surface micromachined thermal shear stress sensor - Semantic Scholar
SURFACE MICROMACHINED THERMAL SHEAR STRESS SENSOR Chang Liu, Yu-Chong Tai Electrical Engineering 116-81 California Institute of Technology Pasadena, California
.
Jin-Biao Huang, Chih-Ming Ho Mechanical, Aerospace and Nuclear Engineering Department University of California at Los Angeles Los Angeles, California
ABSTRACT
We have designed, fabricated, and tested a new type of thermal shear-stress sensor. The sensor consists of a 2 m wide polysilicon thermistor (with a range of lengths, 80-200 m) on a 1.2 m thick silicon-nitride diaphragm, which is on top of a vacuum cavity 2002002 m3 in size. The vacuum cavity provides a good thermal isolation between the resistive element and the substrate; shear-stress sensitivity of 15 V/kPa at a power consumption of 12 mW (constant current = 2 mA) is obtained. The typical thermal time constant of our sensors is 350 s, which translates into a bandwidth of 500 Hz. INTRODUCTION
In an on-going University Research Initiative (URI) project, we are developing a distributed micro-electromechanical system (MEMS) for surface drag reduction in a turbulent ow eld. In our system, sensors will collect 2-D ow- eld information on its surface. This information will be processed, and command signal is passed to microactuators (Liu et al., 1994a) to control the vortices within the boundary layer. Accordingly, the sensors that provide the ow- eld information must have a ne spatial resolution (100 m ) and a fast frequency response (> 1 kHz). Among various possible
ow-sensing schemes, surface shear-stress sensing is the focus of this work.
.
There are many reported shear-stress sensors made by conventional manufacturing techniques. These sensors use either direct or indirect methods for shear-stress detection (Haritonidis, 1989). In the direct method, a tangential force on a surface oating balance gives the shear stress directly (Dhawan, 1953 and Mabey, 1975). In the indirect methods, shear stress is extracted from other physical measurands that are indirectly related to the shear stress. Such indirect measuring devices include Preston tubes (Preston, 1953), Stanton tubes (Hool, 1956 and East, 1966), and hot-wire/ lm surfacemounted sensors (Rubesin et al., 1975 and Sandborn, 1979). These conventional shear-stress sensors, however, can only be hand-made one at a time that we can not use them in our system where at least thousands of sensors are required. To solve this problem, we have looked for answers from silicon micromachining, a new but promising technology. Silicon micromachining, which has evolved from silicon integrated-circuit (IC) techniques, is a new technology that enables both miniaturization and mass production of electro-mechanical devices suitable for our use. In fact, silicon micromachining has already produced many new devices for uid measurements with much improved performance over their conventional counterparts. For example, micromachined surface oating balances for direct shear-stress measurements were reported (Schmidt et al., 1988 and Shajii et al., 1992). The measured sensitivity was 52 V(ac)/Pa for gas sensing using a di erential capacitor readout and 13.7 V/V-kPa for liquid sensing using
DESIGN AND FABRICATION
Fig. 1 shows the schematic design of our shear-stress sensor. A polysilicon resistor (80{200 m long, 2 m wide, and 0.45 m thick) is located at the center of a cavity diaphragm, typically 200200 m2 in size and about 2 m above the bottom of the cavity. The resistors are uniformly doped to a low sheet-resistance value of 50 =2, with typical resistances between 2{5 k . The micromachining fabrication process is shown in Fig. 2. A 200 nm LPCVD silicon nitride is rst deposited on a 4-inch wafer and patterned to de ne the cavities (as 200200 m2 windows in the silicon-nitride layer). The windows are etched down 600 nm into the silicon substrate by wet silicon etchant, and then planarized by the following thermal-oxide growth (1.3 m) on the exposed silicon. A 400 nm sacri cial phosphosilicate glass (PSG) is then deposited and patterned, followed by a blank deposition of 1 m low-stress silicon nitride as the diaphragm material. Next, etching holes are opened in the silicon-nitride layer to expose the sacri cial PSG. The sacri cial PSG and the underlying thermal oxide, if any, is etched by (49%) hydro u-
I
POLYSILICON RESISTOR
Unfortunately, none of these devices can be directly adopted for our project. In this paper, we present our work on the development of a new, surfacemicromachined thermal shear-stress sensor, with the fundamental design of a hot wire on a free-standing silicon nitride diaphragm. The novel aspect of the sensor is that this diaphragm is on a vacuum cavity which minimizes the heat conduction from the diaphragm to the substrate through the gap. From our work, it is found that this approach does signi cantly improve the thermal isolation; an unprecedented shear-stress sensitivity of about 15 V/kPa under constant current drive (2mA, 12 mW) from our micro-sensors is obtained.
oric acid for about 20 minutes. The wafer is then dried, and a silicon-nitride layer (400 nm) is deposited at 300 mTorr (0.04 Pa) to seal the cavity under vacuum (Liu and Tai, 1994b). The thickness of the cavity diaphragm is approximately 1.2 m .
METAL LINE
a piezoresistive readout. Flow sensors based on heattransfer principles (van Oudheusden, 1992; Petersen, 1985; Tai and Muller, 1988) were also demonstrated using free-standing beams (Mastrangelo, 1987 ; Tobata, 1986; Goldberg H.D. et al., 1994), free-standing diaphragms (van Oudheusden, 1990), and low thermalconductivity layer such as polyimide (Lofdahl et al., 1989 and Stemme, 1986) for better thermal isolation. These features were used to solve the common but major problem that the heat loss to the substrate limits the performance of the conventional thermal shear-stress sensors.
I (a)
POLYSILICON RESISTOR
SILICONNITRIDE DIAPHRAGM
CAVITY
(b)
AIR FLOW
silicon substrate
Fig. 1 Schematic (a) top and (b) cross-sectional view (along line I-I) of a thermal shear-stress sensor. A vacuum cavity is created underneath the diaphragm to minimize heat loss to the substrate. Resistors on silicon substrate without the vacuum cavity underneath have also been made.
To form the thermistors, a 450 nm polysilicon layer is deposited and patterned. The following doping is done by ion implantation with phosphorus using a total dose of 1 1016 =cm2 . The wafer is then annealed at 1000 C for 1 hour to activate the dopant and to reduce the intrinsic stress in the as-deposited polysilicon. A 100 nm layer of LPCVD silicon nitride is deposited for passivating the polysilicon resistors to prevent their resistance drift (Saraswat and Singh, 1982). Finally, aluminum metalization forms the leads. Micrographs of the fabricated devices are shown in Fig. 3; since the cavity is held under vacuum, the diaphragm is bent down by the external atmospheric pressure so that optical interference patterns (Newton rings in Fig. 3a) can be seen under the microscope. It is important that the shear-stress sensors should be at so that they do not cause surface-roughness effects such as signal uctuation (Goldstein, 1983). Surface pro les of our sensors are examined using a surface pro lometer (Tencor Instrument Alpha-Step 200) with a stylus force of 4 mg. Fig. 4 shows that, even de ected
by both the stylus force and the atmospheric pressure, the device has a roughness of only 450 nm. This roughness nicely falls within our design speci cation of a few micrometers. In addition, this surface pro le also con rms that the diaphragm is not in contact with the cavity bottom, which is 2 m below the surface.
etching channel Cavity
Resistor
etch hole
nitride
(1)
substrate
(a)
(5) sealing
oxide
(2)
Metal lead
Al CONTACT
(6) PSG
polysilicon 1234 1234
(3)
RESISTOR
(7) silicon nitride
metal 1234 1234
(4)
(8)
Fig. 2 Fabrication steps. (1) Silicon nitride is deposited and windows are opened; (2) fully-recessed local thermal oxidation of silicon (LOCOS) to planarize the surface; (3) sacri cial phosphosilicate glass (PSG) deposition and patterning; (4) low-stress silicon nitride deposition for diaphragms; (5) etch hole opening and sacri cial layer removal; (6) vacuum cavity sealing; (7) polysilicon deposition and patterning to form hot wires; (8) metalization to form leads.
(b)
Fig. 3 Micrographs of a fabricated shear-stress sensor with a 200200 m2 diaphragm. (a) An optical micrograph of the shear-stress sensor with clear interference patterns(Newton ring). (b) An SEM picture of a polysilicon resistor.
WIND TUNNEL SETUP
Shown in Fig. 5, the wind-tunnel is approximately 5 m long, 60 cm wide and 2.5 cm high. It has a maximum mean-stream velocity of about 25 m/s. The sensor package is about 2 2 cm2 in area and is designed such that the sensor surface is ush-mounted with the wind-tunnel side wall. The sensor chip is located in the mid-span on the top wall of the wind tunnel, in a downstream region where turbulence ow is fully developed. The long axis of the resistor bar is perpendicular to the mean- ow direction.
Polysilicon resistor
Stepheight = 450 nm
Diaphragm size = 200 µm
Fig. 4 Surface roughness of a shear-stress sensor as measured by a surface pro lometer with a stylus force of 4 mg. The diaphragm is de ected by about 450 nm at the center due to the combined stylus force and the atmospheric pressure on the diaphragm.
R=2.5181 (1+ 0.001307 t) 3
ga sf
lo w
2.5cm
shear-stress sensor
60 cm
WIND TUNNEL
Resistance (kOhm)
5m
2.8
2.6
20
40
60
80
100
120
140
160
180
200
Temperature (degree C)
Fig. 5 Schematic diagram of the wind-tunnel and the shear-stress sensor mounting.
Fig. 6 Resistance vs temperature measurement for a resistor on the silicon substrate. The resistor is 100 m long, 2 m wide and 450 nm thick. The temperature coecient of the resistor is 0.13 %/ C .
EXPERIMENTAL RESULTS Electrical and Thermal Characteristics
Temperature coecient of resistance (TCR) for the resistors is measured in still-air using a hot chuck. Fig. 6 shows measurement data of the resistance as a function of the temperature . To obtain TCR, we use the relation that R(T ) = R0 (1+ 4T ) to t the data, where 4T is the temperature change and is the TCR. A resistor has a typical positive TCR value of approximately 0.13 %/ C . The next subject is to determine the power biasing. There is a maximum allowable power consumption associated with each sensor. Over the limit, either the diaphragm will fracture due to thermal heating or the resistor will have electrical breakdown. In our experiments, we normally set the bias for the resistors below 15 mW to ensure long-term reliable operations. The performance of the thermal isolation for our shear-stress sensors is evaluated here. For comparison, we xed the size of the resistors used in our experiments (100 m long, 2 m wide and 0.45 m thick). The IV characteristics of the resistor on a nitride diaphragm over a vacuum cavity, over an air- lled cavity and on the silicon substrate (without a cavity) are given in Fig. 7. It is clear that the resistor on the vacuum cavity has the largest non-linearity because the vacuum cavity offers the best thermal isolation.
Current (mA)
3
2
1
Resistor #1 over Silicon substrate Resistor #2 over a vacuum cavity Resistor #2 over an air-filled cavity 0 0
1
2
3
4
5
Voltage (V)
Fig. 7 I-V characteristics of a resistor (100 m long, 2 m wide and 450 nm high) located over a vacuum cavity, over an air- lled cavity and directly over the substrate.
Fig. 8a then shows the resistance as a function of power using the data from Fig. 7 and the resistor's TCR. The solid lines are theoretical ttings with 4T = P and R(T ) = R0 (1 + P ), where is the thermal resistance and P is the power. As a result, the thermal resistances (over the vacuum cavity, over the air- lled cavity, and over the substrate) are quanti ed from the slopes of the lines in Fig. 8b. It is found that of the resistor over the cavity is 2:4 104 C/W , 1:8 104 C/W for over the air- lled cavity, and 2:7 103 C/W for over the substrate. It is clear that a vacuum cavity improves the thermal isolation about 1.33 times over an air- lled cavity, and about 9 times over just the silicon substrate. However, it should be noted that the thermal resistance ratios depend on sizes of the cavity diaphragms and the resistors.
Normalized Resistance ( R / Ro)
1.5
R over vacuum cavity R over air-filled cavity R on substrate
1.4
09
20
.03
0 e=
p
slo
1.3
e= slop
1.2
1.1
29
228
0.0
5
slope = 0.003
Wind-tunnel tests of the micromachined shearstress sensors are conducted using constant-current (CC) mode with the circuit shown in Fig. 10. Fig. 11 shows the measured steady-state sensor voltage output (time average) with respect to the mean-stream velocity under CC mode within a velocity range between 5 to 25 m/s. As shown, the over-all slope of the response is 1.069 mv= (m=s). driving pulse
1 0
2
4
6
8
10
12
14
16
Power (W)
(a)
voltage response
τ1=8 µs τ2=350 µs 4
0 x1
pe
(C
500 µs/div.
500 ms/div.
.4 =2
slo
200
4
e
slop
10 .8x
/W)
=1
100
03 ( C/W)
slope = 2.7x1 0 0
2
4
6
8
10
12
14
(a)
80
(C
Output Voltage (mV)
Temperature Difference ( C)
)
/W
R over vacuum cavity R over air-filled cavity R on direct substrate
300
16
60
40
20
Power (W)
(b) Fig.8 Thermal measurement data converted from Fig. 7. (a) Resistance vs power, and (b) temperature vs power. Frequency Response
In order to measure the time constant of the resistors, we pass a step current through the resistors and observe the change of the voltage drop across the resistor as a function of time (Fig. 9). For a resistor with a positive TCR, the output voltage increases. Experimentally, the response curves are recorded and then tted with exponential rises using as many time constants as necessary. It is found that there are two measurable time constants: 1 the electrical time constant, and 2 , the thermal time constant. Fig. 9a shows that for a resistor 100 m long, 2 m wide and 0.45 m high, 1 is about 8 s while 2 being 350 s. The two corner frequencies in the frequency response curve, 500 Hz and 20 kHz, correspond well with the thermal and electrical time constants. Wind Tunnel Tests
101
102
103
104
105
106
Frequency (Hz)
(b)
Fig. 9 (a) Voltage output for the time-constant measurements in still air for a typical sensor with the polysilicon resistors being 100 m long, 2 m wide and 0.45 m thick. (b) Frequency response curve of the same sensor driven at 1 mA(ac). Rh
+15 V SQ Wave S 10K
1K
Rp
(50K)
Bal (20K)
1K
A1 +
Vout
Fig. 10 The circuit used for constant-current mode windtunnel measurements. Rh is the shear-stress sensor resistor.
In order to correlate the output voltage with the wall shear stress ( w ), two theoretical methods (Hussain, 1970) are used here. In the rst method, w in a fully developed channel ow is related to the streamwise pressure gradient by dPx dx
= , hw ,
(1)
Fig. 13. As the mean-stream velocity increases, the local shear-stress uctuation also increases.
85
80
85
Vout [mv] = 58.12 + 14.89 [mv/ Pa] * T [Pa]
Output voltage (mv)
Output voltage (mv)
Vout [mv] = 57.39 + 1.069 [mv/ m/s] * V [m/s]
75
70
65
80
75
70
65
5
10
15
20
25
0.2
0.4
Mean Flow Velocity (m/s)
where Px is the local pressure, x is the stream-wise coordinate, and h is the half height of the wind tunnel. In the second method, an empirical relationship between the Reynolds number and the wall shear stress in a fully developed channel ow is obtained, by using ,0:089 ,
(2)
and w
= u 2 ;,
1
1.2
1.4
(3)
where u is the shear velocity, u1 is the mean stream velocity, Re is the Reynolds number calculated by using the central-line mean stream velocity, and ; is the air density. We rst measure the stream-wise pressure gradient and calculate w by using Eq. 1. Afterwards, we use the second method given by Eds. 2 and 3 to calculate the wall shear stress again. Good agreement is found between shear-stress values calculated by using both methods. For example, at free stream velocities of 9.96 m/s, 14 m/s, and 20 m/s, the rst method gives w 's of 0.2388 Pa, 0.5384 Pa and 0.945 Pa, while the second method gives w 's of 0.2813Pa, 0.5232 Pa, and 1.029 Pa. This agreement gives us con dence in directly estimating w from the mean stream velocity. Re-plotting Fig. 11 into Fig. 12 with the sensor-output voltage as a function of w using method 1, we nd a typical shearstress sensitivity of 15 V/kPa under constant-current mode for a typical sensor. Fluctuating sensor signals are also recorded since the sensor is subjected to a fully-developed turbulent ow. The root-mean-square (RMS) value of the sensor output is then plotted against the free-stream velocity in
3
2.5
2
1.5
1
5
10
15
20
25
Mean Stream Velocity (m/s)
Root-mean-square voltage vs. mean ow velocity in a fully turbulent ow eld.
Finally, a conventional hot-wire anemometer is placed approximately 1 mm above the shear-stress sensor and the output signals from both sensors are compared to see if there is any correlations. As shown in Fig. 14, the shear-stress sensor output does not have high frequency contents as compared to the hot-wire anemometer. However, it is very clear that major features in the anemometer response do correspond well with those of the shear-stress sensor. Output voltage (2 mv/div.)
1 = 0:1079Re
0.8
Fig. 12 Output voltage as a function of the wall shear stress converted from Fig. 11 using Eq. 1. RMS Output Voltage (mv)
Fig. 11 Output voltage as a function of the wind-tunnel mean- ow velocity for a type-1 shear-stress sensor biased at 12 mW (constant current 2 mA). The resistor is 100 m long, 2 m wide and 0.45 m thick.
u u
0.6
Wall Shear Stress (Pa)
shear-stress sensor
hot-wire anemometer flow
~1mm hot-wire anemometer
Time
shear-stress sensor chip
Windtunnel Wall
(1 ms/div.)
Fig. 14 Simultaneous voltage output from a shear-stress sensor and a hot-wire anemometer placed 1 mm apart. CONCLUSION
A surface-micromachined shear-stress sensor based on heat transfer mechanisms has been developed. A unique vacuum cavity is fabricated underneath the heating element to reduce the heat loss to the substrate. Wind
tunnel tests show a shear-stress sensitivity of 15 V/kPa (without signal ampli cation) for a typical sensor under constant-current mode operation (I=2 mA and P= 12 mW). ACKNOWLEDGEMENTS
This project is supported by AFOSR (under the University Research Initiative program) and ARPA. We thank Mr. Fu-Kang Jiang and Mr. Trevor Roper at the Caltech micromachining lab for help with processing and testing. We also thank Dr. Steve Tung and John Mai at UCLA for wind-tunnel tests. REFERENCES
Dhawan, S., 1953, \Direct Measurements of Skin Friction," NACA TN 2567. East, L.F., 1966, \Measurement of Skin Friction at Low Subsonic Speeds by the Raxor-Blade Technique," R & M 3525, Aero. Res. Counc. London. Goldstein, R.J., 1983, Fluid Mechanics Measurements, Hemisphere Publishing Co., p. 61. Haritonidis, J.H., 1989, \The Measurement of Wall Shear Stress," Advances in Fluid Mechanics Measurements, Springer-Verlag, pp. 229{261. Hool, J.H., 1956, \Measurement of Skin Friction Using Surface Tubes," Aircr. Eng., Vol. 28, p. 52. Hussain, A.K.M.F., Reynolds, W.C., \The Mechanics of a Perturbation Wave in Turbulent Shear Flow," AFOSR Scienti c Report, AFOSR 177SR-1655TR. Liu, C., Tsao, T., Tai, Y.C., and Ho, C.M., 1994, \Surface Micromachined Magnetic Acautors," Proceedings, 7th IEEE Workshop on Micro Electro-Mechanical Systems, MEMS'94, Oiso, Japan, pp. 57-62. Liu, C., Tai, Y.C., 1994, \Studies on the Sealing of Micromachined Cavities using CHemical Vapor Deposition Materials," Technical Digest, Solid-State Sensor and Actuator Workshop, Hilton Head Island, South Carolina, pp. 103{106. Lofdahl, L., Stemme, G., and Johansson, B., 1989, \A Sensor based on Silicon Technology for Turbulence Measurements," J. Physics E: Sci. Instrument, Vol. 22, pp. 391{393. Mabey, D.G., Gaudet, L., 1975, \Some Performance of Small Skin Friction Balances at Supersonic Speeds," J. Aircr., Vol. 12, pp. 819{825.
Mastrangelo, C., 1987, Master Thesis, University of California at Berkeley, Berkeley, CA. Petersen, K., Brown, J., and Legtenberg, R., 1985, \High-Precision, High-Performance Mass Flow Sensor with Integrated laminar Flow Micro-Channels," Proceedings, Int. Conf. on Solid-state Sensors and Actuators 1985, pp. 361{363. Preston, J.H., 1953, \The Determination of Turbulent Skin Friction by Means of Pitot Tubes," J. R. Aero. Soc., Vol. 58, pp. 109{121. Rubesin, M.W., Okuno, A.F., Mateer, G.G., and Brosh, A., 1975, \A Hot-Wire Surface Gage for Skin Frictio and Spearation Detection Measurements," NASA TM X-62, p. 465. Sandborn, V.A., 1979a, \Surface Shear-Stress Fluctuations in Turbulent Boundary Layers," Second Symp. on Turbulent Shear Flows, London. Saraswat, K.C., Singh, H., 1982, \Thermal Oxidation of Heavily Phosphorous-Doped Thin Films of Polycrystalline Silicon," J. Electrochem. Soc., Vol. 129, pp. 2321-2326. Schmidt, M.A., Howe, R.T., Senturia, S.D., and Haritonidis, J.H., 1988, \Design and Calibration of a Microfabricated Floating-Element Shear-Stress Sensor," IEEE Trans. ED, 35(6), pp. 750{757. Shajii, J., Ng, K.Y., and Schmidt, M.A., 1992, \A Microfabricated Floating-Element Shear-Stress Sensor Using Wafer-Bonding Technology," J. Microelectromechanical Sys., Vol. 1, No. 2. Stemme, G.N., 1986, \A Monolithic Gas Flow Sensor with Polyimide as Thermal Insulator," IEEE Trans. Elec. Devices, ED-33, p. 1470. Tai, Y.C., and Muller, R.S., 1988, \Lightly-Doped Polysilicon Bridge as a Flow Meter", Sensors and Actuators, Vol. 15, pp. 63-75. Tobata,O., 1986, \Fast-Response Silicon Flow Sensor with an On-Chip Fluid Temperature Sensing Element," IEEE Trans. Elec. Devices, Vol. 33, pp. 361{365. van Oudheusden, B.W., van Herwaarden, A.W., 1990, \High Sensitivity 2-D Flow Sensor with an Etched Thermal Isolation Structure," Sensors and Actuators, A 21{23, pp. 425{430. van Oudheusden, B.W., 1992, \Silicon Thermal Flow Sensors," Sensors and Actuators, A. 30, pp. 5{26.
.
Jin-Biao Huang, Chih-Ming Ho Mechanical, Aerospace and Nuclear Engineering Department University of California at Los Angeles Los Angeles, California
ABSTRACT
We have designed, fabricated, and tested a new type of thermal shear-stress sensor. The sensor consists of a 2 m wide polysilicon thermistor (with a range of lengths, 80-200 m) on a 1.2 m thick silicon-nitride diaphragm, which is on top of a vacuum cavity 2002002 m3 in size. The vacuum cavity provides a good thermal isolation between the resistive element and the substrate; shear-stress sensitivity of 15 V/kPa at a power consumption of 12 mW (constant current = 2 mA) is obtained. The typical thermal time constant of our sensors is 350 s, which translates into a bandwidth of 500 Hz. INTRODUCTION
In an on-going University Research Initiative (URI) project, we are developing a distributed micro-electromechanical system (MEMS) for surface drag reduction in a turbulent ow eld. In our system, sensors will collect 2-D ow- eld information on its surface. This information will be processed, and command signal is passed to microactuators (Liu et al., 1994a) to control the vortices within the boundary layer. Accordingly, the sensors that provide the ow- eld information must have a ne spatial resolution (100 m ) and a fast frequency response (> 1 kHz). Among various possible
ow-sensing schemes, surface shear-stress sensing is the focus of this work.
.
There are many reported shear-stress sensors made by conventional manufacturing techniques. These sensors use either direct or indirect methods for shear-stress detection (Haritonidis, 1989). In the direct method, a tangential force on a surface oating balance gives the shear stress directly (Dhawan, 1953 and Mabey, 1975). In the indirect methods, shear stress is extracted from other physical measurands that are indirectly related to the shear stress. Such indirect measuring devices include Preston tubes (Preston, 1953), Stanton tubes (Hool, 1956 and East, 1966), and hot-wire/ lm surfacemounted sensors (Rubesin et al., 1975 and Sandborn, 1979). These conventional shear-stress sensors, however, can only be hand-made one at a time that we can not use them in our system where at least thousands of sensors are required. To solve this problem, we have looked for answers from silicon micromachining, a new but promising technology. Silicon micromachining, which has evolved from silicon integrated-circuit (IC) techniques, is a new technology that enables both miniaturization and mass production of electro-mechanical devices suitable for our use. In fact, silicon micromachining has already produced many new devices for uid measurements with much improved performance over their conventional counterparts. For example, micromachined surface oating balances for direct shear-stress measurements were reported (Schmidt et al., 1988 and Shajii et al., 1992). The measured sensitivity was 52 V(ac)/Pa for gas sensing using a di erential capacitor readout and 13.7 V/V-kPa for liquid sensing using
DESIGN AND FABRICATION
Fig. 1 shows the schematic design of our shear-stress sensor. A polysilicon resistor (80{200 m long, 2 m wide, and 0.45 m thick) is located at the center of a cavity diaphragm, typically 200200 m2 in size and about 2 m above the bottom of the cavity. The resistors are uniformly doped to a low sheet-resistance value of 50 =2, with typical resistances between 2{5 k . The micromachining fabrication process is shown in Fig. 2. A 200 nm LPCVD silicon nitride is rst deposited on a 4-inch wafer and patterned to de ne the cavities (as 200200 m2 windows in the silicon-nitride layer). The windows are etched down 600 nm into the silicon substrate by wet silicon etchant, and then planarized by the following thermal-oxide growth (1.3 m) on the exposed silicon. A 400 nm sacri cial phosphosilicate glass (PSG) is then deposited and patterned, followed by a blank deposition of 1 m low-stress silicon nitride as the diaphragm material. Next, etching holes are opened in the silicon-nitride layer to expose the sacri cial PSG. The sacri cial PSG and the underlying thermal oxide, if any, is etched by (49%) hydro u-
I
POLYSILICON RESISTOR
Unfortunately, none of these devices can be directly adopted for our project. In this paper, we present our work on the development of a new, surfacemicromachined thermal shear-stress sensor, with the fundamental design of a hot wire on a free-standing silicon nitride diaphragm. The novel aspect of the sensor is that this diaphragm is on a vacuum cavity which minimizes the heat conduction from the diaphragm to the substrate through the gap. From our work, it is found that this approach does signi cantly improve the thermal isolation; an unprecedented shear-stress sensitivity of about 15 V/kPa under constant current drive (2mA, 12 mW) from our micro-sensors is obtained.
oric acid for about 20 minutes. The wafer is then dried, and a silicon-nitride layer (400 nm) is deposited at 300 mTorr (0.04 Pa) to seal the cavity under vacuum (Liu and Tai, 1994b). The thickness of the cavity diaphragm is approximately 1.2 m .
METAL LINE
a piezoresistive readout. Flow sensors based on heattransfer principles (van Oudheusden, 1992; Petersen, 1985; Tai and Muller, 1988) were also demonstrated using free-standing beams (Mastrangelo, 1987 ; Tobata, 1986; Goldberg H.D. et al., 1994), free-standing diaphragms (van Oudheusden, 1990), and low thermalconductivity layer such as polyimide (Lofdahl et al., 1989 and Stemme, 1986) for better thermal isolation. These features were used to solve the common but major problem that the heat loss to the substrate limits the performance of the conventional thermal shear-stress sensors.
I (a)
POLYSILICON RESISTOR
SILICONNITRIDE DIAPHRAGM
CAVITY
(b)
AIR FLOW
silicon substrate
Fig. 1 Schematic (a) top and (b) cross-sectional view (along line I-I) of a thermal shear-stress sensor. A vacuum cavity is created underneath the diaphragm to minimize heat loss to the substrate. Resistors on silicon substrate without the vacuum cavity underneath have also been made.
To form the thermistors, a 450 nm polysilicon layer is deposited and patterned. The following doping is done by ion implantation with phosphorus using a total dose of 1 1016 =cm2 . The wafer is then annealed at 1000 C for 1 hour to activate the dopant and to reduce the intrinsic stress in the as-deposited polysilicon. A 100 nm layer of LPCVD silicon nitride is deposited for passivating the polysilicon resistors to prevent their resistance drift (Saraswat and Singh, 1982). Finally, aluminum metalization forms the leads. Micrographs of the fabricated devices are shown in Fig. 3; since the cavity is held under vacuum, the diaphragm is bent down by the external atmospheric pressure so that optical interference patterns (Newton rings in Fig. 3a) can be seen under the microscope. It is important that the shear-stress sensors should be at so that they do not cause surface-roughness effects such as signal uctuation (Goldstein, 1983). Surface pro les of our sensors are examined using a surface pro lometer (Tencor Instrument Alpha-Step 200) with a stylus force of 4 mg. Fig. 4 shows that, even de ected
by both the stylus force and the atmospheric pressure, the device has a roughness of only 450 nm. This roughness nicely falls within our design speci cation of a few micrometers. In addition, this surface pro le also con rms that the diaphragm is not in contact with the cavity bottom, which is 2 m below the surface.
etching channel Cavity
Resistor
etch hole
nitride
(1)
substrate
(a)
(5) sealing
oxide
(2)
Metal lead
Al CONTACT
(6) PSG
polysilicon 1234 1234
(3)
RESISTOR
(7) silicon nitride
metal 1234 1234
(4)
(8)
Fig. 2 Fabrication steps. (1) Silicon nitride is deposited and windows are opened; (2) fully-recessed local thermal oxidation of silicon (LOCOS) to planarize the surface; (3) sacri cial phosphosilicate glass (PSG) deposition and patterning; (4) low-stress silicon nitride deposition for diaphragms; (5) etch hole opening and sacri cial layer removal; (6) vacuum cavity sealing; (7) polysilicon deposition and patterning to form hot wires; (8) metalization to form leads.
(b)
Fig. 3 Micrographs of a fabricated shear-stress sensor with a 200200 m2 diaphragm. (a) An optical micrograph of the shear-stress sensor with clear interference patterns(Newton ring). (b) An SEM picture of a polysilicon resistor.
WIND TUNNEL SETUP
Shown in Fig. 5, the wind-tunnel is approximately 5 m long, 60 cm wide and 2.5 cm high. It has a maximum mean-stream velocity of about 25 m/s. The sensor package is about 2 2 cm2 in area and is designed such that the sensor surface is ush-mounted with the wind-tunnel side wall. The sensor chip is located in the mid-span on the top wall of the wind tunnel, in a downstream region where turbulence ow is fully developed. The long axis of the resistor bar is perpendicular to the mean- ow direction.
Polysilicon resistor
Stepheight = 450 nm
Diaphragm size = 200 µm
Fig. 4 Surface roughness of a shear-stress sensor as measured by a surface pro lometer with a stylus force of 4 mg. The diaphragm is de ected by about 450 nm at the center due to the combined stylus force and the atmospheric pressure on the diaphragm.
R=2.5181 (1+ 0.001307 t) 3
ga sf
lo w
2.5cm
shear-stress sensor
60 cm
WIND TUNNEL
Resistance (kOhm)
5m
2.8
2.6
20
40
60
80
100
120
140
160
180
200
Temperature (degree C)
Fig. 5 Schematic diagram of the wind-tunnel and the shear-stress sensor mounting.
Fig. 6 Resistance vs temperature measurement for a resistor on the silicon substrate. The resistor is 100 m long, 2 m wide and 450 nm thick. The temperature coecient of the resistor is 0.13 %/ C .
EXPERIMENTAL RESULTS Electrical and Thermal Characteristics
Temperature coecient of resistance (TCR) for the resistors is measured in still-air using a hot chuck. Fig. 6 shows measurement data of the resistance as a function of the temperature . To obtain TCR, we use the relation that R(T ) = R0 (1+ 4T ) to t the data, where 4T is the temperature change and is the TCR. A resistor has a typical positive TCR value of approximately 0.13 %/ C . The next subject is to determine the power biasing. There is a maximum allowable power consumption associated with each sensor. Over the limit, either the diaphragm will fracture due to thermal heating or the resistor will have electrical breakdown. In our experiments, we normally set the bias for the resistors below 15 mW to ensure long-term reliable operations. The performance of the thermal isolation for our shear-stress sensors is evaluated here. For comparison, we xed the size of the resistors used in our experiments (100 m long, 2 m wide and 0.45 m thick). The IV characteristics of the resistor on a nitride diaphragm over a vacuum cavity, over an air- lled cavity and on the silicon substrate (without a cavity) are given in Fig. 7. It is clear that the resistor on the vacuum cavity has the largest non-linearity because the vacuum cavity offers the best thermal isolation.
Current (mA)
3
2
1
Resistor #1 over Silicon substrate Resistor #2 over a vacuum cavity Resistor #2 over an air-filled cavity 0 0
1
2
3
4
5
Voltage (V)
Fig. 7 I-V characteristics of a resistor (100 m long, 2 m wide and 450 nm high) located over a vacuum cavity, over an air- lled cavity and directly over the substrate.
Fig. 8a then shows the resistance as a function of power using the data from Fig. 7 and the resistor's TCR. The solid lines are theoretical ttings with 4T = P and R(T ) = R0 (1 + P ), where is the thermal resistance and P is the power. As a result, the thermal resistances (over the vacuum cavity, over the air- lled cavity, and over the substrate) are quanti ed from the slopes of the lines in Fig. 8b. It is found that of the resistor over the cavity is 2:4 104 C/W , 1:8 104 C/W for over the air- lled cavity, and 2:7 103 C/W for over the substrate. It is clear that a vacuum cavity improves the thermal isolation about 1.33 times over an air- lled cavity, and about 9 times over just the silicon substrate. However, it should be noted that the thermal resistance ratios depend on sizes of the cavity diaphragms and the resistors.
Normalized Resistance ( R / Ro)
1.5
R over vacuum cavity R over air-filled cavity R on substrate
1.4
09
20
.03
0 e=
p
slo
1.3
e= slop
1.2
1.1
29
228
0.0
5
slope = 0.003
Wind-tunnel tests of the micromachined shearstress sensors are conducted using constant-current (CC) mode with the circuit shown in Fig. 10. Fig. 11 shows the measured steady-state sensor voltage output (time average) with respect to the mean-stream velocity under CC mode within a velocity range between 5 to 25 m/s. As shown, the over-all slope of the response is 1.069 mv= (m=s). driving pulse
1 0
2
4
6
8
10
12
14
16
Power (W)
(a)
voltage response
τ1=8 µs τ2=350 µs 4
0 x1
pe
(C
500 µs/div.
500 ms/div.
.4 =2
slo
200
4
e
slop
10 .8x
/W)
=1
100
03 ( C/W)
slope = 2.7x1 0 0
2
4
6
8
10
12
14
(a)
80
(C
Output Voltage (mV)
Temperature Difference ( C)
)
/W
R over vacuum cavity R over air-filled cavity R on direct substrate
300
16
60
40
20
Power (W)
(b) Fig.8 Thermal measurement data converted from Fig. 7. (a) Resistance vs power, and (b) temperature vs power. Frequency Response
In order to measure the time constant of the resistors, we pass a step current through the resistors and observe the change of the voltage drop across the resistor as a function of time (Fig. 9). For a resistor with a positive TCR, the output voltage increases. Experimentally, the response curves are recorded and then tted with exponential rises using as many time constants as necessary. It is found that there are two measurable time constants: 1 the electrical time constant, and 2 , the thermal time constant. Fig. 9a shows that for a resistor 100 m long, 2 m wide and 0.45 m high, 1 is about 8 s while 2 being 350 s. The two corner frequencies in the frequency response curve, 500 Hz and 20 kHz, correspond well with the thermal and electrical time constants. Wind Tunnel Tests
101
102
103
104
105
106
Frequency (Hz)
(b)
Fig. 9 (a) Voltage output for the time-constant measurements in still air for a typical sensor with the polysilicon resistors being 100 m long, 2 m wide and 0.45 m thick. (b) Frequency response curve of the same sensor driven at 1 mA(ac). Rh
+15 V SQ Wave S 10K
1K
Rp
(50K)
Bal (20K)
1K
A1 +
Vout
Fig. 10 The circuit used for constant-current mode windtunnel measurements. Rh is the shear-stress sensor resistor.
In order to correlate the output voltage with the wall shear stress ( w ), two theoretical methods (Hussain, 1970) are used here. In the rst method, w in a fully developed channel ow is related to the streamwise pressure gradient by dPx dx
= , hw ,
(1)
Fig. 13. As the mean-stream velocity increases, the local shear-stress uctuation also increases.
85
80
85
Vout [mv] = 58.12 + 14.89 [mv/ Pa] * T [Pa]
Output voltage (mv)
Output voltage (mv)
Vout [mv] = 57.39 + 1.069 [mv/ m/s] * V [m/s]
75
70
65
80
75
70
65
5
10
15
20
25
0.2
0.4
Mean Flow Velocity (m/s)
where Px is the local pressure, x is the stream-wise coordinate, and h is the half height of the wind tunnel. In the second method, an empirical relationship between the Reynolds number and the wall shear stress in a fully developed channel ow is obtained, by using ,0:089 ,
(2)
and w
= u 2 ;,
1
1.2
1.4
(3)
where u is the shear velocity, u1 is the mean stream velocity, Re is the Reynolds number calculated by using the central-line mean stream velocity, and ; is the air density. We rst measure the stream-wise pressure gradient and calculate w by using Eq. 1. Afterwards, we use the second method given by Eds. 2 and 3 to calculate the wall shear stress again. Good agreement is found between shear-stress values calculated by using both methods. For example, at free stream velocities of 9.96 m/s, 14 m/s, and 20 m/s, the rst method gives w 's of 0.2388 Pa, 0.5384 Pa and 0.945 Pa, while the second method gives w 's of 0.2813Pa, 0.5232 Pa, and 1.029 Pa. This agreement gives us con dence in directly estimating w from the mean stream velocity. Re-plotting Fig. 11 into Fig. 12 with the sensor-output voltage as a function of w using method 1, we nd a typical shearstress sensitivity of 15 V/kPa under constant-current mode for a typical sensor. Fluctuating sensor signals are also recorded since the sensor is subjected to a fully-developed turbulent ow. The root-mean-square (RMS) value of the sensor output is then plotted against the free-stream velocity in
3
2.5
2
1.5
1
5
10
15
20
25
Mean Stream Velocity (m/s)
Root-mean-square voltage vs. mean ow velocity in a fully turbulent ow eld.
Finally, a conventional hot-wire anemometer is placed approximately 1 mm above the shear-stress sensor and the output signals from both sensors are compared to see if there is any correlations. As shown in Fig. 14, the shear-stress sensor output does not have high frequency contents as compared to the hot-wire anemometer. However, it is very clear that major features in the anemometer response do correspond well with those of the shear-stress sensor. Output voltage (2 mv/div.)
1 = 0:1079Re
0.8
Fig. 12 Output voltage as a function of the wall shear stress converted from Fig. 11 using Eq. 1. RMS Output Voltage (mv)
Fig. 11 Output voltage as a function of the wind-tunnel mean- ow velocity for a type-1 shear-stress sensor biased at 12 mW (constant current 2 mA). The resistor is 100 m long, 2 m wide and 0.45 m thick.
u u
0.6
Wall Shear Stress (Pa)
shear-stress sensor
hot-wire anemometer flow
~1mm hot-wire anemometer
Time
shear-stress sensor chip
Windtunnel Wall
(1 ms/div.)
Fig. 14 Simultaneous voltage output from a shear-stress sensor and a hot-wire anemometer placed 1 mm apart. CONCLUSION
A surface-micromachined shear-stress sensor based on heat transfer mechanisms has been developed. A unique vacuum cavity is fabricated underneath the heating element to reduce the heat loss to the substrate. Wind
tunnel tests show a shear-stress sensitivity of 15 V/kPa (without signal ampli cation) for a typical sensor under constant-current mode operation (I=2 mA and P= 12 mW). ACKNOWLEDGEMENTS
This project is supported by AFOSR (under the University Research Initiative program) and ARPA. We thank Mr. Fu-Kang Jiang and Mr. Trevor Roper at the Caltech micromachining lab for help with processing and testing. We also thank Dr. Steve Tung and John Mai at UCLA for wind-tunnel tests. REFERENCES
Dhawan, S., 1953, \Direct Measurements of Skin Friction," NACA TN 2567. East, L.F., 1966, \Measurement of Skin Friction at Low Subsonic Speeds by the Raxor-Blade Technique," R & M 3525, Aero. Res. Counc. London. Goldstein, R.J., 1983, Fluid Mechanics Measurements, Hemisphere Publishing Co., p. 61. Haritonidis, J.H., 1989, \The Measurement of Wall Shear Stress," Advances in Fluid Mechanics Measurements, Springer-Verlag, pp. 229{261. Hool, J.H., 1956, \Measurement of Skin Friction Using Surface Tubes," Aircr. Eng., Vol. 28, p. 52. Hussain, A.K.M.F., Reynolds, W.C., \The Mechanics of a Perturbation Wave in Turbulent Shear Flow," AFOSR Scienti c Report, AFOSR 177SR-1655TR. Liu, C., Tsao, T., Tai, Y.C., and Ho, C.M., 1994, \Surface Micromachined Magnetic Acautors," Proceedings, 7th IEEE Workshop on Micro Electro-Mechanical Systems, MEMS'94, Oiso, Japan, pp. 57-62. Liu, C., Tai, Y.C., 1994, \Studies on the Sealing of Micromachined Cavities using CHemical Vapor Deposition Materials," Technical Digest, Solid-State Sensor and Actuator Workshop, Hilton Head Island, South Carolina, pp. 103{106. Lofdahl, L., Stemme, G., and Johansson, B., 1989, \A Sensor based on Silicon Technology for Turbulence Measurements," J. Physics E: Sci. Instrument, Vol. 22, pp. 391{393. Mabey, D.G., Gaudet, L., 1975, \Some Performance of Small Skin Friction Balances at Supersonic Speeds," J. Aircr., Vol. 12, pp. 819{825.
Mastrangelo, C., 1987, Master Thesis, University of California at Berkeley, Berkeley, CA. Petersen, K., Brown, J., and Legtenberg, R., 1985, \High-Precision, High-Performance Mass Flow Sensor with Integrated laminar Flow Micro-Channels," Proceedings, Int. Conf. on Solid-state Sensors and Actuators 1985, pp. 361{363. Preston, J.H., 1953, \The Determination of Turbulent Skin Friction by Means of Pitot Tubes," J. R. Aero. Soc., Vol. 58, pp. 109{121. Rubesin, M.W., Okuno, A.F., Mateer, G.G., and Brosh, A., 1975, \A Hot-Wire Surface Gage for Skin Frictio and Spearation Detection Measurements," NASA TM X-62, p. 465. Sandborn, V.A., 1979a, \Surface Shear-Stress Fluctuations in Turbulent Boundary Layers," Second Symp. on Turbulent Shear Flows, London. Saraswat, K.C., Singh, H., 1982, \Thermal Oxidation of Heavily Phosphorous-Doped Thin Films of Polycrystalline Silicon," J. Electrochem. Soc., Vol. 129, pp. 2321-2326. Schmidt, M.A., Howe, R.T., Senturia, S.D., and Haritonidis, J.H., 1988, \Design and Calibration of a Microfabricated Floating-Element Shear-Stress Sensor," IEEE Trans. ED, 35(6), pp. 750{757. Shajii, J., Ng, K.Y., and Schmidt, M.A., 1992, \A Microfabricated Floating-Element Shear-Stress Sensor Using Wafer-Bonding Technology," J. Microelectromechanical Sys., Vol. 1, No. 2. Stemme, G.N., 1986, \A Monolithic Gas Flow Sensor with Polyimide as Thermal Insulator," IEEE Trans. Elec. Devices, ED-33, p. 1470. Tai, Y.C., and Muller, R.S., 1988, \Lightly-Doped Polysilicon Bridge as a Flow Meter", Sensors and Actuators, Vol. 15, pp. 63-75. Tobata,O., 1986, \Fast-Response Silicon Flow Sensor with an On-Chip Fluid Temperature Sensing Element," IEEE Trans. Elec. Devices, Vol. 33, pp. 361{365. van Oudheusden, B.W., van Herwaarden, A.W., 1990, \High Sensitivity 2-D Flow Sensor with an Etched Thermal Isolation Structure," Sensors and Actuators, A 21{23, pp. 425{430. van Oudheusden, B.W., 1992, \Silicon Thermal Flow Sensors," Sensors and Actuators, A. 30, pp. 5{26.
