Supplementary Material
Suppressing high-frequency temperature oscillations in microchannels with surface
structures
Yangying Zhu, Dion S. Antao, David W. Bian, Sameer R. Rao, Jay Sircar, Tiejun Zhang, and
Evelyn N. Wang
Contents:
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
Microchannel Design and Fabrication
RTD Calibration
Experimental Setup
Heat Loss Measurement and Heat Flux Calculation
Error Estimation
Experiment Repeatability, Boiling Curves and Pressure Drop
Temperature Oscillations at Low Heat Fluxes
Flow Rate Oscillation due to the Peristaltic Pump
T Oscillation Range Calculation
FFT calculation
Page 1
Page 3
Page 4
Page 5
Page 6
Page 7
Page 8
Page 9
Page 10
Page 11
1. Microchannel Design and Fabrication
We designed single microchannels with characteristic dimensions of 10 mm in length, 500 μm in width,
and 500 μm in height. Micropillar arrays (diameter d=10 μm, pitch l=30 μm and height h=25 μm) were
designed at the bottom surface of the microchannel. This d/h and l/d ratio was chosen to optimize
capillary wicking according to a numerical model.1 The microchannel top cover was transparent to allow
flow visualization. To emulate the heat flux from a high performance electronic device, we integrated a
thin-film metal heater (8.6 mm long × 380 μm wide) directly underneath the microchannel. In addition to
the heating element, four thin-film resistance temperature detectors (RTDs) were incorporated along the
length of the heater to measure the microchannel backside surface temperature at different locations.
Figure S1a shows the schematic (to scale) of the heater and RTDs on the backside of the microchannel
device. Specifically, the distances x of RTD1 through RTD4 from the inlet of the microchannel were
0 mm, 1.4 mm, 5.7 mm and 10 mm respectively. The entire device including the inlet and outlet region is
1 cm wide and 3 cm long.
Standard silicon MEMS fabrication processes were used to create the microchannel test devices and are
summarized in Figure S1. The micropillars were etched in a 500 μm thick Si wafer using deep reactive
ion etching (DRIE) of the microchannel bottom surface (Figure S1b). A second 500 μm thick Si wafer
was etched through using DRIE to define the microchannel sidewalls (Figure S1c). Inlet and outlet ports
were created on a 500 μm thick Pyrex wafer by laser drilling (Figure S1d). The two Si wafers were first
bonded using direct Si-Si fusion bonding at room temperature (two minutes), followed by annealing at
1
1000 °C for one hour. Subsequently, a 1 μm silicon dioxide (SiO2) layer was thermally grown on the Si
surface as a hydrophilic coating on the channel walls and as an electrical insulation layer on the backside.
The Pyrex wafer was bonded onto the Si wafers using anodic bonding (Figure S1e). Finally, a layer of
~170 nm thick platinum (Pt) was deposited on the backside of the microchannel with electron-beam
evaporation and patterned by lift-off technique to serve as the heater and RTDs (Figure S1f). In addition
to the microstructured microchannels, we also fabricated microchannels with smooth surfaces following a
similar procedure, however a polished Si wafer (roughness < 50 nm) was used instead of the micropillar
wafer in the initial step.
Figure S1. Design and fabrication process of the microchannel device. (a) Schematic (to scale) of the
heater and RTDs on the backside of the microchannel device. The dotted sections are the electrical
connection lines to the contact pads. (b) Micropillars of 25 μm height were etched in Si using deep
reactive ion etching (DRIE). (c) A Si wafer was etched through using DRIE to define the channel. (d)
Inlet and outlet ports were laser-drilled on a Pyrex glass wafer. (e) The Si layers were bonded using direct
Si-Si bonding. A silicon dioxide (SiO2) layer was thermally grown on the Si surface. The Pyrex layer was
bonded to the top Si layer using anodic bonding. (f) A platinum (Pt) layer was deposited on the backside
of the microchannel using electron-beam evaporation and patterned to form the heater and RTDs.
Figure S2a and S2b show the front and backside of a fabricated microchannel device. The two air
chambers next to the microchannel were incorporated to minimize heat loss via conduction and to better
2
isolate the effect of flow boiling in the microchannel. Figure S2c shows a magnified view of the Pt heater
and an RTD on the backside of the microchannel.
Figure S2. Images of a fabricated microchannel. Optical images of the (a) front and (b) backside of a
device. (c) Optical microscope image of the heater and RTD4 on the backside of the microchannel.
2. RTD Calibration
Before the experiment, the Pt heater and RTDs were annealed at 400 °C for 1 hour to avoid resistance
drift. Higher annealing temperature may be more effective in avoiding resistance drift, but was not tested
in our experiment. All the RTDs were calibrated in an oven (1330GM, VWR). The resistance of the
RTDs was measured at least two hours after each set temperature, and the oven temperature was
measured using a standard Ultra Precise RTD Sensor (P-L-A-1/4-6-1/4-T-6, Omega). The calibration
results are shown in Figure S3. The wiring (FTAKC06010, Kurt Lesker) used for connecting the RTDs
to the measurement system is less than 0.7 m long and has a resistance of 0.045 ohms (resistivity of
64 ohms/km), which is negligible compared to the resistance of the RTDs. A linear correlation between
the resistance and temperature was observed. The symbols are the measured data and the lines are linear
fits. The errors between the linear fits and the measurements are less than 0.1 °C.
Figure S3. Calibration results (temperature vs. resistance) of the RTDs for (a) the smooth surface and (b)
the structured surface microchannels. The symbols are the measured data and the lines are linear fits.
3
3. Experimental Setup
We developed a closed loop test rig to characterize the microchannel test devices during flow boiling
(Figure S4). The loop consists of a liquid reservoir, a pump, pre-heaters to minimize subcooling, a test
fixture to interface with the test device, and various sensors. Degassed and high purity water (WX0004-1,
OmniSolv) was used as the working fluid. Throughout the experiment, water in the liquid reservoir was
heated and degassed to saturated conditions under atmospheric pressure (Twater,res = 100 °C for Pwater,res =
1 atm). The fluid was pumped through the loop using a peristaltic pump (7528-30, Cole Parmer
MasterFlex L/S) to avoid contamination of the working fluid. We used the smallest pump tubing available
(tube ID = 0.8 mm, L/S 13, Masterflex) for the pump used in the study to minimize flow oscillation
amplitudes. In order to measure the flow rate with a liquid flow meter (L-50CCM-D, Alicat Scientific for
G=300 kg/m2s and L-5CCM-D, Alicat Scientific for G=100 and 200 kg/m2s) which has a maximum
working temperature of 60 °C, the degassed liquid was cooled via heat exchange with the ambient to
below 60 °C as it passed through the metal tubing (the line in Figure S4 between the liquid reservoir and
the flow meter). Since there is heat loss to the ambient as the 100 °C water from the reservoir flows
though the metal tubing, the pump and the flow meter, we added pre-heaters (FGR-030, OMEGALUX) to
reheat the liquid. The preheaters maintained a minimum liquid subcooling (10 °C) while also avoiding
boiling at the entrance of the test fixture. Thermocouples (K type, Omega) and pressure transducers
(PX319-030A5V, Omegadyne, Inc.) were used to monitor the loop conditions. The microchannel test
devices were placed in an Ultem test fixture that interfaces with the loop. The test fixture was placed in an
inverted microscope (TE2000-U, Nikon) and the flow was captured using a high speed camera (Phantom
v7.1, Vision Research) at 2000 frames/s. Thermal insulation (cotton) was wrapped around the tubing
connecting the preheater and the text fixture, as well as the test fixture and the liquid reservoir. The
purpose of the thermal insulation is to reduce heat loss to the environment of the liquid entering the
microchannel, and to reduce condensation after the two-phase flow exits the microchannel, respectively.
Figure S4. Schematic of the experimental setup. T, P and M are temperature, pressure and flow rate
sensors, respectively.
4
4. Heat flux Calculation and Heat Loss Calibration
4.1 Heat flux calculation
The heat flux was obtained from the input electrical power (DC voltage and current), the surface area of
the microchannel bottom wall and accounting for the loss to the environment as,
q" 
0.95UI  Ploss
Amc
(S1)
where U is the input voltage, I is the input current, Ploss is the calibrated loss to the environment (Figure
S5a), Amc is the microchannel bottom wall surface area (500 μm × 10 mm). The heat generated in the
electrical connection lines to the contact pads (enclosed in the dotted line in Figure S1a) was 5% of the
total heat from the power supply. The factor of 0.95 (5% loss) in equation (S1) was obtained based on the
geometry of this resistance relative to the total resistance using,
Rc
Rtotal
Lc
Wc

 0.05
Lc Lheater

Wc Wheater
(S2)
where Rc and Rtotal are the resistance of the connection lines and the total heater resistance, respectively, Lc,
Lheater, Wc and Wheater are the length and width of the connection lines and the heater respectively.
4.2 Heat loss calibration
During the flow boiling experiment, most of the heat generated by the heater would be dissipated by the
two-phase flow because of the low thermal resistance of the microchannel compared with the lowthermal-conductivity Ultem fixture (i.e., q” >> q”loss). The temperature (at the heater surface) dependent
heat loss to the environment Ploss (in equation (S1)) was obtained from experiments where the test device
in the fixture was heated with the flow loop evacuated (i.e., no working fluid). Under this condition all the
heat generated by the heater was dissipated into the environment and was considered a close
approximation of the actual heat loss during flow boiling experiment.
Figure S5b shows the calibrated heat loss Ploss (W) as a function of the average microchannel backside
surface temperature Tave (°C). Since T2, T3 and T4 are the backside surface temperatures near the inlet,
mid-point and outlet of the microchannel, respectively, if we approximate the first half of the
microchannel backside surface temperature as 0.5(T2+T3), and the second half as 0.5(T3+T4), the average
microchannel backside surface temperature can be approximated as,
Tave  0.25T2  0.5T3  0.25T4
(S3)
A 2nd order polynomial (close to linear) fit (R2 = 1) between Ploss and Tave was determined from the
experimental data as,
Ploss  4.5 105 Tave  0.0163Tave  0.299
2
5
(S4)
The actual heat loss during flow boiling is difficult to obtain and this calibrated heat loss is a conservative
approximation of the real heat loss.
Figure S5 (a) Schematic of heat flux dissipated by the microchannel during flow boiling, and heat loss to
the environment. (b) The calibrated heat loss as a function of average microchannel backside surface
temperature Tave (°C).
5. Error Estimation
The error bars in the experiments were estimated based on the uncertainty of the measurement from the
instrument error (errorinstrument) and the standard deviation (STD) of multiple data points for the timeaveraged data, described by equation (S5). For transient measurement, only instrument error was
considered.
error  errorinstrument  STD 2
2
(S5)
The instrument error included the resolution of the pressure transducers (±300 Pa), the uncertainty of the
temperature measured by the RTDs (±1 °C, predominantly from data acquisition via the DAQ card, since
calibration error is less than 0.1 °C), the power supply (0.06 V and 0.4 mA), and the flow meter
(±1 ml/min).
Specifically, the heat flux uncertainty mainly originated from the uncertainties of the measured heater
voltage U and current I (0.06 V and 0.4 mA, equation (S1)). An input power of 40 W (microchannel
heater area = 500 μm × 10 mm) corresponds to an uncertainty of 0.05 W (0.13%) from the current and
voltage measurement.
A calibrated heat loss Ploss was also considered in the calculation of the heat flux. The measured Ploss as a
function of temperature is shown in Figure S5b. Using the uncertainty of 0.06 V and 0.4 mA for U and I,
the uncertainty for the measured Ploss is 0.011 W for Ploss=1.35 W and 0.018 W for Ploss =3.36 W, which
were both less than 1%. Since Ploss is a function of T, an uncertainty of 1°C for T results in an uncertainty
of 0.017 W in the Ploss estimate, based on the slope of the Ploss-T curve. Therefore, the total uncertainty of
Ploss was maximally 0.035 W (as T approaches 160 °C), which is negligible compared to the input powers
(30 W – 40 W) at high temperature.
6
However, the actual heat loss during flow boiling was difficult to obtain and this calibrated heat loss is
only an approximation of the real heat loss. The actual error may be larger due to this approximation. An
upper bound of the heat loss can be obtained by using the highest measured temperature (T3) to represent
the average temperature of the heater. This resulted in an upper bound of the error to be ~5%.
6. Experiment Repeatability, Boiling Curves and Pressure Drops
To verify that the boiling curve measurement is repeatable, we performed multiple experiment for the
smooth surface microchannels at q”=100 W/cm2 and the results are shown in Figure S6. The measured
data are within the error bars. Figure S7 and Figure S8 shows the boiling curves and pressure drop
curves respectively for the smooth surface and the structured surface microchannels at all of the measured
mass flux G=100, 200 and 300 kg/m2s. Boiling curves measurements were stopped when the temperature
rise approached 60 °C to avoid device failure due to temperature gradient and high temperature.
Specifically, in some experiments we have observed mechanical cracks near the inlet of the microchannel
as the temperature rise exceeded 70 °C. Factors that may have caused the crack include a large
temperature variation between RTD1 (cooled by the inlet liquid) and RTD2 (heated by the heater), and
thermal expansion of the test fixture which could induce uneven forces to the microchannel device.
Figure S6 Repeated boiling curve measurements for the smooth surface microchannel at G=100 kg/m2s.
7
Figure S7 The boiling curves at mass fluxes of (a) 100 kg/m2s, (b) 200 kg/m2s and (c) 300 kg/m2s. The
shaded areas indicate the temperature oscillation range.
Figure S8 Pressure drop as a function of heat flux at mass fluxes of (a) 100 kg/m2s, (b) 200 kg/m2s and (c)
300 kg/m2s.
7. Temperature Oscillations at Low Heat Fluxes
Figure S9 shows the temperature oscillation of the microchannels at low heat fluxes (G=300 kg/m2s,
q”=110 W/cm2) for a smooth surface (Figure S9a) and a structured surface (Figure S9b) microchannel.
The temperature increased during single phase flow and the corresponding pressure drops were small.
The temperature decreased during bubble growth and departure which caused large pressure fluctuations.
Visualization of the flow at a low heat flux (q”=80 W/cm2) can be found in the Supplementary Video.
8
Figure S9 Temperature and pressure drop oscillations of the (a) smooth and (b) structured surface
microchannels at low heat fluxes (G=300 kg/m2s, q”=110 W/cm2).
8. Flow Rate Oscillation due to the Peristaltic Pump
Figure S10 shows the temperature T1-T4 measured by RTD1-RTD4, respectively, the pressure drop
measured by the pressure transducers, and the mass flux after the pump for the smooth surface
microchannel at (a) G=100 kg/m2s, q”≈376 W/cm2, and (b) G=200 kg/m2s, q”≈424 W/cm2. Apart from
the RTDs which were fabricated on the microchannel, the pressure transducers and the flow meters were
at a large distance to the microchannel due to their physical sizes. The pressure transducers were located
approximately 10 cm to the inlet and outlet of the microchannels, and the flow meter was located
approximately 20 cm after the pump. The mass flux in Figure S10 reflects the mass flux near the outlet of
the pump, rather than the mass flux in the microchannel. The frequency of the mass flux oscillation
agreed with the rotational speed of the pump head. Although the mass flux oscillations were large, its
frequency was much lower than that of T and P oscillations. Therefore, the T and P measurements were
not affected by the peristaltic pump induced mass flux oscillations.
9
Figure S10 The temperatures T1-T4 measured by RTD1-RTD4, respectively, the pressure drop, and the
mass flux measured after the pump for the smooth surface microchannel at (a) G=100 kg/m2s,
q”=396 W/cm2, and (b) G=200 kg/m2s, q”=420 W/cm2.
9. T Oscillation Range Calculation
T oscillation range was taken as the difference between the 95th percentile and the 5th percentile of the
measured temperature (1000 data points per second for at least 2 minutes at the steady state). The 95th
percentile (the upper bound) is the value below which 95% of all the measured temperature falls.
Similarly, the 5th percentile (the lower bound) is the value below which 5% of all the measured
temperature falls. T oscillation range thus implies 90% of all the measured temperatures were between
this range. We did not use the difference between the maximum and minimum values (i.e., the 0th and
100th percentile) to allow for some occasional outliers due to noise. An example of the T oscillation
Range is shown in Figure S11 for smooth surface microchannel at G=100 kg/m2s and q”=397 W/cm2.
Figure S11 T3 oscillation of the smooth surface microchannel at G=100 kg/m2s and q”=397 W/cm2. The
blue lines are the value of the 95 percentile and 5 percentile respectively, to represent the oscillation
bounds for the majority of the measured data.
10. FFT calculation
Fast Fourier transform (FFT) method was used to determine the frequency of the T oscillation. A
Gaussian fit was used to determine the peak frequency f and the standard deviation σ, which was used as
the error bars for f.
10
Figure S12 (a) The FFT of the T oscillation for smooth surface at G = 100 kg/m2s, q” ≈ 400 W/cm2. The
peak is labeled by the red arrow. (b) A Gaussian fit was used to determine the peak frequency (the center
frequency of the peak) and the standard deviation.
Reference:
1
Y. Zhu, D.S. Antao, Z. Lu, S. Somasundaram, T. Zhang, and E.N. Wang, Langmuir 32, 1920 (2016).
11