International Journal of Heat and Mass Transfer 45 (2002) 4761–4771

www.elsevier.com/locate/ijhmt

High-speed visualization of boiling from an enhanced structure

C. Ramaswamy a, Y. Joshi b,*
, W. Nakayama c, W.B. Johnson d

a
IBM Corporation, Hopewell Junction, NY 12533, USA
b
G.W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA
c
ThermTech International, Kanagawa, 255-0004, Japan
d
Laboratory for Physical Sciences, College Park, MD 20740, USA
Received 14 January 2002; received in revised form 16 May 2002

Abstract
Using high-speed photography (1500 frames/s) bubble growth data on microporous structures immersed in a pool of
dielectric coolant (FC-72) were obtained. Wafer dicing and wet etching was used to fabricate a net of interconnected
microchannels on a 10 mm
10 mm piece of silicon wafer. The resultant structure has pores that communicate the
interior of microchannels to the liquid pool. The pore diameter was varied in a range 0.12–0.20 mm and the pore pitch
in 0.7–1.4 mm. The data were collected maintaining the system pressure at one atmosphere and increasing the wall
superheat up to 12 K.
Among the geometrical parameters, the pore diameter was found to be most influential on the bubble departure
diameter. The findings about the bubble growth rate, the bubbling frequency, and the bubble site density were largely in
accord with the previously reported data. However, the coverage of wider ranges of wall superheat and the structural
parameters in the present study revealed new bubble characteristics that was used in implementing an analytical model
for boiling heat transfer on the porous structure.
Ó 2002 Elsevier Science Ltd. All rights reserved.

1. Introduction long, with distilled water as the working fluid. Using a

high-speed movie camera (6400 frames/s) the departure
The mechanism of pool boiling is a complex one with diameter, frequency and nucleation site density was
several factors contributing to the high heat transfer measured. The data showed that latent heat was the
rates. These include latent heat transfer, natural con- dominant mechanism of heat transfer in the nucleate
vection, microconvection and Marangoni flow. Several boiling regime. The study also showed that the bubble
investigators have used high-speed visualization as a departure diameter, frequency and nucleation site den-
tool to quantify the relative contribution of the above sity, for a particular heat flux value, could be repre-
mentioned components. For example, Han and Griffith sented by statistical functions. Recently, Ammerman
[1] used high-speed photography (2400 frames/s) to and You [3] developed a single-photo/laser doppler
study boiling from a cylindrical copper block (30.2 mm anemometry (LDA) technique to investigate the volu-
diameter) plated with a 0.125 mm layer of gold, with metric vapor flow from a heated platinum wire in FC-72.
distilled water as the working fluid. They measured the A random set of five frames was chosen at each heat flux
bubble departure diameters and frequency of bubble level and the volumetric flow rate determined within an
generation to predict the latent heat flux. Paul and area of interest. Combined with the measurements of the
Abdel-Khalik [2] carried out visualization of boiling LDA, the average bubble diameter was computed. Their
from a platinum wire, 0.3 mm in diameter and 30 mm analysis confirmed that latent heat transport was the
dominant mechanism in nucleate boiling.
Investigators have also used high-speed visualization
*
Corresponding author. Tel.: +1-404-385-2810; fax: +1-404- for qualitative analysis of the boiling mechanism. Sun
894-8496. and Lienhard [4] carried out high-speed visualization,
E-mail address: [email protected] (Y. Joshi). using a movie camera, on nichrome wires 101.6 mm in
0017-9310/02/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 1 7 - 9 3 1 0 ( 0 2 ) 0 0 1 9 6 - 5
4762 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771

Nomenclature

B width of the enhanced structure (mm) t time (s)

D instantaneous bubble diameter (mm) Dtg growth period for an individual bubble (s)
Dd bubble departure diameter (mm) Dtw waiting time for bubble initiation (s)
f frequency (Hz) Tsat saturation temperature of FC-72 (°C)
g acceleration due to gravity (9.81 m/s2 ) Twall temperature at the bottom of the enhanced
Ht channel depth (mm) structures (°C)
Hs height of the enhanced structure (mm) DTsup wall superheat (Twall  Tsat ), (°C)
L length of the enhanced structure (mm) Wt channel width––also pore size (mm)
Pt channel pitch––also pore pitch (mm)

length and diameter in the range 0.125–2 mm. Four There is sufficient evidence in literature demonstrating
different organic liquids: acetone, methanol, benzene the very high heat transfer rates (100 W/cm2 ) achiev-
and isopropanol were used. The emphasis of the study able with enhanced surfaces (see Webb [6]). However,
was to understand the vapor and liquid interaction at studies on understanding the mechanism of boiling from
critical heat flux. A correlation was derived to predict these surfaces are limited. Nakayama et al. [7–9] were
the thickness of the vapor blanket layer for this state. among the first to carry out visualization of the boiling
Katto and Yakoya [5] studied the boiling of saturated process from a structured surface. The surface consisted
water at one atmosphere from a horizontal, 10 mm di- of a rectangular channel covered with a thin sheet that
ameter, heated disc. They found that the vapor mass had pores at a regular pitch (Fig. 1) and the entire
above the heated surface can be classified into three structure was immersed in R-11. They used a laser probe
categories depending on the influence between consec- to measure the frequency of bubble formation and the
utive bubbles. In the first category, consecutive vapor population density of the bubbles. They found that
masses are independent of each other. In the second, the liquid in the channel evaporated to form vapor regions,
growing vapor mass is almost attached to the preceding which were then ejected as bubbles from the pores. As
one and is much smaller in size. The final category the bubbles ejected, liquid filled into the channels
consists of vapor masses that are separated from the through the pores and spread axially along thin menisci
preceding one but grow under the influence of the flow along the sharp corners of the channel. With addition of
generated by the preceding one and are almost columnar heat, the frequency of bubbles and the number of bub-
in shape. bles increased, but the basic phenomenon remained the
The above mentioned studies and similar ones in same. An analytical model was formulated to predict a
literature are mostly for plain polished surfaces only. few of these parameters.

Fig. 1. Cross-sectional view of structures used by authors for boiling visualization.

 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771 4763

Fig. 2. Enhanced structures employed in this study.

Arshad and Thome [10] conducted flow visualization 2. Enhanced structures

from similar surfaces (Fig. 1) to understand the mech-
anism of boiling inside the channels. This study con- The enhanced structure, which was the focus of this
firmed that the primary mechanism of heat transfer study, is shown in Fig. 2 and its construction is similar to
inside the channels was evaporation of the thin liquid the one studied by Nakayama et al. [12]. The structures
menisci in sharp corners. The study also showed that the have an array of rectangular channels cut on either side
vapor initiation was from one of the sharp corners and of a 10 mm
10 mm silicon substrate and aligned at an
quickly spread to occupy a large portion of the channel angle of 90° to each other. The depth of the channels is
volume. The vapor was then ejected in the form of more than half the thickness of the substrate, resulting in
bubbles from the surface pores. Recently, Chien and the channels intersecting at discrete locations to provide
Webb [11] conducted high-speed visualization (1000 square pores. In this study, silicon substrates were used
frames/s) of boiling in R-123 from a finned-tube which to fabricate the enhanced structures. The advantage of
was covered with a thin sheet and had pores at regular using silicon is that the enhanced structures can be di-
intervals (Fig. 1). The bubble departure diameter, fre- rectly bonded to the passive side of a heat dissipating
quency of bubble formation and the bubble site density silicon chip, resulting in significantly lower thermal re-
were measured. The bubble growth data showed bubbles sistance and eliminating coefficient of thermal expansion
departing at a faster rate compared to those on a plain (CTE) mismatch.
surface. The study concluded that the bubble formation Two methods of microfabrication, compatible with
phenomenon was thus different than that on a plain chip manufacturing, were used for silicon––wet etching
polished surface. The bubble data were used to develop and wafer dicing. In wet etching, a silicon wafer with
a semi-analytical model for the boiling process. h1 1 0i crystal orientation was patterned using photoli-
The above mentioned studies [7–11] on structured thography and then etched using a 40% KOH solution.
surfaces have been carried out for a limited range of wall In this technique, the channel walls are defined by the
superheat (0–4 °C). With fluorinert liquids (e.g. FC-72), orientation of h1 1 1i crystal planes. These planes are
wall superheats as high as 30 °C can be reached in ap- aligned at an angle of 109.53° and hence the pores are
plications such as electronic cooling, with the wall trapezoidal in shape (Fig. 3a).
temperature below 85 °C. Hence studies for a wider In wafer dicing, a high-speed saw, with diamond
range of wall superheats are required to understand the impregnated nickel blades, was used to mill the channels
boiling mechanism with these liquids. The current study one at a time. The width of the channels was varied
involves visualization of boiling from one such enhanced using blades of different thickness. This resulted in pores
structure in the wall superheat range 4–30 °C. The en- of different sizes. Since this technique is independent of
hanced structure consists of an array of channels and the crystal orientation, the channels were orthogonal
pores that make the structure highly porous. Details of resulting in square pores (Fig. 3b). The structures thus
this structure are presented in the next section. The pa- fabricated were then bonded to a silicon base plate using
rameters measured were bubble diameter as a function thin aluminum foil (2 lm), sandwiched between the two
of time, departure diameter, frequency of bubble for- and annealed at 570 °C. Silicon forms a eutectic with
mation and the bubble site density (nucleation density). aluminum at this temperature, forming a permanent
The ultimate goal was to use the data for formulating a bond. A more detailed discussion on these techniques is
semi-analytical model that predicts the heat transfer presented in Ramaswamy et al. [13] and Ramaswamy
performance of these structures. A framework for de- [14]. Structures with different channel size and pitch were
veloping such a model has been outlined using a flow- used in the current study and the geometrical details are
chart and sample results are included. listed in Table 1.
4764 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771

Fig. 3. Top view of microfabricated enhanced structures (a) wet-etched (b) wafer-diced.

3. Experimental set-up
Table 1
Enhanced structures employed in the visualization A schematic of the test facility is shown in Fig. 4. The
Structure Channel Channel Channel Thickness test set-up consists of evaporator and condenser sections
identification width, pitch, Pt height, of structure, separated by flexible poly-fluoro-alkoxy tubing with an
number Wt (mm) (mm) Ht (mm) Hs (mm) internal diameter of 3.2 mm. This provides a ther-
S-0.20-1.0-W 0.20 1.0 0.26 0.50 mosyphon loop with the condenser placed at a higher
S-0.20-0.7-D 0.20 0.7 0.26 0.51 elevation compared to the evaporator for liquid return
S-0.15-1.4-D 0.15 1.4 0.26 0.51 by gravity. Fig. 4 also shows a detailed sketch of the
S-0.15-0.7-D 0.15 0.7 0.26 0.51 evaporator section. The evaporator consists of a plexi-
S-0.12-0.7-D 0.12 0.7 0.26 0.51 glass enclosure (12.7 mm thick walls) of outside
D: wafer dicing; W: wet chemical etching. dimensions 50 mm
50 mm
40 mm. The heater as-
sembly consists of an Omega cartridge heater embedded

Fig. 4. Schematic of the experimental set-up (a) thermosyphon loop, (b) detailed view of the evaporator section.
 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771 4765

into a square copper rod (1.1 cm2 cross-sectional area). structure. The lens was capable of a variable magnifi-
Omegathermâ ‘‘201’’ high thermal conductivity, high cation of 1–6.5 times the original size of the object. A
temperature paste was applied between the heater and 150 W fiber optic light guide was used for illuminating
the copper rod for good contact. The copper rod was the structure.
insulated with plexiglass and additional styrofoam in-
sulation to minimize heat loss. The enhanced structure
was soldered to one end of this heater assembly. 4. Experimental procedure
Four copper–constantan sheathed thermocouples
(0.08 mm diameter) were embedded at the center of the Every experimental run was preceded by a degassing
copper rod starting from 4 mm below the boiling surface operation. The vent valve was opened and a heat input
at intervals of 4 mm. These thermocouples were used to of 70 W was supplied to the heater (to start vigorous
calculate the heat flux at the base of the enhanced boiling). Boiling was sustained for about 1 h and then
structure (1 cm2 ) by a one-dimensional heat conduction the system was allowed to cool down for 2 h. The vent
analysis. Pool temperature in the evaporator section was valve was kept closed during this period. The system
measured using an embedded copper–constantan ther- reached room temperature by this time in all the runs.
mocouple of 0.08 mm diameter. The thermocouple Before commencement of the test runs, the valve was
measurements were acquired using a National Instru- again opened to ensure atmospheric conditions inside
ments data acquisition system interfaced to a personal the system at all heat input levels. This procedure en-
computer and controlled using LabVIEW software. sured that the dissolved gas concentration of non-con-
A variac (0–140 V) supplied power to the heater. densibles was uniform, if not zero, in all the runs.
Voltage taps were placed on the heater lead wires very The initial heat input was set at 1 W and incremented
near the sheath. This provided precise measurement of in steps of 1 W until boiling was observed. Subsequently,
voltage drop across the heater. A precision resistor the increments were increased to 2.5 W. The image cap-
(1  0:01 X) was connected in series to the power supply ture was carried out along with the temperature mea-
to measure the line current. A Fluke 87 True RMS surements. For each value of power input, the system
multimeter measured the voltages. was allowed to reach steady state (defined by a rise of less
A plate fin condenser was employed with the fin than 0.2 °C in a span of 10 min). The image capture
spacing of 6 mm optimized for operation in natural software was used to start the capture and 500 images
convection. A Robinair pressure gauge ()101 kPa to 0.8 were acquired sequentially at a speed of 1500 frames/s
MPa) was attached to the loop near the exit of the (4:1 aspect ratio). The capture sequence was replayed and
condenser to monitor the loop pressure during operation the number of images to be stored adjusted to capture at
and during leak test of the set-up (in vacuum and under least 10 cycles of bubble growth and departure. The
pressure). An ice cooled spiral tube reflux condenser was images were then transferred to permanent storage on
used to trap any vapor leaving the condenser. It also the computer. The temperatures were then recorded at
provided a vent for the system to maintain it at atmo- two-second intervals for 1 min. The averaged values were
spheric pressure throughout the tests. used to calculate the temperature gradient and the heat
The flow visualization set-up consisted of a high- input. The voltages across the heater and the precision
speed CCD camera (model––Phantom) capable of frame resistor were recorded during the data acquisition and
rates up to 1600 frames/s with a split field. The CCD used to calculate the electrical input to the system.
image size was 512 pixels across by 512 pixels. With an This procedure was repeated at each heat flux until
aspect ratio of 1:1 a maximum of 500 frames/s was the liquid inside the evaporator was agitated so vigor-
captured. Setting the aspect ratio to 2:1 gave two sub- ously (because of bubble generation) that the captured
fields for each normal field, allowing two times the frames did not provide any meaningful pictures. The
number of images (of half the height). This resulted in magnification was kept fixed for each experimental run
1000 images of 512 pixels across by 256 pixels (at the (depending on the channel size). The images were ana-
same 500 frames/s speed). The camera had a personal lyzed using the image analysis software provided by the
computer integrated into it, which was used to control camera vendor (Phantom) and the parameters measured
the capturing and storing of images. The camera vendor were bubble departure diameter, frequency and growth
(Vision Systems, Inc.) supplied the front-end software, period. The nucleation site density was measured by
which was used to control the capture sequence. The direct observation.
input parameters were speed of capture, number of
frames to store after triggering the capture and the as-
pect ratio of the image. An instant replay allowed re- 5. Measurements and uncertainty analysis
viewing the captured images in slow motion.
A zoom lens (Navitar 6000) was used to magnify the The thermocouples and the data acquisition system
area of interest and work at close distances to the boiling were calibrated against a precision mercury thermometer
4766 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771

at ice point to an uncertainty of 0.1 °C. The uncertainty

in voltage measurement was specified by the manufac-
turer as 0.7% of the reading. The precision resistor used
to measure the line current was accurate to 1%. Un-
certainty estimates for the derived quantities were made
using the Kline and McClintok [15] technique. The un-
certainty in the wall superheat values was 0.12 °C. As a
result of the linear curve fit of the temperature data, from
the inline thermocouples, the maximum uncertainty in
the heat flux was 13.6%.
For the flow visualization studies, the maximum
uncertainty in measuring the bubble diameters was
23% at initiation of the bubble growth and 7.8% at
departure. The uncertainty in frequency measurement
was estimated as 3.5% and for the nucleation site
density it was 11.2%.

6. Experimental results

6.1. Boiling regimes

Fig. 5. (a) Isolated bubble regime (laminar regime) for boiling
Zuber [16] classified boiling from plain surfaces into from structure ‘S-0.15-1.4-D’, DTsup ¼ 7:3 °C, (b) mushroom
several regimes. The first of these was termed ‘laminar’. cloud formation at high wall superheat values (25–30 °C) for
In this regime, the bubbles are intermittently formed at structure ‘S-0.2-1.0-W’.
active cavities. Each bubble maintains its identity from
formation to departure. The second regime was termed
as ‘turbulent’ where a few bubbles merge with their The experiments conducted by Chien and Webb [11]
predecessors. Hence there is some amount of vertical were all in the isolated bubble regime (laminar regime).
coalescence. In the third regime almost all bubbles co- For this regime, a switching mechanism was observed
alesce with their predecessors and vapor columns are where some of the active pores suddenly become inactive
formed at the active cavities. This regime eventually ends and some inactive pores become active. This phenome-
up with two or more vertical columns combining to form non was observed in the current study for structures
a larger mushroom cloud above the boiling surface. with larger pores only (Dp ¼ 200 lm). For the other
The current observations show that boiling from pore sizes, the active pores remained active throughout
enhanced structures has some similarity. At low wall the experiments. Chien and Webb [11] also observed
superheat values (4–7 °C), bubbles maintaining distinct that the bubbles departing from the same pore did not
identities were generated from the active pores at a have a constant diameter. A similar observation was
constant rate (Fig. 5a). However, the wake from a de- made in the current study. In addition, the bubbles from
parting bubble did exert a pulling force on its successor. different pores also displayed a size distribution. Hence a
With a rise in the wall superheat values (up to 12 °C), the large sample of bubbles was collected, in this study, to
number of active sites increased and at some sites ver- arrive at a mean departure diameter.
tical coalescence was observed. The formation of vapor
columns at the active sites, beyond this point, depended 6.2. Bubble departure diameter
on the pore pitch. For structures with a large pore pitch
(e.g. ‘S-0.15-1.4-D’) vapor stems were formed similar to A still picture of the structure was taken at the be-
the observation of Zuber [16] before leading to a ginning of each experiment. Using this as the reference
mushroom cloud. However, as the pitch was reduced image, the conversion factor from pixels to mm was first
(e.g. ‘S-0.15-0.5-W’ and ‘S-0.2-0.5-W’), the bubbles co- evaluated. The video sequence was played frame by
alesced laterally before vertical coalescence occurred. frame and the diameter measured for the frame where
Hence the spacing of the pores defines the coalescence the bubble departed from the pore. As mentioned in the
mechanism to some extent, unlike in plain surfaces previous section, the bubble diameter varied from the
where it is a random event. For all the structures, at high same pore and between different pores. Hence a large
wall superheats (20–30 °C), large mushroom clouds were sample size (50 bubbles) was taken to arrive at a mean
formed (Fig. 5b). Hence in this range the enhanced departure diameter. The bubble departure diameters
structures seem to behave like a plain surface. were found to fit a normal distribution. A few examples
 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771 4767

isolated reentrant cavities etched on a silicon surface.

Their study also shows that in the range 5–30 °C, the
departure diameter increases with an increase in the wall
superheat. Zuber [16] suggested that for very low wall
superheats the bubble departure is a function of the
buoyancy and surface tension forces only. With an in-
crease in the wall temperature, the surface tension re-
duces (for most fluids). This would result in a decrease in
the departure diameter. However, as the wall superheat
is increased further, dynamic forces (e.g. inertia forces)
start dominating the bubble growth and hence the
bubble diameter shows an increasing trend with increase
in wall superheat.
In addition to the above observation, the current
study shows that for a fixed pore size the pore pitch very
weakly influences the bubble diameter. For a fixed pore
pitch, a larger pore resulted in a larger departure di-
ameter.

6.3. Frequency measurement

To measure frequency, a sample size of at least 10

Fig. 6. Distribution of bubble departure diameter (a) S-0.2-0.7-
D, DTsup ¼ 4:8 °C, (b) S-0.15-1.4-D, DTsup ¼ 12:3 °C.
bubbles was used. The time elapsed from the beginning
of the first bubble growing to departure of the 10th
bubble was measured and the average time for one
are shown in Fig. 6. The average values were then bubble cycle evaluated. This was done for at least 10
plotted as a function of the wall superheat (Fig. 7). sequences and once again an average value of frequency
The results show that the bubble diameters increase calculated.
with an increase in the wall superheat values. The studies The results for average frequency as a function of the
by Nakayama et al. [9] and Chien and Webb [11] show wall superheat are shown in Fig. 8. The frequency of
an opposite trend where the bubble departure diameter bubble generation increased initially with an increase in
reduced with increase in the wall superheat. Kolev [17] the wall superheat. However, at intermediate wall su-
developed a bubble departure model for boiling on plain perheats (8–12 °C), the frequency actually decreased. At
surfaces. The model shows that the bubble diameters low wall superheat values (4–6 °C), the bulk liquid in the
reduce with increasing wall superheat in the range of 0–3 evaporator was nearly stationary similar to pool boiling
°C. Beyond 3 °C, the values start increasing and this and the departing bubbles did not get affected by the
trend continues up to 20 °C. After this the bubble liquid flowing in from the return line. However, with an
diameter again starts decreasing very sharply. Recently, increase in the wall superheat, the liquid returning into
Fournelle et al. [18] conducted flow visualization on the evaporator induced a sweeping motion on the

Fig. 7. Average bubble departure diameter vs. wall superheat. Fig. 8. Average frequency vs. wall superheat.
4768 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771

enhanced structure. This process was observed to be

random and inhibited bubble generation for brief peri-
ods of time. It is hypothesized that for boiling in large
volumes of liquid, this decrease in frequency might not
occur.
The results also show that the frequency of bubble
generation reduces with an increase in the pore diameter
or the pore pitch. It is attempted to interpret these
findings on the basis of dynamic modeling of the events.
More details are presented in an accompanying paper
that describes the semi-analytical model [19].

6.4. Growth rate

The process of bubble generation in the isolated

bubble regime is intermittent. In this regime, a bubble
cycle typically consisted of three phases––waiting phase,
growth phase and an intake phase. The growth period
was measured by replaying the cine sequences frame by
frame and monitoring approximately 10 bubbles from
the point where they protrude from the pores until they
depart. An average growth period was then evaluated.
The liquid intake period was less than the resolution of
the camera and is assumed to be much smaller than the
waiting and growth periods. The waiting time was then
evaluated using the following equation.
Dtw ¼ f 1  Dtg ð1Þ

Sample pictures of a bubble cycle (from beginning to Fig. 10. Instantaneous bubble diameters (a) ‘S-0.15-1.4-D’,
end) are shown in Fig. 9. The instantaneous bubble di- DTsup ¼ 7.3 °C, (b) ‘S-0.2-1.0-W’, DTsup ¼ 5.3 °C.
ameters were plotted as a function of time. Two cases
are presented in Fig. 10. The results show a nearly linear
growth rate. The growth rate from plain surfaces is transfer controlled phase where the bubble grows [20].
usually divided into two phases––an inertia driven phase The inertia driven phase is typically significantly smaller
where the bubble initiation takes place, and a heat than the heat transfer controlled phase and most studies

Fig. 9. Bubbles growing at a pore for structure S-0.15-1.4-D, DTsup ¼ 7:3 °C.
 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771 4769

assume that the entire growth period is heat transfer because of an increase in the amount of liquid evapo-
controlled. For the inertia driven phase, the instanta- rated in the tunnels.
neous diameter varies linearly with time. On the other The results also show that as the pore pitch was in-
hand, for the heat transfer controlled regime, the vari- creased, the nucleation density decreases. With an in-
ation is proportional to the square root of time. The crease in the pitch, the number of pores on the surface is
current study shows that the mechanism of bubble reduced and hence the nucleation density is also smaller.
growth is indeed different from that on plain surfaces
and is inertia driven for the entire growth period (linear
with time). This confirms the findings of Chien and 7. Framework for semi-analytical model
Webb [11] who made a similar observation with their
structured surface. A semi-empirical model for prediction of heat
transfer rates from the enhanced structure was devel-
6.5. Nucleation site density oped that used the bubble data recorded above. Fig. 12
shows how the measured data and the other boiling
The number of bubbles generated on the surface was parameters feed into each other to form a semi-analyt-
calculated by direct observation. Since the pores on all ical model. Sub-models for the bubble departure diam-
the structures were on a square grid, the number of eter, frequency and nucleation site density have been
bubbles generated at the outermost rows (or columns) developed in conjunction with a correlation for liquid
shadowed the ones generated at the center. Hence an evaporation inside the channels [19]. The semi-analytical
approximate method was used to calculate the number model predicts the total heat transfer rate at a certain
of bubbles. The number of bubbles on the outermost wall superheat.
rows was first recorded (by viewing the front and back Fig. 13 shows the comparison between measured
of the structure). An average value was thus calculated data and prediction based on this model, for a few
for number of bubbles generated per row. Next, the structures (Table 2). The predicted results for heat flux
number of active rows was recorded by viewing the are within 30% of the experimental ones. Details on
structure from the sides (left and right). The total prediction accuracy of the sub-models are included in
number of bubbles was evaluated by multiplying the two the accompanying paper [19]. Accurate prediction of the
numbers. The nucleation site density was evaluated by evaporation inside the channels seems to be the most
dividing the number of bubbles generated by the exter- dominant factor in the efficacy of the model.
nal surface area of the structure.
The resulting nucleation site density is plotted as a
function of wall superheat in Fig. 11. The results show
that the number of bubbles formed increases with an
increase in the wall superheat. For plain surfaces, this
phenomenon occurs because sites with larger cavity sizes
get activated with an increase in wall superheat. For the
structured surfaces, the cavity size is fixed (by the pore
size) and hence the increase in number of bubbles occurs

Fig. 12. Schematic of the semi-analytical model showing the

Fig. 11. Nucleation site density vs. wall superheat. various components and their interdependency.
4770 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771

pools this reduction will not be observed. The fre-

quency reduced with an increase in the pore pitch
and pore diameter.
(d) The nucleation site density increased with an in-
crease in the wall superheat (for all structures). A
larger pitch resulted in fewer bubbles because of
fewer pores. The pore size had negligible effect ex-
cept for one structure where the number of bubbles
increased. The nucleation site density is a function
of the volume evaporated inside the tunnels and
the average departure diameter of the bubbles. With
a change in the pore size, interplay of these two pa-
rameters leads to reduction/no change/increase in
the nucleation site density.

Fig. 13. Comparison of predicted and experimental heat flux

(structures S-0.09-0.7-D, S-0.12-0.7-D and S-0.32-0.7-D).
Acknowledgements

Table 2 This work was performed while the authors were at

Geometrical details of enhanced structures in Fig. 13 the University of Maryland, College Park. The authors
Structure Channel Channel Channel Thickness would like to acknowledge support of this work through
identification width, pitch, Pt height, of structure, the members of CALCE Electronic Products and
number Wt (mm) (mm) Ht (mm) Hs (mm) Systems Center and the Defense Advanced Research
S-0.09-0.7-D 0.09 0.7 0.26 0.51
Projects Agency contract N00164-99-C-0039 and Semi-
S-0.12-0.7-D 0.12 0.7 0.26 0.51 conductor Research Corporation contract 99-NJ-649.
S-0.32-0.7-D 0.32 0.7 0.26 0.51 We also thank Professor Jim Duncan, at the University
of Maryland, for providing us with the high-speed
D: wafer dicing.
visualization equipment. We thank Professor John
Melngailis and his staff for providing us access to the
8. Conclusions cleanroom and assisting us in fabricating the enhanced
structures.
A flow visualization study was conducted to under-
stand the mechanism of boiling from enhanced struc-
tures in the wall superheat range 4–30 °C. The following References
conclusions were drawn based on the study:
[1] C.Y. Han, P. Griffith, The mechanism of heat transfer in
(a) There are distinct boiling regimes for enhanced nucleate pool boiling, Int. J. Heat Mass Transfer 8 (1965)
structures similar to that for plain surfaces. For 887–914.
[2] D.D. Paul, S.I. Abdel-Khalik, A statistical analysis of
low to intermediate wall superheat values (4–12
saturated nucleate boiling along a heated wire, Int. J. Heat
°C), the boiling was in the isolated bubble regime.
Mass Transfer 26 (1983) 509–519.
With an increase in the wall superheat, coalescence [3] C.N. Ammerman, S.M. You, Identification of pool boiling
occurs, leading eventually to formation of mush- heat transfer mechanisms from a wire immersed in satu-
room clouds. The coalescence phenomenon was con- rated FC-72 using a single-photo/LDA method, J. Heat
trolled to some extent by the pore pitch. Transfer 118 (1996) 117–123.
(b) The average bubble departure diameter increased [4] K.-H. Sun, J.H. Lienhard, The peak boiling heat flux on
with an increase in the pore size (for same wall su- horizontal cylinders, Int. J. Heat Mass Transfer 13 (1970)
perheat). The effect of pore pitch was very small. 1425–1439.
For a certain pore size, the departure diameter in- [5] Y. Katto, S. Yakoya, Behavior of a vapor mass in
saturated nucleate and transition boiling in pool boiling,
creased with an increase in the wall superheat.
Heat Transfer––Japanese Res. 5 (1976) 45–65.
(c) The frequency of bubble generation increased mar-
[6] R.L. Webb, Principles of Enhanced Heat Transfer, John
ginally with an increase in the wall superheat. At in- Wiley & Sons, New York, 1994.
termediate wall superheats (12 °C), the frequency [7] W. Nakayama, T. Daikoku, H. Kuwahara, T. Nakajima,
showed a decreasing trend. This decrease might be Dynamic model of enhanced boiling heat transfer on
a manifestation of the induced flow inside the com- porous surfaces, part I: experimental investigation, J. Heat
pact evaporator. It is hypothesized that for large Transfer 102 (1980) 445–450.
 C. Ramaswamy et al. / International Journal of Heat and Mass Transfer 45 (2002) 4761–4771 4771

[8] W. Nakayama, T. Daikoku, H. Kuwahara, T. Nakajima, [14] C. Ramaswamy, A compact two-phase thermosyphon emp-
Dynamic model of enhanced boiling heat transfer on loying microfabricated boiling enhancement structures, Ph.D.
porous surfaces, part II: analytical modeling, J. Heat Thesis, University of Maryland, College Park, MD, 1999.
Transfer 102 (1980) 451–456. [15] S.J. Kline, F.A. McClintock, Describing uncertainties in
[9] W. Nakayama, T. Daikoku, T. Nakajima, Effects of pore single sample experiments, Mech. Eng. 75 (1953) 3–8.
diameters and system pressure on saturated pool boiling [16] N. Zuber, Nucleate boiling––the region of isolated bub-
heat transfer from porous surfaces, J. Heat Transfer 104 bles––similarity with natural convection, Int. J. Heat Mass
(1982) 286–291. Transfer 6 (1963) 53–56.
[10] J. Arshad, J.R. Thome, Enhanced boiling surfaces: heat [17] N.I. Kolev, The influence of mutual bubble interaction on
transfer mechanism and mixture boiling, in: Proceedings of the bubble departure diameter, Exp. Thermal Fluid Sci. 8
ASME-JSME Thermal Engineering Joint Conference, vol. (1994) 167–174.
1, 1983, pp. 191–197. [18] G. Fournelle, S.H. Bhavnani, R.C. Jaegar, Optical Study
[11] L.-H. Chien, R.L. Webb, Measurement of bubble dynam- of Enhanced Heat Transfer from a Heat Sink for Micro-
ics on an enhanced boiling surface, Exp. Thermal Fluid Sci. electronic Applications, in: D. Agonafer, M. Saka, Y.-C.
16 (3) (1998) 177–186. Lee (Eds.), Advances in Electronic Packaging,, EEP-Vol.
[12] W. Nakayama, T. Nakajima, S. Hirasawa, Heat sink studs 26-2, 1999.
having enhanced boiling surfaces for cooling microelec- [19] C. Ramaswamy, Y. Joshi, W. Nakayama, W.B. Johnson,
tronic components, ASME Paper No.84-WA/HT-89, Semi-analytical model for boiling from enhanced struc-
1984. tures, Int. J. Heat and Mass Transfer, submitted for
[13] C. Ramaswamy, Y. Joshi, W. Nakayama, W.B. Johnson, publication.
Compact thermosyphons employing microfabricated com- [20] B.B. Mikic, W.M. Rohsenow, A new correlation of pool-
ponents, Microscale Thermophys. Eng. 3 (4) (1999) 273– boiling data including the effect of heating surface charac-
282. teristics, J. Heat Transfer 91 (1969) 245–250.