colorado at Boulder.
[39] Creare Inc., “Heat Exchangers”, http://www.creare.com/
services/fluid/heat_ex.html
[40] Deane, G. et al, US patent applications 12/462,200,
12/462,206, and 12/462,213, filed 7/29/2009.
[41] , D. P., 2003, “Fundamentals of
Heat Exchanger Design”, Chapter 12.
[42] Kays, W. M., and Crawford, M. E., 1980, Convective Heat
and Mass Transfer, 2nd edition, p.103, Table 8-2.
[43] Zaheed, L. and Jachuck, R. J. J., 2004, “Review of
Polymer Compact Heat Exchangers with Special Emphasis on a
Polymer Film unit”, Applied Thermal Engineering, 24, pp.
2323-2358.
[44] Cohen, A.E., and Moerner, W. E., 2008, “Controlling
Brownian Motion of Single Protein Molecules and Single
Fluorophores in Aqueous Buffer”, Optics Express, 16(10), pp.
6941-6956.
[45] Cantu-Perez, A., and Gavriilidis, A., 2008, “Mixing and
Residence Time Distribution Studies in Microchannels with
Floor Herringbone Structure”, Proc. European COMSOL
2008 Conference, Hannover, Germany.
[46] Bansal, B., and Chen, X. D., “Fouling of Heat Exchangers
by Dairy Fluids – A Review”, 2005, Proc. 6th Intl Conference
on Heat Exchanger Fouling and Cleaning, RP2, article 23.
Also “A Critical Review of Milk Fouling in Heat Exchangers”,
2006, Comprehensive Reviews in Food Science and Food
Safety, 5, pp. 27-33.
[47] Advanced Materials Components Express, LLC product
literature (www.amcx.com).
[48] www.reeis.usda.gov/web/crisprojectpages/203049.html
[49] Puri, V. M., and Jun, S., 2003-4, “Carbon Nanotubes
(CNTs) for Quantum Gains in Efficiency during Thermal
Processing of Foods”, Penn State University Seed Grant
Program Abstract.
[50] Balasubramanian, S., and Puri, V. M., “Reduction of Milk
Fouling in a Plate Heat Exchanger System Using Food-Grade
Surface Coating”, Transactions of the ASABE. 52(5): 1603-
1610.
[51] Balasubramanian, S., and Puri, V. M., 2009, “Thermal
Energy Savings in Pilot-Scale Plate Heat Exchanger System
during Product Processing using Modified Surfaces”, J. Food
Engineering, 91, pp. 608-611.
[52] General Magnaplate Corp. product literature (web site
magnaplate.com)