International Journal of Heat and Mass Transfer 48 (2005) 1572–1582

www.elsevier.com/locate/ijhmt

Phase change phenomena in silicon microchannels

Lian Zhang *, Evelyn N. Wang, Kenneth E. Goodson, Thomas W. Kenny
Department of Mechanical Engineering, Stanford, CA 94305-4021, USA

Received 26 February 2004; received in revised form 6 September 2004

Available online 22 December 2004

Abstract

Understanding the boiling process and two-phase flow behavior in microchannels is the key to developing micro-
channel heat sinks for high-power microprocessors. We conducted experiments in micromachined silicon channels with
a range of 27–171 lm hydraulic diameters and varying surface roughnesses. Bubble nucleation, flow patterns, wall tem-
perature, as well as transient pressure fluctuations were recorded and analyzed. We observed both typical nucleate boil-
ing and eruption boiling with large amounts of wall superheat in these channels, and recorded up to 138 kPa transient
pressure fluctuations due to bubble nucleation. We found the boiling mechanism is strongly dependent on the wall sur-
face roughness, and we explained the boiling mechanism in sub-150 lm diameter channels with HsuÕs model.

2004 Elsevier Ltd. All rights reserved.

1. Introduction ture must be maintained in the approximate range of

80–100 C for proper functionality and reliability. By
In accordance with MooreÕs Law [1], the number of that time, air cooling will no longer satisfy the needs
transistors integrated on Intel microprocessors has in- of these high-power chips. Scientists and researchers
creased from 3.1 · 106 (Pentium, 1993) to 9.5 · 106 (Pen- have long been pursuing solutions for high-power chip
tium III, 1999) to 178 · 106 (Pentium 4, 800 MHz, 2004) thermal management. Since Tuckerman and Pease [4]
in the past decade [2]. The continual increase in the inte- proposed the idea of liquid cooling with microchannels
gration density causes a rapid rise in power generation in 1981, microchannel heat sinks have been considered
from microprocessors. With 82 W of heat generation one of the most promising cooling devices for future
by a Pentium 4 processor, the noise produced by the high-power chips [5]. The latest idea is to directly attach
cooling fan is approaching the acceptable limits for con- a microchannel heat sink to the back side of the micro-
sumers. According to ITRS (International Technology processor while a coolant is pumped into the channels
Roadmap for Semiconductor), the heat dissipation from [6]. The heat generated by the chip is absorbed through
a single cost-performance chip is expected to exceed the liquid-vapor phase change and is carried by the
100 W by the year 2006 [3]; however the chip tempera- vapor into a condensing chamber. The vapor condenses
and is re-circulated back into the microchannel heat
sink. Taking advantage of the latent heat associated with
phase change, the heat absorption is much more efficient
*
Corresponding author. Present address: Molecular Nano- than that of a single-phase liquid. The recent develop-
systems, 977 Commercial Street, Palo Alto, CA 94303, USA. ment of electroosmotic pumps even allows a miniature
E-mail address: [email protected] (L. Zhang). closed-loop cooling system [7].

0017-9310/$ - see front matter
2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijheatmasstransfer.2004.09.048
 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582 1573

Nomenclature

A Cross sectional are a of a microchannel Tsat liquid saturation temperature

cp specific heat Tw wall temperature
DH hydraulic diameter of a microchannel, DTsup wall superheat
DH = 4A/Pw vl specific volume of the liquid at saturation
hfg latent heat temperature
m_ liquid mass flow rate vv specific volume of the vapor at saturation
Pw Wetted perimeter of a microchannel temperature
q heat rate
q* dimensionless heat rate Greek symbols
rc, cavity radius of an effective nucleation site d thermal boundary layer thickness
T0 liquid entrance temperature r liquid surface tension
T1 bulk liquid temperature

However, the phase change phenomena in micro- 2. Experimental system

channels are relatively poorly understood due to the lack
of experimental studies, especially the visualization of Thermally isolating the microchannel is important to
the two-phase flow conditions. Jiang et al. [8] studied minimize experimental uncertainties. In our previous
water-vapor phase change in 26 and 53 lm V-shaped sil- study, we have successfully carried out heat transfer
icon microchannels. They observed bubble nucleation in experiments in a single microchannel fabricated on a
large manifolds at the entrance to the microchannels. thermally isolated silicon beam [14]. On the basis of
Steinke and Kandlikar [9] examined flow boiling pat- this concept, we designed single-channel devices with
terns in 207 lm diameter channels machined into copper 5–10 X/C temperature sensitivity and 0–10 W heating
plates. They suggested conventional flow boiling pat- capability for both steady state and transient measure-
terns also occur in these channels, such as nucleation, ments. The test system allows simultaneous heat flux
bubbly flow, counter flow, churn flow, and dry-out. control, wall temperature measurement, pressure drop
Peng et al. [10] conducted phase change experiments in measurement, as well as optical imaging of the two-
0.6 · 0.7 mm stainless steel channels. They reported that phase flow conditions.
the heat transfer in the single-phase regime was pro-
longed, and fully developed boiling seemed to appear 2.1. Instrumented microchannels
suddenly at unusually high heat fluxes without nucle-
ation or partial boiling process. Kenning and Yan [11] As shown in Fig. 1, the 6.5 · 2 cm microchannel chip
designed a single vertical channel with 2 · 1 mm cross has two mounting pads and a 2 · 0.2 cm freestanding
section and 248 mm length. They recorded bubble beam. On each device, a single rectangular microchannel
growth in the channel and detected pressure pulses with inlet and outlet reservoirs is micromachined into
caused by the acceleration of liquid slugs resulting from the narrow beam. The freestanding beam and the two
confined bubble growth. Peles et al. [12,13] studied the large reservoirs provide thermal isolation to the heated
phase change in 50–200 lm diameter silicon channels. test channel. The microchannels are 2 cm long, with a
They recorded an oscillating evaporation front between hydraulic diameter in the range of 27–171 lm. The chan-
the vapor bubble and the liquid and associated pressure nels are fabricated with deep reactive ion etching
fluctuations. They also proposed that explosive boiling (DRIE) on the front side of the substrate, with seven
might occur at small Peclet numbers, which leads to fine doped silicon temperature sensors and aluminum heat-
bubble growth, blockage of the channel, and vapor ers on the back side. Transparent glass cover slides are
explosion. In the previous research, bubble nucleation used to seal the channels, allowing high-speed photogra-
has been visualized in both larger than 1 mm diame- phy of the internal flow conditions. The design parame-
ter channels and large areas such as manifolds in micro- ters of the device are listed in Table 1.
channel test devices. However, the much smaller
channels, where opposite observations and a unique 2.2. Experimental setup
‘‘explosive boiling’’ are more often reported, are of more
interest to the study of microchannel heat sinks. In this Fig. 2 shows the schematic of the experimental setup.
paper, we focus on how the phase change occurs and The microchannel test device is mounted to a fixture
how the two-phase flow develops in micromachined sil- with internal flow channels and directors. The heaters
icon channels with less than 150 lm hydraulic diameters. and temperature sensors are connected through a
1574 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582

Fig. 2. Schematic of experimental setup.

3. Phase change in flat-wall channels

Fig. 1. Microchannels with integrated heaters and temperature In macroscale channels, the boiling process always
sensors. (a) Front side of the microchannel chip with the begins with bubble growth on the channel walls. The
channel, inlet and outlet reservoirs. (b) Silicon temperature two-phase flow transitions along the flow direction into
sensors and aluminum heater on the back side. four regimes—bubbly flow (liquid flow carrying vapor
bubbles), slug/plug flow (liquid flow with coalesced bub-
surface mounting board attached to the device [14]. De-
bles), annular flow (a fast-moving vapor core wrapped by
ionized (DI) water is pumped into the channel with a
a thin, slowly-moving liquid layer), and mist flow (a mix-
constant flow rate syringe pump. By applying a voltage
ture of vapor and tiny liquid droplets moving at the same
to the on-chip aluminum heater, the microchannel is
speed) [15]. Among all the two-phase flow regimes, annu-
uniformly heated. The channel wall temperatures are
lar flow has the highest heat transfer coefficients due to
measured through the resistance change of doped silicon
thin film evaporation [15]. In flat-wall silicon microchan-
temperature sensors, which are evenly distributed along
nels, we observed two different nucleation mechanisms in
the channel in seven locations. The temperature sensors
channels larger than 100 lm and smaller than 50 lm in
are 300–350 lm away from the liquid-channel interface
hydraulic diameter. The two-phase flow directly develops
and the temperature gradient in the bottom wall is
into annular flow or mist flow regime, without a clear
approximately 0.3 C with the maximum power used
transition from bubbly or slug/plug flows. Annular flows
in our experiments. The temperature-resistance correla-
in microchannels also achieve higher heat transfer rates
tion for each temperature sensor is calibrated before
than other flow regimes.
experiments. These temperature sensors have 5–10 X/
C temperature sensitivity in the range of 20–150 C,
with ±0.5 C accuracy. In addition, an MSP-400 pres- 3.1. Phase change visualization
sure transducer is used to measure the global pressure
drop along the channel. A SONY 2735 CCD camera Using a SONY 2735 CCD camera with controlled
(30 frames/s, 1/100,000 s maximum shutter speed) re- shutter speed, we recorded the nucleation process in
cords the two-phase flow conditions simultaneously. channels with various hydraulic diameters. In all of these
The heating power, channel wall temperature, and the channels, bubbly and slug/plug flow regimes are absent,
pressure signals are collected and recorded with a 16- because a single bubble can quickly grow to the size of
bit A/D system. the channel diameter and suppress the two regimes.

Table 1
Device structural dimensions
Structure Dimension
Chip 6.5 cm long, 2 cm wide, silicon substrate 480 lm thick, glass slide 520 lm thick
Beam 2 cm long, 2 mm wide
Reservoirs 1 cm long, 600 lm wide, 250 lm deep
Channels 10–150 lm wide, 10–200 lm deep, 2 cm long
Temperature sensor (back side) 150 lm long, 110 lm wide, 1 lm junction depth, 8 kX resistance
Heater (back side) 2 cm long, 50 lm wide (10 lines), 2 lm thick, 50 X resistance
 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582 1575

the amount of wall superheat. In our experiment, the

flow rates are selected so that the global pressure drop
is less than 35 kPa, and the saturation temperature of
water is calculated from the measured pressure drop just
before the onset of boiling.
With typical bubble nucleation in larger channels, the
annular flow eventually develops with a measurably
higher heat transfer rate. Fig. 4 shows the wall tempera-
ture and global pressure measurement with increasing
input heat rate in a 113 lm diameter channel. The wall
temperature increases linearly with increasing heat
rate in the single-phase region. Typical bubble nucle-
ation is recorded with less than 5 C wall superheat. A
steady annular flow then develops, resulting in a long
temperature plateau in the two-phase region. Between
the single-phase and two-phase regions, a secondary pla-
teau appears due to liquid out-gassing. While bubbles
are seen in out-gassing period, the global pressure drop
does not increase because phase change is not involved.
The onset of boiling in the plot is defined as the
point when a sharp increase in global pressure drop is
recorded. The heat loss in this experiment was approxi-
mately 42%, estimated from a thermal resistance model
for the experimental system [14]. Subtracting the heat
loss, the thermal resistances are 95.2 C/W in the sin-
gle-phase region and 15.6 C/W in the fully developed
Fig. 3. Bubble nucleation mechanisms and two-phase flow two-phase region. Therefore, the heat transfer rate in-
regimes. (a) Bubble nucleation in a 250 lm wide, 130 lm deep creases about 5 times due to the two-phase annular flow.
(171 lm hydraulic diameter) channel, with 0.5 ml/min DI water
In less than 50 lm diameter channels, where eruption
flow rate, 94–96 C wall temperature; (b) Vapor eruption in a
boiling happens, mist flow usually develops and there is
50 lm wide, 40 lm deep (44 lm hydraulic diameter) channel,
with 0.02 ml/min DI water flow rate, 120 C wall temperature. no apparent temperature plateau in the two-phase re-
gion. Fig. 5 is the wall temperature and global pressure
measurement with increasing input heat rate in the
We also observed two distinct bubble nucleation mech- 44 lm channel shown in Fig. 3b. In the single-phase re-
anisms—typical nucleation and eruption boiling. In gion, the wall temperature also increases linearly with
Fig. 3a, a bubble forms and departs in 0.2 s in a
171 lm diameter channel as a typical nucleation process.
120 50
The bubble becomes a transient annular flow upon Single- Two-phase 45
departure, and a steady annular flow develops when 100 phase
40
Wall temperature (C)

more bubbles are departing. In a 44 lm diameter chan- 16.65mm

35
Pressure (kPa)

nel shown in Fig. 3b, eruption boiling occurs with no 80

detectable bubble nucleation period. The vapor eruption 8.18mm 30
process completes within 0.033 s, and two-phase flow 60
19.48mm
25
appears as a single-phase liquid and a two-phase mist 20
separated by a fast oscillating interface in the longitudi- 40
15
2.53mm
nal direction. The two nucleation mechanisms are repre- 10
20 Pressure
sentative for larger than 100 lm diameter (typical 5
nucleation) and smaller than 50 lm diameter (eruption
0 0
boiling) channels respectively. Channels with 50– 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
100 lm diameters exhibit transitional behaviors, show- Heat rate (W)
ing both typical nucleation and eruption mechanisms.
Fig. 4. Wall temperature change from four temperature sensors
and global pressure drop as a function of input heat rate in
3.2. Wall temperature measurement a 100 lm wide, 130 lm deep (113 lm hydraulic diameter)
channel at 0.1 ml/min DI water flow rate. The numbers
The nucleation mechanism further determines the represent the distance between the sensor and the entrance of
two-phase flow regime in the microchannel as well as the microchannel.
1576 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582

180 100 9200

160 90 9100
16.65mm
Wall temperature (C)

140 9000

Resistance (Ohm)
8.18mm 80

Pressure (kPa)
120 8900 8.18mm
19.48mm 70
100 8800
16.65mm 60
80 8700
50 19.48mm
60 8600 2.53mm
2.53mm 40
40 8500
Pressure
20 30 8400
Single-phase Two-phase Single-phase Nucleation Two-phase
0 20 8300
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0 5 10 15 20 25 30
Heat rate (W) (a) Time (s)

Fig. 5. Wall temperature change from four temperature sensors 19.48 mm

9025
and global pressure drop as a function of input heat rate in a
9020
50 lm wide, 40 lm deep (44 lm hydraulic diameter) channel at

R (Ohm)
9015
0.02 ml/min DI water flow rate. The numbers represent the 9010
distance between the sensor and the entrance of the micro- 9005
channel. The short ‘‘plateaus’’ are artificial lines between input 9000
heat rate points, not measured temperature plateaus. 8995
0 0.2 0.4 0.6 0.8 1

2.53 mm
the heat rate. However, the single-phase region is ex- 8705
tended to nearly 120 C when vapor eruption suddenly 8700
R (Ohm)

occurs. With the same DI water, there is no apparent 8695

out-gassing period as in the 113 lm diameter channel. 8690
The wall temperature continues to increase with increas- 8685
ing heat rate in the two-phase region. More than 20 C 0 0.2 0.4 0.6 0.8 1
wall superheat is characteristic to the eruption mecha- (b) T (s)
nism. Because of the large amount of superheat, the high Fig. 6. High speed time traces of the sensor resistance change
wall temperature quickly turns single-phase water flow during the phase change process, showing bubble nucleation
into a mist or even pure vapor flow in these small-diam- induced pressure fluctuations in a 100 lm wide, 130 lm deep
eter channels. Mist/vapor flows are known to have poor (113 lm hydraulic diameter) channel with 0.1 ml/min DI water
thermal transport properties, and a temperature plateau flow rate. The numbers represent the distance between the
is usually not expected. With 10% heat loss in the mea- sensor and the entrance of the microchannel. (a) Transient
surement, the thermal resistance in both single- and resistance change after a constant input heat rate is applied at
two-phase regions is approximately 600 C/W. t = 0.5 s. The bubble nucleation begins at t = 11 s, and a steady
annular flow develops at t = 18 s. (b) Highly correlated resis-
tance fluctuations at the inlet and outlet sensor locations when
3.3. Bubble induced pressure fluctuations
bubbles forming and departing in the channel.

Because of the small internal volume of microchan-

nels, individual bubbles can affect important parameters ble nucleation are significantly higher than what have
such as local pressure. Fig. 6a shows high-speed time been measured in macroscale channels.
traces of temperature sensor resistances when a constant In this experiment, water enters the channel at room
heat power is applied to a 113 lm diameter channel. temperature, therefore boiling only occurs in the middle
When bubbles begin to form, there are noticeable fluctu- of the channel. However, as shown in Fig. 6b, the signal
ations in all locations. The fluctuation frequency is al- fluctuation is highly correlated along the entire channel,
ways equal to the bubble departing frequency, which including the entrance of the channel which is under sin-
has been confirmed from simultaneous video recording. gle-phase flow. The fluctuations are determined to be
The fluctuations begin with 3–5 Hz frequency, and in- pressure signals rather than temperature signals because
crease to 30–40 Hz within a few seconds, indicating only pressure waves are fast enough to cause correlated
more bubble generation and the development of a steady signals along the length of the channel. Since doped sil-
annular flow. These signals are found to be transient icon resistors are also sensitive to strain, the temperature
pressure fluctuations on the order of 69–138 kPa, caused sensors can respond to large transient pressure waves.
by individual bubbles forming and departing in the Calibration shows that the sensors have 10 X/C temper-
channel. Transient pressure fluctuations induced by bub- ature sensitivity and 0.145 to 0.29 X/kPa pressure
 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582 1577

sensitivity. The two sensitivities function in different fre- sites and help initiate the phase change. HsuÕs model
quency domains. Temperature measurements are per- introduces the concept of waiting period, the time re-
formed in steady state with less than 1 Hz sampling quired for the liquid to be heated to a higher tempera-
frequency. With 69 kPa to 1 C equivalence in the two ture than that of a bubble embryo, so that the bubble
measurements, the maximum pressure fluctuations only can absorb the heat from surrounding liquid and con-
cause about 1–2 C uncertainty in the steady state tem- tinue to grow [17]. Only the cavities with a finite waiting
perature. The temperature sensitivity decreases quickly period, referred to as effective sites, can potentially be-
for signals above 1 Hz due to the necessary thermal come active nucleation sites. He further provided the
diffusion time in the system. Therefore, the AC (Alter- relationship between the wall superheat and the size
nating Current) signals on top of the DC (Direct Cur- range (rc,max, rc,min) of effective sites:
rent) temperatures are considered transient pressure 2
components. d 4 T sat  T 1
rc;max ; rc;min ¼ 1
2C 1 Tw  T1
s ffi3
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
4. Phase change in enhanced-wall channels
T sat  T 1 8rT sat vv C 3 5
 1 
Eruption boiling is not commonly seen in macroscale Tw  T1 hfg dDðT w  T 1 Þ
channels, but has been reported by several research ð2Þ
groups in 50–700 lm diameter channels [10,13]. In some
previous experiments, this phenomenon has been as- where C1 and C3 are constants related to the shape of
sumed and/or confirmed to be related to channel dimen- the cavity and the contact angle of the bubble embryo.
sions. In this study, we believe that eruption boiling is For simplicity, Hsu [17] assumes the bubble embryo
caused by the lack of active nucleation sites in micro- has the same diameter as the cavity mouth, and obtains
channels. Our further experiments indicate that the boil- the values as 2 and 1.6 respectively. Eq. (2) indicates that
ing mechanism is determined by the wall surface with a given type of liquid and wall superheat, only a
condition rather than the internal space of a given range of nucleation sites can be activated. There-
microchannel. fore, given the cavity sizes in a microchannel, the
amount of wall superheat must satisfy the equation in
4.1. Hypotheses order to initiate the phase change.
Experiments were designed to compare the two
One existing hypothesis about the eruption boiling is hypotheses. We used surfactant to reduce the liquid sur-
that a vapor bubble must overcome extra pressure im- face tension to study whether the wall superheat can be
posed by channel walls, and bubble nucleation will be
suppressed if the internal space of a microchannel is
smaller than an ‘‘evaporating space’’ [16]. Using Clau-
sius–Clapeyron equation and a correction factor for
the space induced pressure, Peng et al. [16] suggested
that the amount of wall superheat DTsup in a microchan-
nel is determined by
4CT sat ðvv  vl Þr
DT sup ¼ T w  T sat P ð1Þ
hfg DH
where C is an empirical constant which is proposed to be
approximately 280 on the basis of earlier experiments.
This equation also indicates that the degree of wall
superheat is an exclusive function of surface tension r
with a given channel dimension and the type of liquid.
On the other hand, in the flat-wall silicon channels,
we found that bubbles only grew on the side walls,
and never on the bottom of the channel. It is known that
the DRIE process creates rough side walls to achieve a
Fig. 7. The notches in the side walls and the cavities in the
highly anisotropic etch. Therefore we postulate that bottom as surface enhancement. They are designed to trap air
the eruption boiling is caused by the wall surface condi- to facilitate nucleation process during the phase change. (a)
tion. In traditional pool boiling theories, the wall surface Notches with 4 m opening, 20 lm depth into the side walls. (b)
condition plays a key role in heterogeneous nucleation 4–8 lm diameter, 150–250 lm deep cavities in the bottom of
because gas-trapping cavities first become nucleation the channel.
1578 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582

reduced accordingly. We also fabricated enhanced-wall 4.3. Wall temperature measurement

devices with 4–8 lm notches and cavities defined in
channel walls, as shown in Fig. 7. The experiments Fig. 9 shows the repeated wall temperature measure-
focus on the boiling mechanism and the wall super- ment in the 44 lm channel with low-surface tension DI
heat in channels with less than 100 lm hydraulic water. The trend and shape of the curves are very similar
diameters. to those in Fig. 5, measured with regular DI water. We
estimate the surface tension by taking pictures of a li-
quid drop on a silicon surface and measuring the contact
4.2. Phase change visualization
angle. Estimation shows that the DI water surface ten-
sion is reduced by approximately 30% in the 200 ppm
We first added surfactant into water (Triton X-100,
Triton X-100 solution. However, vapor eruption oc-
200–10,000 ppm weight concentration) to reduce the
curred in both experiments with more than 20 C wall
liquid surface tension and repeated the phase change
superheat without apparent reduction for the low-
experiment in the 44 lm diameter channel shown in
surface tension liquid. Repeated experiments in a few
Fig. 3b. However, the same type of eruption boiling
small-diameter channels with different flow rates yielded
was observed.
the same result. According to Eq. (1), the amount of wall
We then conducted phase change experiments in four
superheat is proportional to the liquid surface tension
enhanced-wall channels with 28, 47, 61, and 72.5 lm
when the channel size is fixed. Therefore eruption boil-
hydraulic diameters. In all four channels, we observed
ing in our experiments is not directly caused by liquid
typical nucleate boiling and confirmed that the cavities
surface tension.
induced bubble nucleation. Also the increased density
In the phase change experiments with enhanced-wall
of nucleation sites help to induce a steady annular flow.
channels, wall temperature measurement indicates that
Fig. 8a shows bubbles generating and departing from
bubble nucleation occurred with less than 5 C wall
the cavities in the bottom of the channel. The sequence
superheat, as shown in the wall temperature plots in
of images in Fig. 8b shows a steady annular flow with
Figs. 10 and 11. When the annular flow is induced in
varying liquid film thickness.
the 72.5 lm diameter channel shown in Fig. 10, a tem-
perature plateau is measured, similar as in Fig. 4. Con-
sidering approximately 40% heat loss, here the thermal
resistance reduces from 105 C/W in the single-phase re-
gion to 22 C/W in the two-phase region. We also found
that in less than 50 lm diameter channels, bubble nucle-
ation causes strong instabilities even if the wall super-
heat is significantly reduced. As a result, the annular

180

160
Wall temperature (C)

140
8.18mm
19.48mm
120

100 16.65mm
80

60

40 2.53mm

20
Single-phase Two-phase
0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Heat rate (W)

Fig. 9. Wall temperature change from four temperature

Fig. 8. Bubble nucleation and annular flow patterns in a sensors as a function of heat rate in a 50 lm wide, 40 lm deep
120 lm wide, 52 lm deep (72.5 lm hydraulic diameter), (44 lm hydraulic diameter) channel at 0.02 ml/min flow rate of
enhanced-wall channel. (a) Bubble nucleation from cavities in 200 ppm TritonX-100 DI water solution. The numbers repre-
the bottom of the channel. (b) Annular flow in the channel with sent the distance between the sensor and the entrance of the
varying liquid film thickness. microchannel.
 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582 1579

160 Table 2
Experimental parameters for representative microchannels
140
8.18mm Parameters Ch1 Ch2 Ch3 Ch4
Wall temperature (C)

120
19.48mm Channel 100 · 130 50 · 40 120 · 50 50 · 44
100 dimension
16.65mm (lm · lm)
80 Channel wall Flat Flat Enhanced Enhanced
2.53mm Hydraulic 113 44 72.5 47
60
diameter
40 (lm)
Mass flow 0.1 0.02 0.1 0.02
20
Single-phase Two-phase rate (ml/min)
0 Flow velocity 0.128 0.167 0.278 0.152
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 (m/s)
Heat rate (W) Reynolds 15.4 7.9 20.8 7.5
number
Fig. 10. Wall temperature change from four temperature
Weber number 0.026 0.017 0.078 0.015
sensors as a function of the heat rate in a 120 lm wide, 52 m
Capillary 1.68e3 2.19e3 3.65e3 1.99e3
deep (72.5 lm hydraulic diameter) channel at 0.1 ml/min DI
number
water flow rate. The numbers represent the distance between the
Bond number 1.74e3 2.64e4 7.16e4 3.01e4
sensor and the entrance of the microchannel.

140 180

120 160
8.18mm
Eruption
Wall temperature (C)

16.65mm Ch2
Wall temperature (C)

140
100
120 Ch4
80
100 Nucleation Ch3
19.48mm
60 80 Ch1
40 60
2.53mm
40
20
Single-phase Two-phase 20
0
0.0 0.5 1.0 1.5 2.0 2.5 0
0 0.5 1 1.5 2 2.5 3 3.5
Heat rate (W)
Dimensionless heat rate q*
Fig. 11. Wall temperature change from four temperature
Fig. 12. Wall temperature change as a function of dimension-
sensors as a function of the heat rate in a 50 lm wide, 44 m
less heat rate from flat-wall and enhanced-wall channels. Ch1:
deep (47 lm hydraulic diameter) channel at 0.02 ml/min DI
113 lm hydraulic diameter, flat-wall channel, 0.1 ml/min DI
water flow rate. The numbers represent the distance between the
water flow rate. Ch2: 44 lm hydraulic diameter, flat-wall
sensor and the entrance of the microchannel.
channel, 0.02 ml/min DI water flow rate. Ch3: 72.5 lm
hydraulic diameter, enhanced-wall channel, 0.1 ml/min DI
flow changes to the mist flow regime well before antici- water flow rate. Ch4: 47 lm hydraulic diameter, enhanced-wall
pated in macroscale channels, which explains the ab- channel, 0.02 ml/min DI water flow rate.
sence of a temperature plateau in Fig. 11.
proven the primary reason for wall superheat in erup-
tion boiling. Wall temperature measurements performed
5. Discussion
in these channels are compared in Fig. 12. Temperature
measurements in the same location (16.65 mm from the
Table 2 summarizes the experimental parameters of
entrance of the channel) in all four channels are plotted
four microchannels discussed in this paper. Microchan-
in one graph with normalized input heat rate to compen-
nel flows usually have much lower flow velocities and
sate the difference in flow rates. The dimensionless heat
Reynolds numbers than those in macroscale channels.
rate q* is defined as
Calculation of the Weber number, Capillary number,
q
and Bond number indicates that surface tension is the q ¼ ð3Þ
dominant force in our experiments, although it is not _ p ðT sat  T 0 Þ
mc
1580 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582

where the heat rate q is divided by the sensible heat of a For the enhanced-wall channels, using 4–8 lm cavity
given mass flow rate m. _ p ðT sat  T 0 Þ
_ The sensible heat mc size in a 75 lm and a 45 lm diameter channel, HsuÕs
is the maximum amount of heat that can be absorbed by model shows 1.5–2 C and 2.5–3 C wall superheat
the liquid before the phase change takes place. The the- respectively, also in very good agreement with the tem-
oretical saturation temperature (boiling point) of DI perature plots in Figs. 10 and 11 (also Ch3 and Ch4 in
water under the experimental conditions ranges from Fig. 12). In addition, although the bubbles prefer to
101 to 103 C. Comparing the four boiling curves, erup- grow from the side walls in flat-wall channels (Fig. 3a),
tion boiling with large amounts of wall superheat is they move to the bottom of the enhanced-wall channels
eliminated by adding cavities to the channel walls, and (Fig. 8a) and are never observed in the side wall notches.
the enhanced surfaces help to establish a steady annular These notches are originally designed as a comparison to
flow regime in larger than 50 lm diameter channels. the cavities. However, because the notches have the
The wall temperature measurements qualitatively same depth as the channel, the actual opening size is tens
support that the nucleation sites affect the wall superheat of microns. From Fig. 13, this size range falls in the inef-
as well as the nucleation mechanism. This theory is also fective region (sizes with an indefinite waiting period),
supported by KandlikarÕs recent study on the influence which well explains the nucleation preference to the bot-
of channel hydraulic diameter on flow boiling heat tom cavities.
transfer mechanisms [18]. Using HsuÕs model, the re- In summary, our experimental results support the
quired wall superheat can be estimated from the cavity nucleation site hypothesis, and large amounts of wall
size. In microchannels with thermally fully developed superheat associated with eruption boiling can still be
flows (test channels have nearly-constant heat flux explained with HsuÕs model. Both typical nucleation
boundary condition [14], therefore thermally fully devel- and eruption boiling are forms of heterogeneous nucle-
oped flow will appear), we use one half of the channel ation where vapor forms on a liquid–solid interface,
hydraulic diameter as thermal boundary layer thickness and the amount of wall superheat is primarily deter-
(d = DH/2), and assume the liquid is saturated mined by the size and number of active nucleation sites.
(T1 = Tsat). In Fig. 13, four plots are generated for the Eruption boiling tends to occur when the wall superheat
range of effective nucleation sites (the region within the exceeds 5–10 C due to fast evaporation in a very limited
parabolas) with 1–100 C wall superheat for a 11 lm, internal volume. Fig. 14 is a proposed boiling mecha-
45 lm, 75 lm, and a 110 lm diameter channel. nism and regime map for sub-150 lm diameter micro-
In the flat-wall silicon channels, we found that the channels with less than 0.5 m/s water flow velocities,
DRIE process resulted in fine features ranging from based on our experiments with flat- and enhanced-wall
0.1 to 0.4 lm in the side walls, with a few defects on channels. Here we assume the condition for eruption
the order of 2–5 lm in larger channels. Assuming 2– boiling is D Tsup P 10 C, also based on our experimen-
5 lm cavity size in a 110 lm channel and 0.1–0.4 lm tal results. Given the channel hydraulic diameter and
cavity size in a 45 lm diameter channel, HsuÕs model wall superheat, the range of effective nucleation sizes
predicts 1.5–3.5 C and 18–70 C wall superheat respec- (rc,min, rc,max) can be calculated from Eq. (2), under
tively, which is in good agreement with our measure- assumptions d = DH/2 and T1 = Tsat. In Fig. 14, the
ments in Figs. 4 and 5 (also Ch1 and Ch2 in Fig. 12).

100.0
100
DH=75µm Eruption/Mist
Cavity size (um)

10 10.0
Cavity size (um)

DH=45µm
Nucleation/Annular
DH=110µm
Nucleation/
1 Effective region
Mist
1.0
DH=11µm
0.1
Eruption/Mist
0.1
0.01 0 50 100 150
1 10 100 Hydraulic diameter (um)
Wall superheat (C)
Fig. 14. Boiling mechanism and two-phase flow regime map for
Fig. 13. The size range of effective nucleation sites as a function sub-150 lm diameter microchannels with less than 0.5 m/s
of wall superheat and channel dimension from HsuÕs model. water flow velocity.
 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582 1581

calculated rc,min and rc,max are plotted as a function of annular flow for microchannel heat sink applications,
DH. The region within the parabola defines the micro- the channel wall must have enough gas-trapping cavities
channels where typical nucleation tends to occur to ensure typical nucleation. Fourth, the mist flow tends
(DTit sup < 10 C). The area is further divided into mist to dominate in sub-50 lm diameter channels even with
flow and annular flow regions according to our observa- wall surface enhancements. The phase change mecha-
tion that mist flow tends to dominant in less than 50 lm nism and regime map in Fig. 14 provides a reference
diameter channels. In microchannels outside of the for design of microchannel heat sinks. These findings
parabola region, eruption tends to occur and mist flow will help to better understand the flow behavior in
is the only dominant flow regime as the result of erup- microchannels, and a new generation of heat sinks for
tion boiling. microprocessors may be expected in the near future.
The ‘‘evaporating space’’ hypothesis was not verified
in our experiments, and we believe this theory may apply
to ideally smooth channel walls, or below a certain size Acknowledgments
threshold in microchannels. The ‘‘evaporating space’’
hypothesis assumes the internal space of the channel as The work was supported by DARPA HERETIC
a single nucleation site, while we treat the defects or cav- program under DARPA Contract F33615-99-C-1442,
ities in the channel walls as nucleation sites. As shown in Stanford Graduate Fellowships (Lian Zhang), and
Fig. 13, the minimum amount of wall superheat is a National Defense Science and Engineering Graduate
function of the thermal boundary layer thickness (chan- Fellowship (Evelyn N. Wang). The devices were fab-
nel diameter), and the actual wall superheat is primarily ricated at Stanford Nanofabrication Facility, funded
determined by the cavity size. For example, for channels by the National Science Foundation under award num-
larger than 11 lm in diameter, the wall superheat can be ber ECS-9731294. We thank Prof. Juan Santiago at
reduced to less than 10 C by creating appropriate cavi- Stanford University and James Maveety at Intel Corpo-
ties in the channel wall. However, for channels smaller ration for their valuable discussions, and Shilajeet
than 11 lm in diameter, the minimum amount of wall Benerjee for his assistance in building the experimental
superheat will be 10 C regardless of cavity sizes and system.
eruption boiling may become the only mechanism for
phase change. We expect an interesting transition in
boiling phenomena in less than 11 lm diameter
References
channels.
[1] G.E. Moore, Cramming more components onto integrated
circuits, Proc. IEEE 86 (1) (1998) 82–85, reprinted from
6. Conclusions Electronics 38 (8) (1965) 114–117.
[2] Intel microprocessor quick reference guide, http://www.
We found that two-phase forced convective theories intel.com/pressroom/kits/quickreffam.htm.
are valid down to 27 lm diameter silicon microchannels, [3] International Technology Roadmap for Semiconductors
including both heat transfer [14] and flow behavior. 2003 update, http://public.itrs.net/Files/2003ITRS/Home2003.
However, due to the scaling of the channel dimensions, htm.
factors that can be neglected in macroscale channels be- [4] D.B. Tuckerman, R.F.W. Pease, High-performance heat
sinking for VLSI, IEEE Electron Dev. Lett. EDL 2 (5)
come dominant, such as the condition of individual
(1981) 126–129.
nucleation sites as well as the formation of individual
[5] C.D. Patel, Enabling pumped liquid loop cooling: justifi-
bubbles. This effect, although not a violation of two- cation and the key technology and cost barriers, in:
phase theories, leads to some unique phenomena in Proceedings of International Conference on High-Density
two-phase flows in microchannels. First, bubbly and Interconnect and Systems Packaging, IMAPS, Reston,
slug flows that exist in macroscale channels are often ab- VA, 2000, pp. 145–152.
sent in microchannels. This is beneficial to microchannel [6] L. Jiang, J. Mikkelsen, J.-M. Koo, D. Huber, S. Yao, L.
heat sinks because the desired annular flow pattern can Zhang, P. Zhou, J.G. Maveety, R. Prasher, J.G. Santiago,
be quickly established. Second, vapor bubbles can cause T.W. Kenny, K.E. Goodson, Closed-loop electroosmotic
69–138 kPa transient pressure pulses at the onset of boil- microchannel cooling system for VLSI circuits, IEEE
Trans. Components Packaging Technol. 25 (3) (2002)
ing. These pressure pulses may damage the device, and
347–355.
are expected to affect local boiling point and equilibrium
[7] S. Zeng, C.-H. Chen, J.C. Mikkelsen, J.G. Santiago,
conditions. Third, the phase change mechanism (typical Fabrication and characterization of electrokinetic micro
nucleation and eruption boiling) is closely associated pumps, in: Proceedings of ASME ITHERM, IEEE,
with wall surface conditions, and the two-phase flow re- Piscataway, NJ, 2000, pp. 31–36.
gime has a strong dependence on the phase change [8] L. Jiang, M. Wong, Y. Zohar, Phase change in micro-
mechanism. In order to induce and maintain a steady channel heat sink under forced convection boiling,
1582 L. Zhang et al. / International Journal of Heat and Mass Transfer 48 (2005) 1572–1582

in: Proceedings of IEEE MEMS Workshop, IEEE, Piscat- tional Conference on Microchannels and Minichannels,
away, NJ, 2000, pp. 397–402. ASME, New York, NY, 2003, pp. 559–566.
[9] M.E. Steinke, S.G. Kandlikar, Flow boiling and pressure [14] L. Zhang, J.-M. Koo, L. Jiang, M. Asheghi, K.E.
drop in parallel flow microchannels, in: 1st International Goodson, J.G. Santiago, T.W. Kenny, Measurements
Conference on Microchannels and Minichannels, ASME, and modeling of two-phase flow in microchannels with
New York, NY, 2003, pp. 567–579. nearly-constant heat flux boundary conditions, J. MEMS
[10] X.F. Peng, B.X. Wang, Forced convection and flow boiling 11 (1) (2002) 12–19.
heat transfer for liquid flowing through microchannels, Int. [15] V. Carey, Liquid–Vapor Phase-Change Phenomena, Hemi-
J. Heat Mass Transfer 36 (14) (1993) 3421–3427. sphere Publishing Corporation, Washington DC, 1992, pp.
[11] D.B.R. Kenning, Y. Yan, Saturated flow boiling of water 483–485.
in a narrow channel: experimental investigation of local [16] X.F. Peng, H.Y. Hu, B.X. Wang, Boiling nucleation
phenomena, IChemE Transactions A, Chem. Eng. Res. during liquid flow in microchannels, Int. J. Heat Mass
Des. 79 (A4) (2001) 425–436. Transfer 41 (1) (1998) 101–106.
[12] Y.P. Peles, L.P. Yarin, G. Hetsroni, Evaporating two- [17] Y.Y. Hsu, On the size range of active nucleation cavities on
phase flow mechanism in micro-channels, Proc. SPIE 3680 a heating surface, ASME J. Heat Transfer (Aug.) (1962)
(1999) 226–229. 207–216.
[13] Y. Peles, Two-phase boiling flow in microchannels-insta- [18] S.G. Kandlikar, Heat transfer mechanisms during flow
bilities issues and flow regime mapping, in: 1st Interna- boiling in microchannels, J. Heat Transfer 126 (2004) 8–16.