Proceedings of the 9th IEEE International

Conference on Nano/Micro Engineered and Molecular Systems

April 13-16, 2014, Hawaii, USA

Design and Fabrication of Micro Hot-wire Flow

Sensor Using 0.35µm CMOS MEMS Technology
Zhuonan Miao1, Christopher Y. H. Chao1, Yi Chiu2, Chia-Wei Lin2, Yi-Kuen Lee1*
1
Dept. of Mechanical & Aerospace Engineering, HKUST, Clear Water Bay, Kowloon, Hong Kong
2
Dept of Electrical and Control Engineering, National Chiao Tung University, Hsinchu, Taiwan
*Tel: +852 2358-8663, E-mail: [email protected]

Abstract—MEMS sensors are promising for Energy Efficient 2 ⁄ ) and Prandtl number (Pr). Among these
Building (EeB) because of the potential low cost and low power empirical heat transfer laws, King’s law [9] and the following
consumption. Various flow sensors based on MEMS technology revised one by Kramers [10] are the most commonly used for
have been fabricated. In this work, we designed and fabricated a
polysilicon micro hot-wire flow sensor using a commercial the hot wire anemometry [11].
0.35μm 2P4M CMOS technology followed by post-CMOS
processing. A post-CMOS MEMS process for a 1.5mm×1.5mm
sensor chip using Deep Reactive Ion Etch (DRIE) and spray
coating was utilized to finish the fabrication. The fabricated flow
sensor was characterized at different flow rates. The fabricated
sensor with a dimension of 300μm×2μm×3.76μm demonstrated a
sensitivity of 23.87 mV/(m/s) and power consumption of 0.79 mW
at Uin=5m/s. The experiment results were consistent with the
theoretical prediction and the best results showed an average
error of only 5%.

I. INTRODUCTION
MEMS flow sensors are one of the most important sensors
for various applications, such as energy efficient building
(EeB), microfluidics, medical instrumentation, fuel cells and
so on [1-4]. In particular, a series of micro hot-wire sensors
have been reported in the literature using MEMS technology
[5-7]. However, to date, very few micro hot-wire sensors
employed CMOS MEMS technology [8]. In this work, we
designed and fabricated a micro hot-wire sensor working in
the Constant Temperature Anemometer (CTA) mode using a
commercial 0.35μm 2P4M CMOS technology followed by
post-CMOS processing.
Fig. 1. (a) Schematics of a micro flow sensor, (b) layout of a micro flow
sensor chip, (c) the on-chip circuit diagram for Constant Temperature
The schematic drawing of a hot-wire sensor and the layout Anemometer (CTA) mode.
design are shown in Fig. 1. The rectangular sensor probe
consists of poly-silicon for sensing and silicon dioxide for Based on both experimental and theoretical work, King
electric insulation and protection with length L, width W, proposed the convective heat transfer of a wire with a uniform
thickness T, electric resistance Rw and equivalent thermal free-stream air velocity of Uin as
conductivity kw. The Constant Temperature Anemometer
(CTA) mode was designed to be achieved by the on-chip / (1)
circuit as shown in Fig. 1(c). The operating temperature of the where A and B are the constants of the fluid related to heat
sensor Tw is set at 323K. conduction and convection, respectively [9].

II. THEORY AND MODELLING Kramers [10] further extended King’s law by considering the
experiments for wires placed in air, water and oil and the
The fundamental principle of hot-wire sensors employ equation is expressed as
convective heat transfer laws which relate the heat convection
coefficient (h), or normalized h, i.e., Nusselt number . . .
0.42 0.57 (2)
(Nu=hDh/ka), of the micro flow sensor to the input flow
velocity in terms of Reynolds number (Re = UinDh/ ν where

978-1-4799-4727-0/14/$31.00 © 2014 IEEE 289

where the film temperature [12] Tf=(Tw+Ta)/2 is selected as As the CMOS process was finished, the post-CMOS
the reference temperature for the fluid properties including MEMS process was carried out to define the final sensor
dynamic viscosity μ, density ρ and thermal conductivity ka. structure as demonstrated in Fig. 2. Firstly the polishing was
Kramer’s law is applicable in the ranges 0.01<Re<10000 and performed to reduce the wafer thickness from 650μm to
0.71<Pr<1000. 300μm as shown in Fig. 2(b). This was carried out by using
the manual polisher at HKUST’s MCPF. Secondly, the
Based on Eq. (2), a 1D analytical model to predict sensor backside silicon was etched by Deep Reactive Ion Etch (DRIE)
output was constructed: the output voltage Vout can be related with spray-coated photoresist as the etch mask as shown in Fig.
to the air velocity Uin based on the conservation of energy: 2(c). Thirdly, through front-side DRIE, the sensor structure
was defined as shown in Fig. 2(d). The difficulties in this
process were to perform photolithography on small area and
2 (3)
2 / then etch through the chip. The spray coating technique was
This 1D model can be used to perform the scaling analysis employed to accomplish the photolithography process. The
and predict the sensor performance as presented in our non-uniform etch rate of front-side silicon dioxide affected the
previous work [13]. fabrication yield. The fabrication of Sensor 1 was successful
while for Sensors 2 and 3, some silicon dioxide could not be
III. DESIGN AND FABRICATION removed which degraded the thermal isolation between the
An optimized geometrical design should achieve both high sensor probe and the substrate.
sensitivity and low power consumption. To optimize our own
sensor design, the target power consumption and sensitivity
were selected based on the performance of commercial flow
sensors. Currently, the power consumption of a typical flow
sensor for energy efficient building system can be as low as
23mW (e.g., Zephyr HAF series, Honeywell, USA). For
typical HVAC system in the building, the air velocity usually
ranges from 0 to 20 m/s. Uin=5m/s was selected as the
operation condition and at this condition, the sensitivity from a
commercial hot-wire anemometer sensor was 20mV/(m/s).
The performance of the fabricated hot-wire sensors and a
commercially available flow sensor for EeB (Zephyr HAF
series) is summarized in Table 1. Based on the EeB
application and our previous scaling analysis [11], we
proposed three sensor designs: (L×W×T) Sensor 1:
300μm×2μm×3.76μm, Sensor 2: 100μm×2μm×3.76μm and
Sensor 3: 100μm×2μm×0.76μm.

TABLE 1. SUMMARY OF TARGET SENSOR PERFORMANCE

@Uin=5m/s Target (Zephyr HAF Series )

Velocity Range, m/s 0~50

Sensitivity, mV/(m/s) * 20

Power (mW) * 1 Fig. 2. Post-CMOS fabrication process: (a) the cross section of the chip after
CMOS process, (b) polishing to reduce the chip thickness from 650μm to
300μm, (c) backside silicon etch by Deep Reactive Ion Etch (DRIE) with
spray coated photoresist as the etch mask.
The layout design of micro flow sensor is in Fig. 1 (b). The
fabrication process of micro flow sensor can be divided into IV. EXPERIMENTAL SETUP
two parts: the CMOS process and the post-CMOS MEMS
As the fabrication process was finished, the fabricated sensor
process. A commercial 2P4M 0.35μm CMOS MEMS process
was packaged by wire bonding was shown in Fig. 3 and then
was employed to fabricate the sensor chip. The advantage of
tested. We first measure the temperature coefficient of
the CMOS MEMS process is manufacturing sensor and circuit
at the same time. Since higher resistivity is desirable, the poly- resistance (TCR) of the fabricated flow sensors in an oven at
silicon layer was used to construct the sensor body and the different temperatures. The linear and quadratic coefficients of
surrounding silicon oxide was used as protection layer. Metal temperature were: 5.59-4 K−1 (PTC1) and 5.76-7 K−2 (PTC2).
layers were used as the mask to define the sensor geometry
and depending on which the layer of metal was used, the The experimental setup used to obtain the relationship
thickness of sensor probe could be controlled. between the sensor output voltage and air flow speed could be

290
 heating. The imbalance of the Wheatstone bridge would be
compensated by the feedback of an Op Amp LM358.
Following the same rule of on-chip circuit, the resistance
values of R1 and R3 were set to be 1/5 of the hot-wire sensor.

V. RESULTS AND DISCUSSION

With the experimental setup and the off-chip circuit, the
relationship between the sensor voltage output and the air flow
speed can be determined. The testing results of Sensor 1 are
shown in Figs. 4 and 5. It was found that the average error
between the theoretical prediction and actual measurement
was about 5% for output voltage and 12 % for power which
demonstrates that our analytical model is accurate in output
prediction. The measured sensitivity showed a large deviation
from the theoretical prediction which might be the result of
not enough data points. However, at a larger flow rate they
Fig. 3. Experimental setup to test the sensor output at different flow
velocities.
shared the same trend.

Velocity (m/s)
0 20 40 60
5

3
Vout (V)

2
Experiment
1 Prediction
0
0 0.2 0.4 0.6 0.8 1
Flow Rate (SLM)
Fig. 5. Voltage output of Sensor 1 (L=300μm, W=2μm and T=3.7μm;
R=10.525kΩ) at Ts=43˚C. SLM = Standard Liter per Min.

Velocity, m/s
Fig. 4. Off-chip circuit diagram (Roffset>>R3, R4).
0 20 40 60
3 200
shown in Fig. 3. Nitrogen with maximum pressure of 15 bar Power (Experiment)
was used as the gas source and by adjusting the valve we can Power (Prediction)
2.4
Sensitivity, mV/(m/s)
control the flow rate and thus the flow speed. A commercial Sensitivity (Experiment) 150
Sensitivity (Prediction)
flow sensor AWM5104VN (Honeywell, USA) was used to
Power, mW

indicate the flow rate and a pressure sensor was used to 1.8
measure the pressure drop. The gas flow system was 100
connected through 6mm pneumatic tube and the 1.2
corresponding connectors which could withstand the pressure 50
up to 7 bar. 0.6

Due to unknown reasons, the on-chip circuit could not 0 0

function and therefore the off-chip feedback circuit was 0 0.2 0.4 0.6 0.8 1
constructed to achieve the CTA mode. The circuit Flow Rate, SLM
Fig. 6. Sensitivity and power consumption of sensor 1 (L=300μm, W=2μm
configuration used in [14] can be shown in Fig. 4. By and T=2μm; R=10.525kΩ) at Ts=43˚C. SLM = Standard Liter per Min.
adjusting the offset voltage and the value of potentiometer, the
voltage across the hot-wire sensor at the no flow condition
could be adjusted to the desirable value to provide enough

291
 TABLE 2. SENSOR PERFORMANCE SUMMARY
Velocity (m/s) @Uin=5m/s Sensor 1 Sensor 2 Target
0 20 40 60 Velocity Range, m/s 0~60 0~60 0~50
5
Sensitivity, mV/(m/s) * 23.87 47 20

4 Power (mW) * 0.79 2.76 1

3
Vout (V)

Experiment (Sensor2)
Theoretical prediction VI. CONCLUSION
2
In conclusion, by making use of the theoretical model, we
1 designed and fabricated a micro hot-wire sensor with on-chip
feedback circuit using a commercial 0.35μm 2P4M CMOS
0 technology followed by post-CMOS processing. The sensor
0 0.2 0.4 0.6 0.8 1 output was obtained at different flow speeds. It was shown in
Flow Rate (SLM) that for Sensor 1 the average error between theoretical
Fig. 7. Voltage output of Sensor 2 (L=100μm, W=2μm and T=3.76μm;
R=3.450kΩ). SLM = Standard Liter per Min. prediction and the best results was 5% and the output
sensitivity is 23.87 mV/(m/s) and power consumption 0.79
mW which met the requirements. Therefore, the fabricated
Velocity (m/s) micro flow sensors using CMOS MEMS technology are
0 20 40 60 promising for EeB. With the CMOS MEMS technology, in
6 the future, more functions and control units can be
incorporated into the flow sensor to provide faster
5 measurement speed, higher output accuracy and higher energy
4 efficiency. It is hoped that the development of CMOS MEMS
Power (mW)

flow sensor will bring us not only a technical breakthrough but

3 also enormous social and economic benefits.
Experiment (Sensor 2)
2 Theoretical prediction ACKNOWLEDGEMENTS
1
The authors would like to acknowledge HKUST Special
0 Research Fund Initiative (SRFI) project “Energy Efficient and
0 0.2 0.4 0.6 0.8 1 Low Carbon Emission Sustainable Buildings.”
Flow Rate (SLM)
Fig. 8. Power output of Sensor 2 (L=100μm, W=2μm and T=3.76μm;
R=3.450kΩ). SLM = Standard Liter per Min.
REFERENCES
[1] W. J. Fisk and A. T. De Almeida, "Sensor-based demand-controlled
The voltage and power output of Sensor 2 can be found in ventilation: a review," Energy and Buildings, vol. 29, pp. 35-45, 1998.
[2] N. T. Nguyen, "Micromachined flow sensors - a review," Flow
Figs. 7 and 8. It was found that the experimental data also Measurement and Instrumentation, vol. 8, pp. 7-16, Mar 1997.
shared the same trend with the output from analytical model. [3] Y. H. Wang, C. P. Chen, C. M. Chang, C. P. Lin, C. H. Lin, L. M. Fu,
However the data showed that real voltage and power to and C. Y. Lee, "MEMS-based gas flow sensors," Microfluidics and
maintain the sensor at preset temperature were almost 100% Nanofluidics, vol. 6, pp. 333-346, Mar 2009.
larger than the theoretical prediction. The large deviation [4] J. T. Kuo, et al., "Micromachined Thermal Flow Sensors—A Review,"
Micromachines, vol. 3, pp. 550-573, 2012.
could be due to the imperfection in the fabrication process.
[5] Y. C. Tai and R. S. Muller, "Lightly-Doped Polysilicon Bridge as a
The remaining oxide resulted in excessive conductive heat Flow Meter," Sensors and Actuators, vol. 15, pp. 63-75, Sep 1988.
loss from the sensor to the substrate. In order to maintain the [6] M. Mischler, et al., "A micro silicon hot-wire anemometer," in 1995
constant temperature mode, the input power or the supply IEEE Region 10 International Conference on Microelectronics and
voltage has to be increased. This is a reasonable explanation of VLSI (TENCON'95), 1995, pp. 20-23.
the large deviation of the results of Sensor 2. [7] M. M. Sadeghi, R. L. Peterson and K Najafi, "Air flow sensing using
micro-wire-bonded hair-like hot-wire anemometry." J. Micromech.
Microeng. vol. 23, 2013
A summary of the sensor performance can be found in [8] H. Baltes, O. Brand, G. K. Fedder, C. Hierold, J. G. Korvink & O.
Table 2: Sensor 1 demonstrates a higher sensitivity of 23.87 Tabata, (Eds.). CMOS-MEMS: Advanced Micro and Nanosystems.
John Wiley & Sons, 2008
mV/(m/s) compared with 20 mV/(m/s) and a lower power
[9] L. V. King, "On the convection of heat from small cylinders in a stream
consumption 0.79 mW compared with 1 mW which shows of fluid: determination of the convection constants of small platinum
that the fabricated sensor meets the design target. wires, with applications to hot-wire anemometry," Proceedings of the
Royal Society of London. Series A, vol. 90, pp. 563-570, 1914.

292
[10] H. Kramers, "Heat transfer from spheres to flowing media," Physica,
vol. 12, pp. 61-80, 1946.
[11] H. H. Bruun, Hot-Wire Anemometry: Principles and Signal Analysis:
Oxford University Press, 1995.
[12] J. Holman, Heat Transfer, 9 ed.: McGraw-Hill, New York, 2002.
[13] Z. Miao, C. Y. H. Chao, Y. Chiu, and Y.-K. Lee, "Scaling Analysis of
Low-Power Micro FlowSensors for Energy-Efficient Buildings," The
Asia-Pacific Conference on Transducers and Micro/Nano Technologies
(APCOT), Nanjing, China, 2012.
[14] F. Jiang, "Silicon-micromachined flow sensors," PhD dissertation,
California Institute of Technology, USA, 1998.

293