Development and characterization of surface ... - Semantic Scholar
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
979
Development and Characterization of Surface Micromachined, Out-of-Plane Hot-Wire Anemometer Jack Chen, Member, IEEE, and Chang Liu, Senior Member, IEEE
Abstract—In this paper, we report the development of a new type of hot-wire anemometer (HWA) realized by using a microfabrication process that combines surface micromachining and an efficient three-dimensional assembly technique. The HWA uses a thermal element (hot wire) that is made of Pt/Ni/Pt film with a measured temperature coefficient of resistance (TCR) of 2400 ppm/ . The thermal element is elevated out of plane by using support beams made of polyimide. In our current design, the thick and up to 1 mm tall, and the length support beam is 2.7 to 200 . Steady-state of the thermal element varies from 50 response to air velocity has been experimentally obtained up to 20 m/s under both constant current (CC) and constant temperature (CT) modes. The transient-state response has been examined using square wave and sinusoidal wave excitation signals in CT modes with the maximum cutoff frequency found to be approximately 10 kHz. This new HWA offers a number of unique materials and performance characteristics. The sensor does not require the use of silicon as either substrate or sensing materials. Using this process, it is possible to form large arrays of HWA on a variety of substrate materials. [910]
C
m
m
m
I. INTRODUCTION
H
OT-WIRE anemometry (HWA) is a well-studied technique for measuring the velocity of fluid flow. It utilizes a thermal element that serves as both a Joule heater and a temperature sensor. Under a constant bias power and zero flow rate, the temperature of the thermal element reaches a steady-state value. If external fluid flow is present around the thermal element, the thermal element will experience forced convective cooling. Accordingly, the temperature of the thermal element provides the means to gauging the cooling rate and the flow velocity. Conventional HWA are assembled individually by mounting a thin wire made of platinum or tungsten onto support prongs. The wires may be thinned (e.g., by etching in acidic solutions) until the desired dimensions are reached (typically a few mm long and a few micrometers in diameter). This active portion of the sensor is then mounted on a long probe with electrical connection for ease of handling. The schematic diagram of a typManuscript received July 26, 2002; revised August 8, 2003. This work was supported by the National Science Foundation under the CAREER program (NSF Award 9984954) and the Smart Skins program (NSF Award 0080639) and also in part by the Air Force Office of Scientific Research under the Bioinspired Concepts program. Subject Editor C.-J. Kim. J. Chen is with the Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801 USA and also with the 319 B Micro and Nanotechnology Laboratory, Urbana, IL 61801 USA (e-mail: [email protected]). C. Liu is with the Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801 USA. Digital Object Identifier 10.1109/JMEMS.2003.820261
Fig. 1. Schematic diagram of a conventional hot-wire anemometer.
ical finished device is shown in Fig. 1. The HWA is noted for its low cost, fast response (in the kilohertz range), small sizes, and low noise [1]. Conventional HWAs suffer from two major shortcomings, however. First, the fabrication and assembly process is delicate and does not guarantee uniformity of performance. Secondly, it is prohibitively difficult to form large arrays of HWA for measuring flow distribution. In the past few years, efforts have been made to apply various micromachining processes to realize HWA for air flow with smaller dimensions, better uniformity, and faster time response (due to smaller thermal masses possible). One strategy is to use the bulk-micromachining technique to produce freestanding cantilever structures. For example, researchers have reported making prongs and hot-wires out of doped polycrystalline silicon and releasing the cantilever by partially removing the silicon substrate [2]. One group used doped silicon for both the prong and the hot-wire along with polyimide hinge to make three-component hot-wire sensors [3]. Another group undercut a polysilicon hot-wire with wet etching so the hot-wire is suspended over a cavity for better thermal insulation [4]. To our knowledge, almost all existing micromachined HWAs for air flow measurement use silicon as the thermal element and involve bulk micromachining of silicon to some extent. Silicon is a relatively expensive material. The doping of silicon (e.g., for creating hot wire), the etching of silicon, as well as the packaging of individual silicon dies require significant expertise and efforts. Most of the micromachined HWAs cannot be realized in large array format efficiently. Researchers have reported using
1057-7157/03$17.00 © 2003 IEEE
980
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
metal thermal element lying on polyimide bulk film [5]; however, insufficient thermal isolation and elevation limits use. Our research is motivated by the needs to reduce the cost of HWA, to allow efficient fabrication and packaging, and to produce arrayed HWAs on potentially flexible substrates. Two strategic approaches are taken. First, we realize HWA by using surface micromachining in conjunction with three-dimensional assembly methods. This circumvents the use of bulk micromachining, which requires relatively long etching time. Bulk etching using anisotropic wet etchants frequently pose concerns of materials compatibility as all materials on a given substrate are required to sustain wet etching for long periods (several hours to etch through typical silicon wafers). Surface micromachining also enables more efficient assembly and allows formation of large arrays of HWAs. Second, we realize thermal elements using nonsilicon, temperature sensitive materials. This could reduce costs as the use of silicon bulk (as substrate) or thin film (as hot wire) is not required. By substituting silicon doping and bulk etching steps, the fabrication process can be realized in more efficient manner. In this work, we applied a new 3-D assembly process, which pairs magnetic actuation with deformable metal hinges [6], to fabricate a hot-wire anemometer using surface machining of metal and polymer materials. The fabrication process was designed carefully so that 1) the maximum temperature required and 2) no etching throughout the process flow is under 350 using concentrated hydrofluoric (HF) acid was necessary. By limiting the overall process temperature, the process can be run on a broad range of substrates, including silicon, glass, and even certain plastics. It should be noted that a prototype surface-micromachined, out-of-plane HWA had been made using hinged polysilicon structure [7], but the assembly to vertical position was done by manual probing and test results were not presented. II. ANALYSIS AND DESIGN The schematic diagram of the new out-of-plane anemometer is shown in Fig. 2. A thermal element is elevated from the substrate to a predetermined height that corresponds to the length of the support prongs. By elevating the thermal element away from the bottom of the velocity boundary layer, the thermal element experiences greater fluid flow velocity and exhibits greater sensitivity. The thermal element is electrically connected to the substrate through the support prongs as well. The new HWA will work as flow sensor whether the thermal element is beyond the boundary thickness or not. In the case the thermal elements are sufficiently tall to stand beyond the boundary layer, the measurement will directly indicate true free-stream velocity. If the thermal element is embedded in the boundary layer, it will provide direct measurement of true local flow speed, which can be used to infer free-stream velocity based on known boundary layer velocity profile, if necessary. An HWA operates by sensing temperature change of a hot-wire resulting from forced convection. The temperature variation can be inferred from the change of the resistance of the sensing element. The resistance of a hot wire is related to its temperature according to (1)
Fig. 2. Schematic diagram of a single out-of-plane HWA. The thermal element is made of metal thin film and support by two support beams, which also provide electrical leads. The thermal element is elevated away from the substrate surface (and the bottom of the velocity boundary layer).
where is the nominal resistance under a reference temperature (e.g., room temperature), and is the temperature coefficient of resistance (TCR) of the thermal element [8]. For metal, the coefficients for second order and beyond are several orders will be conof magnitude smaller than . In this paper only sidered and will be generally referred to as . We briefly discuss the theoretical basis of the HWA sensor to motivate our design. The heat balance equation for the thermal element under electrical Joule heating is [9]–[11] (2) is the rate of heat storage, is the generated (bias) where is the rate of heat loss due power from Joule heating, represents the sum of conducto forced convection, and tive losses (e.g., through support prongs). For a HWA, the term involves (1) end loss, which is unavoidable in the presence of support prongs, as well as (2) longitudinal thermal conduction along the thermal element. Under a given set of bias and fluid flow rate, it is important to maximize power while minimizing in order to obtain greater sensitivity to velocity changes. To minimize the conductive heat loss also means that the HWA sensor can be operated in a more thermally efficient manner. This is especially important if a large array of such sensors is used. Several dimensional and materials parameters are relevant relative to [12]. Assuming that the for increasing thermal element is a very long wire with a cylindrical cross section, the ratio between conductive and convective heat transfer is (3)
with variables defined in Table I. Although the formula provided above is derived based on an ideal case of infinitely long thermal elements and does not apply to thermal elements with finite length, they nonetheless provide valuable design guidelines.
CHEN AND LIU: SURFACE MICROMACHINED, OUT-OF-PLANE HOT-WIRE ANEMOMETER
981
TABLE I NOMENCLATURE
An important design parameter is the length of the hot wire . It is possible to increase relative to by invaries slowly creasing the length of the hot wire, because changes roughly linearly with , whereas the value of with respect to . However, is limited by fabrication practicality and yield considerations. The longer the thermal elements, the more difficult it is to realize and withstand high flow. Another important design parameter is . In the common case that the thermal element has a rectangular cross-section, the term is the equivalent diameter that yields the cross-section area. By reducing the value of , the surface-to-volume ratio of the hot wire is increased, hence encouraging more convection while confining the conductive component. However, there exists a practical limit to the minimal diameter of the hot wire as well, due to mechanical rigidity concerns. Our current study is aimed at establishing the fundamental design, process feasibility, and reliability. It was not our intention to conduct exhaustive optimization of the dimensional and materials parameters at this stage. Nevertheless, a wide range of dimensional parameters has been accommodated in our proto, 100 , type design. Devices with hot-wire lengths of 50 , and 200 , as well as heights up to 1000 can be 150 fabricated. The cross-section area of the hot wire is a composite of a high TCR metal on an insulating polyimide beam. The cross wide by 0.12 thick, while section of the metal film is 4 wide by 2.7 thick. The polythe polyimide beam is 6 imide piece provides the mechanical support. The cross section of this composite thermal element is comparable to that of commercially available hot wire sensors. III. FABRICATION PROCESS The first step of developing the fabrication process is to select suitable materials from the perspectives of both performance requirements and process simplicity and compatibility. We must first choose materials for the hot wire that provides a high TCR.
TABLE II TABLE OF MATERIAL PROPERTY. ALL VALUES ARE CITED FROM [22] EXCEPT FOR THE THERMAL CONDUCTIVITY OF POLYIMIDE, WHICH IS CITED FROM [23]
We first tried to use Pt as the hot-wire filament because of the high TCR value of 3900 ppm/ . However, the measured electrical resistivity of the evaporated 1000- -thick Pt was several times the theoretical value (see Table II) possibly due to electron scattering at grain boundary [13]. This results in a reduced effective TCR of 780 ppm/ . This could be improved by annealing to increase grain size. However, the annealing , which is incompatible temperature of Pt is above 600 with integrated circuits and the polyimide film [14]. We have therefore chosen Ni as the primary thermal element material, because the experimentally measured TCR of as deposited Ni ) is greater than what thin films (approximately 3750 ppm/ we measured for Pt. This TCR value is also higher than using a polysilicon one, which has reported value of only 1460 ppm/ [15] or 800 ppm/ [3]. The hot wire consists of temperature-sensitive metal thin film on a supporting polyimide piece. The polyimide support is used because it provides the hot wire with needed structural rigidity without increasing thermal conductance of the thermal element. The thermal conductivity of polyimide is low, almost
982
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
three orders of magnitude lower than that of a metal, e.g., nickel (see Table II). In this work, the thickness of the polyimide due to processing concerns. If the thin film is roughly 2.7 thickness is much lower than this value, the mechanical rigidity will likely be degraded. On the other hand, if the thickness is much greater, there is concern that the polyimide support will decrease the frequency response of the HWA due to added thermal mass. The fabrication process utilizes an efficient 3-D assembly method called the Plastic Deformation Magnetic Assembly (PDMA), which was discussed in detail in [6]. A brief discussion of this method is provided in the following. The PDMA process utilizes surface micromachined structures that are anchored to substrates with cantilever beams made of ductile metal materials (e.g., gold and aluminum). The microstructure is attached to pieces of electroplated ferromagnetic material (e.g., Permalloy). By applying an external magnetic field, the ferromagnetic material is magnetized and interacts with the field to bend the microstructure out of plane. If the amount of bending is significant, the cantilever support hinges will be plastically deformed, resulting in permanently bent microstructures even after the magnetic field is removed. The process is very efficient and can be realized in parallel on the wafer scale. The overall fabrication process is shown in Fig. 3. A precursor of the process has been discussed previously [16]. Numerous improvements have been implemented since then. For example, was measurement of flow responses at high velocity not attempted using the first-generation devices due to concerns toward the fragility of support hinges. Later, we improved upon the process so that the sensors are based on strengthened mechanical supports and can sustain higher wind velocity. The starting wafer is silicon. However, the process can be performed on glass or polymer substrates, as the overall temperature of the process is intentionally kept low. First, a chrome/copper/titanium metal stack is evaporated and patterned as the sacrificial layer [see Fig. 3(a)]. The 10-nm-thick chrome film serves as the adhesion layer. The 250- -thick titanium thin film reduces the in-process oxidation of the 2500- -thick copper film [17] during the subsequent polyimide curing. A 2.7- -thick photo-definable polyimide (HD-4000) is [ see spun-on, patterned via lithography, and cured at 350 Fig. 3(b)] for 2 h. This polyimide layer forms the support prong and part of the hot wire. A Cr/Pt/Ni/Pt film is then evaporated and patterned to form the thermal element [see Fig. 3(c)]. The thickness of the Cr layer, which serves as an adhesion layer, is 200 . A 800-thick Ni resistor is sandwiched between two 200- -thick Pt films, which are used to reduce possible oxidation of Ni while in operation because Pt is relatively inert at high temperature of operations. Optionally, the Pt can be patterned separately to complete cover the Ni, including sidewalls, to prevent long-term degradation due to oxidation We then evaporate and pattern a 5000- -thick Cr/Au film [see Fig. 3(d)] to serve as a mechanical bending element as well as electrical leads of the hot-wire filament. We electroplate a 4- -thick Permalloy thin film on portions of the cantilever support prongs [see Fig. 3(e)].
Fig. 3. The fabrication process of a surface micromachined hot-wire anemometer: (a) deposition of sacrificial layer, (b) pattern polyimide structure, (c) deposit nickel sensor, (d) deposit gold, (e) electroplate permalloy, (f) strip electroplating mold, (g) pattern photoresist (PR) layer for postrelease plating, (h) sacrificial layer etching, and (i) PDMA assembly.
Sacrificial layer release is performed by using a solution containing acetic acid and hydrogen peroxide to selectively remove the copper thin film. PDMA assembly is carried out to lift the entire sensor out of plane [see Fig. 3(i)] by placing a permanent magnet (field strength 800 Gauss) at the bottom of the substrate. To finish the process, the device chip is then rinsed in deionized water and dried.
CHEN AND LIU: SURFACE MICROMACHINED, OUT-OF-PLANE HOT-WIRE ANEMOMETER
983
to the substrate while Ni is locally electroplated (see Fig. 5). The plating time is approximately 5 min at 100 per thick Ni at the device, which gives us approximately a 3–4 hinge. At present, this plating process is not optimized and is performed one device at a time. However, with proper wiring layout and plating control, we expect this process to be scalable to the wafer level. An SEM micrograph of a representative device after release and assembly is shown in Fig. 6. Fig. 7 shows the SEM of the Au hinge with and without the post-release plating. The improved device after plating was tested at wind speed up to 25 m/s without failure. Higher speed has not been used because of limitation in test instruments. IV. STEADY-STATE OUTPUT RESPONSES Fig. 4. Au-polyimide separation, shows Ti as a poor adhesion layer.
The adhesion between the Au to the polyimide is very important. Without adequate adhesion, the polyimide and the gold film would separate during the PDMA assembly. One of the ways to improve adhesion was to use an adhesion layer reactive ion etching and treat the polyimide layer by using (RIE) before the metal deposition. Cr seems to be the adhesion layer of choice, and the RIE treatment creates a hydrophilic structure on the polyimide surface that enhances adhesion [18]. Ti was initially used as the adhesion material for both the Au and the hot-wire element due to its good stability in chemical etchants and higher electrical resistivity. However, we found that the metal layer peels off from the polyimide during PDMA assembly process. A scanning electron microscopy (SEM) image of this effect is shown in Fig. 4. The curling of the polyimide is due to the intrinsic stress created during curing. The straight Au frame is maintained because of the low-stress nature of electroplated Permalloy. No such separation was observed after Cr was used as the adhesion layer with the RIE treatment. Both the thermal element and the support prongs have relatively small frontal area. The momentum imparted by fluid on the sensor is minimal. We have demonstrated that the sensor after the PDMA assembly can withstand airflow with mean stream velocity lower than 5 m/s without being damaged. However, at such velocity the support prongs visibly vibrate. Other than low-speed applications, it is imperative that the ductile bent hinge is reinforced to withstand the force due to the airflow. Toward this end, we added an additional post-release Ni electroplating step. Ni is selectively electroplated at the bent hinges to strengthen the hinges. Before the removal of the Cu sacrificial layer, photoresist is spun on and patterned to prevent electroplated Ni from plating on the sensing hot wire [see Fig. 3(g)] The only metal exposed is the Au hinges and the electroplating contacts. The sample is then put into the Cu etchant until the Cu sacrificial layer is completely undercut. Afterwards, the sample is removed from the etchant solution, thoroughly rinsed in deionized water, and immediately placed into a Ni plating bath. External magnetic field is applied normal
We tested the steady-state performance of the sensor in a wind tunnel. The fabricated hot-wire anemometer chip is attached to a PC board with etched wiring traces. The PC board is in turn placed on a stage located in the middle of a wind-tunnel test section (see Fig. 8). The chip is located near the edge of the packaging, with the hot-wire element located only 1 mm away from the leading edge. We have characterized the response of HWAs in both constant current and constant temperature modes. Because the HWA are located close to the leading edge, the even boundary layer at that distance is very small for very low Reynolds number flow. The steady-state flow measurement tests were conducted under laminar flow with Reynolds number between 100 to 2000. For higher flow rate during wind tunnel measurement, the thermal element is above the boundary layer. The TCR value of the thermal element (Cr/Pt/Ni/Pt film and 100 . stack) is found to be 2400 ppm/ between 25 It should be noted that this TCR value is lower than cited TCR values of all compositional metal films (list in Table II). This discrepancy could be contributed by many factors. However, a detailed discussion and resolution of this issue is outside the scope of this paper. Steady-state output of an infinitely long, cylindrical hot wire ) should follow King’s law. For an HWA that has (large ratio, we follow the noncylindrical geometry and smaller common practice of collapsing the experimental hot-wire data in the form (4) and are the temperature in the sensor and fluid, where is the resistance of the hot-wire sensor. The respectively, and term is the overall heat transfer coefficient (5) where is a fitting factor that accounts for the smaller aspect ratio and wire geometry [9], [12], [19]. The heat transfer coefficient accounts for the conduction end loss as well as natural is the forced convection term. convection, and
984
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
Fig. 5. Electroplating in a bath under microscope to reinforce hinge: (a) Device placed in bath and external magnetic field applied to bring the structure to 90 , (b) power supplied to the hinge and plating started at 2 mA for 5 min, and (c) once plating is completed, sample is removed from the bath and resist is stripped.
(a) (a)
(b) Fig. 6. SEM micrographs of (a) an HWA immediately after sacrificial release and PDMA assembly, and (b) an HWA after postrelease strengthening using electroplating.
(b) Fig. 7. SEM graphs of the plastically deformed hinge (a) without postrelease Ni plating and (b) with postrelease plating.
A. Constant Current Response In the constant current (CC) mode, the response can be derived from (5) by holding the current constant. A calibration equation for constant current mode can be written as (6a) where (6b) (6c)
The terms and are velocity independent constants, and is the change in voltage drop across the hot-wire. The is defined as the resistance overheat ratio term at velocity . A CC driving circuitry is shown in Fig. 9(a), using an LM334 current source in conjunction with a diode to reduce temperature dependence of the circuit. Due to a positive TCR, the voltage across the hot wire decreases as the flow rate increases. The sensitivity of output voltage with respect to air velocity increases with increasing overheat ratio because of the increase in temperature difference between hot wire and the fluid.
CHEN AND LIU: SURFACE MICROMACHINED, OUT-OF-PLANE HOT-WIRE ANEMOMETER
985
Fig. 8. Photo of the wind tunnel measurement setup. An anemometer chip is placed on a glass holder that is positioned on a stage located in the test section. The finished package is mounted onto the test fixture. The yaw and pitch angle of the chip can be controlled from outside of the wind tunnel.
Fig. 10. The response of a 400-m-high and 200-m-long, hot-wire anemometer at various overheat ratio (defined at U = 0 m=s) operating in the constant current mode. Solid lines are the curve fit using (6a)–(6c). The air velocity U is calibrated using the Traceable hot-wire anemometer.
(a)
(b)
Fig. 9. (a) Constant current circuit, LM334 is an IC current mirror. (b) Constant temperature circuit. Bridge gain G R =R + 1, R R ,V and R not used during steady-state operation.
=
The output response of a representative hot wire that is long and 400 height is shown in Fig. 10. The three 150 . The current-overheat curves correspond to three different correspondence is determined by voltage-current division at every data point with an Agilent 4155 parameter analyzer, is obtained at 5 mV. Curve fitting with (6a) verifies where good correspondence between the measured data and the theoretical relation at velocity above 3 m/s, with equals 1. At low velocity, because a larger portion of the heat transfer is due to natural convection, the deviation between the data and curve fit is greater. As expected from (6a)–(6c), the voltage drop is larger at a high overheat, but flattens out in all three overheats at high velocity. Velocity response of HWA at various lengths operating at is show in Fig. 11, curve fit with . A comparratio is more revealing. From (5), is not a ison of the function of flow velocity, but of the wire geometry. The higher the B/A ratio, the greater the heat loss attributed to forced con, and it has vection. The longest micromachined HWA is 200
Fig. 11. The response of 400-m-tall hot-wires of various lengths operating in the constant current mode. The devices operate at resistance overheat a = 0:3. The heat transfer ratio B=A defined in (5) is determined from B=A = D =(1 + a ).
a ratio of 0.11, compared to the shortest HWA of 100 , which has a B/A ratio of 0.063. This comparison is in agreement ratio will increase with length with (3), where because there is more loss from forced convection. Commercial HWA has a B/A of around 0.15, which is higher than the micromachined ones made here [19]. This is because most com, while maintaining mercial HWA has a length of over 1000 a wire diameter of a few microns. However, commercial HWA has a geometric exponential factor of 0.5, which means more degradation at higher velocity, compare to the micromachined HWA presented here. For all HWA of various lengths tested, the ratio of the same hot-wire operating at various does not change significantly. B. Constant Temperature Response The steady-state equation for constant temperature (CT) mode operation can be solved in a similar way as the CC mode. is placed on one leg of a Whetstone The hot-wire sensor is kept constant by the feedback bridge. In CT mode, op-amp circuit shown in Fig. 9(b).
986
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
m
m
Fig. 12. Response of a 400- -high, 200- -long, hot-wire anemometer operating in the constant temperature mode. The bridge ratio R =R of the CT circuit is 10, and overheat ratio a is set at 0.1, 0.2, and 0.3.
The bridge circuit is balanced when . The reis a variable resister used to set the hot-wire temperasistor ture (7) is the resistance overheat The term at all air velocity , and is the resistance of the hot-wire at room temperature. Rewriting (4), it can be shown that the voltage output will vary with flow speed according to the relation below:
m
Fig. 13. The response of 400- -tall hot-wires of various lengths operating in the constant temperature mode. The devices operates at resistance overheat a = 0:3. The heat transfer ratio B=A = D =C .
of the hot wire and the conductive thermal resistance. Commercial HWA typically have bandwidth of a few hundred hertz in CC mode, and can approach tens of kilohertz in CT mode. Frequency response for CT operation also has two methods of supply. electrical testing, sine wave or square wave using the The schematic diagram of the constant temperature circuit is shown in Fig. 9(b). Assuming the HWA has only one time constant and the op-amp has a single pole at , the output voltage is derived [2], [20] as a second-order transfer function
(8)
(10a)
where with (9) is the voltage at The terms and are constants from (5), , and is the bridge gain defined as . At low fluid velocity, the calibration curve does not correspond well with the data because of free convection. The CT output response of the same HWA in the CC test (Fig. 10) with curve fitting from (8) is shown in Fig. 12. The fitted lines are the theoretical curves, and they agree well with the measured data except at very low velocity. Curve fitting gives a value in the CT mode, compare to reported of by Jiang et al. [2]. The velocity response of HWA with length to 200is plotted in Fig. 13 operating varying from 100. To compare the three HWAs, the at an overheat ratio between the heat transfer coefficient of conductive loss and force convection is used. The B/A ratio increases from 0.063 for the 100- -long HWA to 0.13 for the 200- -long HWA. V. TRANSIENT RESPONSES When there is a sudden change in the fluid velocity, the output signal will lag behind. It is important to understand the valid frequency range of the HWA, especially for high Re flow measurement. The time constant is determined by the thermal mass
(10b) (10c) is the dc voltage gain, is the velocity sensitivity, The term is the quality factor, is the time constant of the hot-wire, is and are the op-amp gain and pole, and overheat ratio, and are the small signal input voltage and velocity. The small and velocity disturbance are signal voltage disturbance decoupled in the numerator of (10a). Therefore, by analyzing in the frequency domain, we can the HWA response due to obtain the HWA frequency limit with velocity fluctuation. This is beneficial because voltage disturbance is simpler to generate than velocity disturbance. The responses of the CT circuit with a 200- -long HWA to a small signal voltage at constant velocity are shown in Fig. 14. is flat up to , then increases at 20 The value of at the peak, dB/decade until the it reaches a approximately beyond which it asymptotes at 20 dB/decade. measured at the peak corresponds to the The second pole corner frequency of a voltage output with a velocity fluctuation
CHEN AND LIU: SURFACE MICROMACHINED, OUT-OF-PLANE HOT-WIRE ANEMOMETER
987
(a)
Fig. 15. Yaw angular response of a 150-m-long, 600-m-high HWA, the velocity reading is based on wind tunnel calibration using (8). The measured data points are placed along side solid line fitted using the cosine relation (10).
(b)
(c)
m
Fig. 14. Frequency response of a 200- -long hot-wire in CT mode with a sine wave disturbance under the following conditions: (a) R =R = 10, = 0:3, U = 2 m=s; (b) R =R = 10, = 0:3, U = 20 m=s; (c) R =R = 1, a = 0:3, U = 20 m=s. Time constant measured with square wave disturbance is embedded in each figure and corresponds to f = 1=(1:3 1 ).
, making the sine wave response a useful measurement for the cutoff frequency (for velocity fluctuation). Beyond , the output due to will drop off at 40 dB/decade. In order to ensure a flat response, the HWA in CT mode should not operate in turbulence near . Using the frequency response plotted in Fig. 14(a) and (b), we can see that increases slightly from 4 dekHz to 5 kHz as the velocity is increased. As the creases [see Fig. 14(c)], increases to 9 kHz. However, this is at the expense of the velocity sensitivity. Another way to measure the cutoff frequency is by examwith a square wave . The square wave test assumes ining varying the heating current can represent the heating and cooling of the HWA with velocity fluctuation. For example, at the onset of a voltage increase, the feedback circuit will try to balance the bridge by decreasing the voltage output (and hence the heating current), this is similar to a sudden decrease in fluid velocity. to settle back to 3% of peak value is The time it takes for defined as , and can be related to by [21] (11) The corresponding square wave response for each sine wave test clearly marked. The is imbedded in the bode plot with the responses from square wave testing verify the cutoff frequency found in all three cases with the sine wave test. This cutoff frequency of new HWA sensors is comparable to that of a commercial anemometer.
Fig. 16. Micrograph of an array of two-component HWA sensor fabricated on a sheet of Kapton thin film.
VI. ANGULAR RESPONSE Provided a sufficiently large aspect ratio for the HWA filament, the yaw response should follow the cosine relation (12) The yaw response is tested using a rotating stage setup shown in Fig. 8. Constant temperature mode is used because of the higher flow sensitivity associated with this mode. In the measurement, and 50 at 5 HWA output is measured at angle between intervals. This voltage output is then converted to velocity using (8) and then plotted in Fig. 15, along with a fitted line based on the cosine relation. The yaw response of the 150- -long HWA seems to fit the cosine relation fairly well. Because the thermal element is made of a composite film, the response to reverse flow is not expected to be symmetrical. This problem can be circumvented by device calibration, by using an array of sensors, or by sandwiching the metal in between two polyimide layers. VII. COST ESTIMATE The main difference between the bulk-micromachined HWA and the surface micromachined HWA approach is the expected cost. The new HWA promises to reduce the costs for realizing arrayed flow sensors. Conventional Silicon micromachined sensors are associated with greater substrate costs and longer time for processing and assembly. Some commercial hot-wire arrays are made by depositing a high TCR material on a thermal insulating Kapton film. As a comparison, we show in Fig. 16 a two-dimensional array of
988
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
HWA fabricated on a flexible Kapton film. The ability to fabricate large arrays on flexible Kapton film, coupled with the low temperature nature of the process, means this type of HWA can potentially be fabricated using roll-to-roll, high throughput processing. VIII. CONCLUSION We have designed and developed an efficient microfabrication process for realizing a novel HWA. The HWA can be fabricated using nonsilicon materials and hence can be potentially produced with low costs and high fabrication efficiency. The HWA can potentially be realized in large array format for distributed flow sensing. If necessary, the sensor can be made on flexible substrates for conformal coating of fluid dynamic surfaces of interests. The sensor has been systematically characterized under both constant current and constant temperature modes. The time constant of a sensor is 10 kHz under constant temperature mode. The main objective of this work is to establish the design and fabrication process, and to characterize the performance to establish the proof-of-concept and performance merits. We have developed a new HWA sensor that offers unique functional advantages over existing hot wire sensors. Comprehensive optimization of design, materials, and associated fabrication methods are quite involved and are not exhaustively addressed in this work. Such issues will be systematically explored to incrementally improve the performance, ease of fabrication, and reliability.
[8] J. O. Hinze, Turbulence: And Introduction to Its Mechanism and Theory. New York: McGraw Hill, 1959, p. 78. [9] C. G. Lomas, Fundamentals of Hot-Wire Anemometry. Cambridge, U.K.: Cambridge University Press, 1986. [10] A. E. Perry, Hot-Wire Anemometry. Oxford, U.K.: Clarendon, 1982. [11] R. J. Emrich, Ed., Methods of Experimental Physics: Fluid Dynamics Part A. New York: Academic Press, 1981, pp. 259–314. [12] H. H. Brunn, Hot-Wire Anemometry: Principles and Signal Analysis. Oxford, U.K.: Oxford University Press, 1995. [13] A. F. Mayadas, M. Shatzkes, and J. F. Janak, “Electrical resistivity model for polycrystalline films: The case of specular reflection at external surfaces,” Applied Physics Letters, vol. 14, p. 343, 1969. [14] S. Sedky et al., “Experimental determination of the maximum postprocess annealing temperature for standard CMOS wafers,” IEEE Trans. Electron Devices, vol. 48, pp. 377–385, Feb. 2001. [15] F. Jiang, Y. C. Tai, C. M. Ho, and W. J. Li, “A micromachined polysilicon hot-wire anemometer,” in Technical Digest, Solid-State Sensor and Actuator Worlshop, Hilton Head Island, SC, 1994, pp. 264–267. [16] J. Chen, J. Zou, and C. Liu, “A surface micromachined, out-of-plane anemometer,” in Proceedings MEMS, Las Vegas, 2002, pp. 332–335. [17] S. A. Chambers and K. K. Chakravorty, “Oxidation at the polyimide/CU interface,” J. Vacuum Sci. Technol. A, vol. 6, no. 5, pp. 3008–3011, Sept. 1988. [18] Y. Nakamura, Y. Suzuki, and Y. Watanabe, “Effect of oxygen plasma etching on adhesion between polyimide films and metal,” Thin Solid Films, vol. 290, pp. 367–369, 1996. [19] F. Jiang, “Silicon Micromachined Flow Sensors,” Ph.D. dissertation, California Inst. Technol., Pasadena, CA, 1998. [20] P. D. Weidman and F. K. Browand, “Analysis of a simple circuit for constant temperature anemometry,” J. Phys. E—Sci. Instrum., vol. 8, no. 2, pp. 553–560, July 1975. [21] P. Freymuth, “Frequency response and electronic testing for constant temperature hot-wire anemometers,” J. Phys. E—Sci. Instrum., vol. 10, pp. 705–709, 1977. [22] G. T. A. Kovacs, Micromachined Transducer Sourcebook. New York: McGraw-Hill, 1998. [23] Pyralin Polyimide Coating for Electronics: PI2610 Series—Product Information and Process Guidelines, HD Microsystems, 1998.
ACKNOWLEDGMENT The authors wish to acknowledge the technical support of J. Hughes, H. Romans, and R. Blaney for maintaining processing equipment at the Micro and Nanotechnology Laboratory of the University of Illinois, and the technical support of A. Broeren at the Department of Aeronautical and Astronautical Engineering at University of Illinois. REFERENCES [1] R. J. Goldstein, Ed., Fluid Mechanics Measurements. Washington, DC: Hemisphere, 1983, pp. 99–145. [2] F. Jiang, Y. C. Tai, C. M. Ho, R. Karan, and M. Garstenauer, “Theoretical and experimental studies of micromachined hot-wire anemometers,” in Technical Digest, International Electron Devices Meeting(IEDM), San Francisco, CA, 1994, pp. 139–142. [3] T. Ebefors, E. Kalvesten, and G. Stemme, “Three dimensional silicon triple hot-wire anemometer based on polyimide joints,” in Proc. MEMS, New York, 1998, pp. 93–98. [4] T. Neda, K. Nakamura, and T. Takumi, “A polysilicon flow sensor for gas flow meters,” Sens. Actuators A, Phys., vol. A54, no. 1–3, pp. 621–631, June 1996. [5] D. Dittmann, R. Ahrens, Z. Rummler, K. Schlote-Holubek, and W. K. Schomberg, “Low-cost flow transducer fabricated with the AMANDAprocess,” in Proc. Transducer, Munich, Germany, 2001, pp. 1472–1475. [6] J. Zou, J. Chen, C. Liu, and J. E. Schutt-Aine, “Plastic deformation magnetic assembly (PDMA) of out-of-plane microstructures: Technology and application,” J. Microelectromech. Syst., vol. 10, no. 2, pp. 302–309, June 2001. [7] K. S. J. Pister, M. W. Judy, S. R. Burgett, and R. S. Fearing, “Microfabricated hinges,” Sensor and Actuators A-Physical, vol. A33, no. 3, 1992, pp. 249–256.
Jack Chen (M’02) received the B.S. degrees in mechanical engineering and electrical engineering from the University of Illinois at Urbana-Champaign, IL, in May 2000. As an undergraduate, his research focused on fabricating microdischarge devices in silicon. He received the M.S. degree in electrical engineering in August 2002. He is currently pursuing the Ph.D. degree at the Micro and Nanotechnology Laboratory at the University of Illinois. His current research focuses on low-cost polymerbased sensors.
Chang Liu (S’92–M’00–SM’01) received the M.S. and Ph.D. degrees from the California Institute of Technology (Caltech), Pasadena, in 1991 and 1996, respectively. In January 1997, he became an Assistant Professor with major appointment in the Electrical and Computer Engineering Department and minor appointment in the Mechanical and Industrial Engineering Department. In 2003, he was promoted to Associate Professor with tenure. His research interests cover micro sensors, microfluidic lab-on-a-chip systems, and applications of MEMS for nanotechnology. He has 13 years of research experience in the MEMS area and has published 100 technical papers. He teaches undergraduate and graduate courses covering the areas of MEMS, solid-state electronics, and heat transfer. Prof. Liu received the NSF CAREER award in 1998 and is currently an Associate Editor of the IEEE SENSORS JOURNAL. In 2002, he has been elected to the “Inventor Wall of Fame” by the Office of Technology Management of the University of Illinois.
979
Development and Characterization of Surface Micromachined, Out-of-Plane Hot-Wire Anemometer Jack Chen, Member, IEEE, and Chang Liu, Senior Member, IEEE
Abstract—In this paper, we report the development of a new type of hot-wire anemometer (HWA) realized by using a microfabrication process that combines surface micromachining and an efficient three-dimensional assembly technique. The HWA uses a thermal element (hot wire) that is made of Pt/Ni/Pt film with a measured temperature coefficient of resistance (TCR) of 2400 ppm/ . The thermal element is elevated out of plane by using support beams made of polyimide. In our current design, the thick and up to 1 mm tall, and the length support beam is 2.7 to 200 . Steady-state of the thermal element varies from 50 response to air velocity has been experimentally obtained up to 20 m/s under both constant current (CC) and constant temperature (CT) modes. The transient-state response has been examined using square wave and sinusoidal wave excitation signals in CT modes with the maximum cutoff frequency found to be approximately 10 kHz. This new HWA offers a number of unique materials and performance characteristics. The sensor does not require the use of silicon as either substrate or sensing materials. Using this process, it is possible to form large arrays of HWA on a variety of substrate materials. [910]
C
m
m
m
I. INTRODUCTION
H
OT-WIRE anemometry (HWA) is a well-studied technique for measuring the velocity of fluid flow. It utilizes a thermal element that serves as both a Joule heater and a temperature sensor. Under a constant bias power and zero flow rate, the temperature of the thermal element reaches a steady-state value. If external fluid flow is present around the thermal element, the thermal element will experience forced convective cooling. Accordingly, the temperature of the thermal element provides the means to gauging the cooling rate and the flow velocity. Conventional HWA are assembled individually by mounting a thin wire made of platinum or tungsten onto support prongs. The wires may be thinned (e.g., by etching in acidic solutions) until the desired dimensions are reached (typically a few mm long and a few micrometers in diameter). This active portion of the sensor is then mounted on a long probe with electrical connection for ease of handling. The schematic diagram of a typManuscript received July 26, 2002; revised August 8, 2003. This work was supported by the National Science Foundation under the CAREER program (NSF Award 9984954) and the Smart Skins program (NSF Award 0080639) and also in part by the Air Force Office of Scientific Research under the Bioinspired Concepts program. Subject Editor C.-J. Kim. J. Chen is with the Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801 USA and also with the 319 B Micro and Nanotechnology Laboratory, Urbana, IL 61801 USA (e-mail: [email protected]). C. Liu is with the Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801 USA. Digital Object Identifier 10.1109/JMEMS.2003.820261
Fig. 1. Schematic diagram of a conventional hot-wire anemometer.
ical finished device is shown in Fig. 1. The HWA is noted for its low cost, fast response (in the kilohertz range), small sizes, and low noise [1]. Conventional HWAs suffer from two major shortcomings, however. First, the fabrication and assembly process is delicate and does not guarantee uniformity of performance. Secondly, it is prohibitively difficult to form large arrays of HWA for measuring flow distribution. In the past few years, efforts have been made to apply various micromachining processes to realize HWA for air flow with smaller dimensions, better uniformity, and faster time response (due to smaller thermal masses possible). One strategy is to use the bulk-micromachining technique to produce freestanding cantilever structures. For example, researchers have reported making prongs and hot-wires out of doped polycrystalline silicon and releasing the cantilever by partially removing the silicon substrate [2]. One group used doped silicon for both the prong and the hot-wire along with polyimide hinge to make three-component hot-wire sensors [3]. Another group undercut a polysilicon hot-wire with wet etching so the hot-wire is suspended over a cavity for better thermal insulation [4]. To our knowledge, almost all existing micromachined HWAs for air flow measurement use silicon as the thermal element and involve bulk micromachining of silicon to some extent. Silicon is a relatively expensive material. The doping of silicon (e.g., for creating hot wire), the etching of silicon, as well as the packaging of individual silicon dies require significant expertise and efforts. Most of the micromachined HWAs cannot be realized in large array format efficiently. Researchers have reported using
1057-7157/03$17.00 © 2003 IEEE
980
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
metal thermal element lying on polyimide bulk film [5]; however, insufficient thermal isolation and elevation limits use. Our research is motivated by the needs to reduce the cost of HWA, to allow efficient fabrication and packaging, and to produce arrayed HWAs on potentially flexible substrates. Two strategic approaches are taken. First, we realize HWA by using surface micromachining in conjunction with three-dimensional assembly methods. This circumvents the use of bulk micromachining, which requires relatively long etching time. Bulk etching using anisotropic wet etchants frequently pose concerns of materials compatibility as all materials on a given substrate are required to sustain wet etching for long periods (several hours to etch through typical silicon wafers). Surface micromachining also enables more efficient assembly and allows formation of large arrays of HWAs. Second, we realize thermal elements using nonsilicon, temperature sensitive materials. This could reduce costs as the use of silicon bulk (as substrate) or thin film (as hot wire) is not required. By substituting silicon doping and bulk etching steps, the fabrication process can be realized in more efficient manner. In this work, we applied a new 3-D assembly process, which pairs magnetic actuation with deformable metal hinges [6], to fabricate a hot-wire anemometer using surface machining of metal and polymer materials. The fabrication process was designed carefully so that 1) the maximum temperature required and 2) no etching throughout the process flow is under 350 using concentrated hydrofluoric (HF) acid was necessary. By limiting the overall process temperature, the process can be run on a broad range of substrates, including silicon, glass, and even certain plastics. It should be noted that a prototype surface-micromachined, out-of-plane HWA had been made using hinged polysilicon structure [7], but the assembly to vertical position was done by manual probing and test results were not presented. II. ANALYSIS AND DESIGN The schematic diagram of the new out-of-plane anemometer is shown in Fig. 2. A thermal element is elevated from the substrate to a predetermined height that corresponds to the length of the support prongs. By elevating the thermal element away from the bottom of the velocity boundary layer, the thermal element experiences greater fluid flow velocity and exhibits greater sensitivity. The thermal element is electrically connected to the substrate through the support prongs as well. The new HWA will work as flow sensor whether the thermal element is beyond the boundary thickness or not. In the case the thermal elements are sufficiently tall to stand beyond the boundary layer, the measurement will directly indicate true free-stream velocity. If the thermal element is embedded in the boundary layer, it will provide direct measurement of true local flow speed, which can be used to infer free-stream velocity based on known boundary layer velocity profile, if necessary. An HWA operates by sensing temperature change of a hot-wire resulting from forced convection. The temperature variation can be inferred from the change of the resistance of the sensing element. The resistance of a hot wire is related to its temperature according to (1)
Fig. 2. Schematic diagram of a single out-of-plane HWA. The thermal element is made of metal thin film and support by two support beams, which also provide electrical leads. The thermal element is elevated away from the substrate surface (and the bottom of the velocity boundary layer).
where is the nominal resistance under a reference temperature (e.g., room temperature), and is the temperature coefficient of resistance (TCR) of the thermal element [8]. For metal, the coefficients for second order and beyond are several orders will be conof magnitude smaller than . In this paper only sidered and will be generally referred to as . We briefly discuss the theoretical basis of the HWA sensor to motivate our design. The heat balance equation for the thermal element under electrical Joule heating is [9]–[11] (2) is the rate of heat storage, is the generated (bias) where is the rate of heat loss due power from Joule heating, represents the sum of conducto forced convection, and tive losses (e.g., through support prongs). For a HWA, the term involves (1) end loss, which is unavoidable in the presence of support prongs, as well as (2) longitudinal thermal conduction along the thermal element. Under a given set of bias and fluid flow rate, it is important to maximize power while minimizing in order to obtain greater sensitivity to velocity changes. To minimize the conductive heat loss also means that the HWA sensor can be operated in a more thermally efficient manner. This is especially important if a large array of such sensors is used. Several dimensional and materials parameters are relevant relative to [12]. Assuming that the for increasing thermal element is a very long wire with a cylindrical cross section, the ratio between conductive and convective heat transfer is (3)
with variables defined in Table I. Although the formula provided above is derived based on an ideal case of infinitely long thermal elements and does not apply to thermal elements with finite length, they nonetheless provide valuable design guidelines.
CHEN AND LIU: SURFACE MICROMACHINED, OUT-OF-PLANE HOT-WIRE ANEMOMETER
981
TABLE I NOMENCLATURE
An important design parameter is the length of the hot wire . It is possible to increase relative to by invaries slowly creasing the length of the hot wire, because changes roughly linearly with , whereas the value of with respect to . However, is limited by fabrication practicality and yield considerations. The longer the thermal elements, the more difficult it is to realize and withstand high flow. Another important design parameter is . In the common case that the thermal element has a rectangular cross-section, the term is the equivalent diameter that yields the cross-section area. By reducing the value of , the surface-to-volume ratio of the hot wire is increased, hence encouraging more convection while confining the conductive component. However, there exists a practical limit to the minimal diameter of the hot wire as well, due to mechanical rigidity concerns. Our current study is aimed at establishing the fundamental design, process feasibility, and reliability. It was not our intention to conduct exhaustive optimization of the dimensional and materials parameters at this stage. Nevertheless, a wide range of dimensional parameters has been accommodated in our proto, 100 , type design. Devices with hot-wire lengths of 50 , and 200 , as well as heights up to 1000 can be 150 fabricated. The cross-section area of the hot wire is a composite of a high TCR metal on an insulating polyimide beam. The cross wide by 0.12 thick, while section of the metal film is 4 wide by 2.7 thick. The polythe polyimide beam is 6 imide piece provides the mechanical support. The cross section of this composite thermal element is comparable to that of commercially available hot wire sensors. III. FABRICATION PROCESS The first step of developing the fabrication process is to select suitable materials from the perspectives of both performance requirements and process simplicity and compatibility. We must first choose materials for the hot wire that provides a high TCR.
TABLE II TABLE OF MATERIAL PROPERTY. ALL VALUES ARE CITED FROM [22] EXCEPT FOR THE THERMAL CONDUCTIVITY OF POLYIMIDE, WHICH IS CITED FROM [23]
We first tried to use Pt as the hot-wire filament because of the high TCR value of 3900 ppm/ . However, the measured electrical resistivity of the evaporated 1000- -thick Pt was several times the theoretical value (see Table II) possibly due to electron scattering at grain boundary [13]. This results in a reduced effective TCR of 780 ppm/ . This could be improved by annealing to increase grain size. However, the annealing , which is incompatible temperature of Pt is above 600 with integrated circuits and the polyimide film [14]. We have therefore chosen Ni as the primary thermal element material, because the experimentally measured TCR of as deposited Ni ) is greater than what thin films (approximately 3750 ppm/ we measured for Pt. This TCR value is also higher than using a polysilicon one, which has reported value of only 1460 ppm/ [15] or 800 ppm/ [3]. The hot wire consists of temperature-sensitive metal thin film on a supporting polyimide piece. The polyimide support is used because it provides the hot wire with needed structural rigidity without increasing thermal conductance of the thermal element. The thermal conductivity of polyimide is low, almost
982
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
three orders of magnitude lower than that of a metal, e.g., nickel (see Table II). In this work, the thickness of the polyimide due to processing concerns. If the thin film is roughly 2.7 thickness is much lower than this value, the mechanical rigidity will likely be degraded. On the other hand, if the thickness is much greater, there is concern that the polyimide support will decrease the frequency response of the HWA due to added thermal mass. The fabrication process utilizes an efficient 3-D assembly method called the Plastic Deformation Magnetic Assembly (PDMA), which was discussed in detail in [6]. A brief discussion of this method is provided in the following. The PDMA process utilizes surface micromachined structures that are anchored to substrates with cantilever beams made of ductile metal materials (e.g., gold and aluminum). The microstructure is attached to pieces of electroplated ferromagnetic material (e.g., Permalloy). By applying an external magnetic field, the ferromagnetic material is magnetized and interacts with the field to bend the microstructure out of plane. If the amount of bending is significant, the cantilever support hinges will be plastically deformed, resulting in permanently bent microstructures even after the magnetic field is removed. The process is very efficient and can be realized in parallel on the wafer scale. The overall fabrication process is shown in Fig. 3. A precursor of the process has been discussed previously [16]. Numerous improvements have been implemented since then. For example, was measurement of flow responses at high velocity not attempted using the first-generation devices due to concerns toward the fragility of support hinges. Later, we improved upon the process so that the sensors are based on strengthened mechanical supports and can sustain higher wind velocity. The starting wafer is silicon. However, the process can be performed on glass or polymer substrates, as the overall temperature of the process is intentionally kept low. First, a chrome/copper/titanium metal stack is evaporated and patterned as the sacrificial layer [see Fig. 3(a)]. The 10-nm-thick chrome film serves as the adhesion layer. The 250- -thick titanium thin film reduces the in-process oxidation of the 2500- -thick copper film [17] during the subsequent polyimide curing. A 2.7- -thick photo-definable polyimide (HD-4000) is [ see spun-on, patterned via lithography, and cured at 350 Fig. 3(b)] for 2 h. This polyimide layer forms the support prong and part of the hot wire. A Cr/Pt/Ni/Pt film is then evaporated and patterned to form the thermal element [see Fig. 3(c)]. The thickness of the Cr layer, which serves as an adhesion layer, is 200 . A 800-thick Ni resistor is sandwiched between two 200- -thick Pt films, which are used to reduce possible oxidation of Ni while in operation because Pt is relatively inert at high temperature of operations. Optionally, the Pt can be patterned separately to complete cover the Ni, including sidewalls, to prevent long-term degradation due to oxidation We then evaporate and pattern a 5000- -thick Cr/Au film [see Fig. 3(d)] to serve as a mechanical bending element as well as electrical leads of the hot-wire filament. We electroplate a 4- -thick Permalloy thin film on portions of the cantilever support prongs [see Fig. 3(e)].
Fig. 3. The fabrication process of a surface micromachined hot-wire anemometer: (a) deposition of sacrificial layer, (b) pattern polyimide structure, (c) deposit nickel sensor, (d) deposit gold, (e) electroplate permalloy, (f) strip electroplating mold, (g) pattern photoresist (PR) layer for postrelease plating, (h) sacrificial layer etching, and (i) PDMA assembly.
Sacrificial layer release is performed by using a solution containing acetic acid and hydrogen peroxide to selectively remove the copper thin film. PDMA assembly is carried out to lift the entire sensor out of plane [see Fig. 3(i)] by placing a permanent magnet (field strength 800 Gauss) at the bottom of the substrate. To finish the process, the device chip is then rinsed in deionized water and dried.
CHEN AND LIU: SURFACE MICROMACHINED, OUT-OF-PLANE HOT-WIRE ANEMOMETER
983
to the substrate while Ni is locally electroplated (see Fig. 5). The plating time is approximately 5 min at 100 per thick Ni at the device, which gives us approximately a 3–4 hinge. At present, this plating process is not optimized and is performed one device at a time. However, with proper wiring layout and plating control, we expect this process to be scalable to the wafer level. An SEM micrograph of a representative device after release and assembly is shown in Fig. 6. Fig. 7 shows the SEM of the Au hinge with and without the post-release plating. The improved device after plating was tested at wind speed up to 25 m/s without failure. Higher speed has not been used because of limitation in test instruments. IV. STEADY-STATE OUTPUT RESPONSES Fig. 4. Au-polyimide separation, shows Ti as a poor adhesion layer.
The adhesion between the Au to the polyimide is very important. Without adequate adhesion, the polyimide and the gold film would separate during the PDMA assembly. One of the ways to improve adhesion was to use an adhesion layer reactive ion etching and treat the polyimide layer by using (RIE) before the metal deposition. Cr seems to be the adhesion layer of choice, and the RIE treatment creates a hydrophilic structure on the polyimide surface that enhances adhesion [18]. Ti was initially used as the adhesion material for both the Au and the hot-wire element due to its good stability in chemical etchants and higher electrical resistivity. However, we found that the metal layer peels off from the polyimide during PDMA assembly process. A scanning electron microscopy (SEM) image of this effect is shown in Fig. 4. The curling of the polyimide is due to the intrinsic stress created during curing. The straight Au frame is maintained because of the low-stress nature of electroplated Permalloy. No such separation was observed after Cr was used as the adhesion layer with the RIE treatment. Both the thermal element and the support prongs have relatively small frontal area. The momentum imparted by fluid on the sensor is minimal. We have demonstrated that the sensor after the PDMA assembly can withstand airflow with mean stream velocity lower than 5 m/s without being damaged. However, at such velocity the support prongs visibly vibrate. Other than low-speed applications, it is imperative that the ductile bent hinge is reinforced to withstand the force due to the airflow. Toward this end, we added an additional post-release Ni electroplating step. Ni is selectively electroplated at the bent hinges to strengthen the hinges. Before the removal of the Cu sacrificial layer, photoresist is spun on and patterned to prevent electroplated Ni from plating on the sensing hot wire [see Fig. 3(g)] The only metal exposed is the Au hinges and the electroplating contacts. The sample is then put into the Cu etchant until the Cu sacrificial layer is completely undercut. Afterwards, the sample is removed from the etchant solution, thoroughly rinsed in deionized water, and immediately placed into a Ni plating bath. External magnetic field is applied normal
We tested the steady-state performance of the sensor in a wind tunnel. The fabricated hot-wire anemometer chip is attached to a PC board with etched wiring traces. The PC board is in turn placed on a stage located in the middle of a wind-tunnel test section (see Fig. 8). The chip is located near the edge of the packaging, with the hot-wire element located only 1 mm away from the leading edge. We have characterized the response of HWAs in both constant current and constant temperature modes. Because the HWA are located close to the leading edge, the even boundary layer at that distance is very small for very low Reynolds number flow. The steady-state flow measurement tests were conducted under laminar flow with Reynolds number between 100 to 2000. For higher flow rate during wind tunnel measurement, the thermal element is above the boundary layer. The TCR value of the thermal element (Cr/Pt/Ni/Pt film and 100 . stack) is found to be 2400 ppm/ between 25 It should be noted that this TCR value is lower than cited TCR values of all compositional metal films (list in Table II). This discrepancy could be contributed by many factors. However, a detailed discussion and resolution of this issue is outside the scope of this paper. Steady-state output of an infinitely long, cylindrical hot wire ) should follow King’s law. For an HWA that has (large ratio, we follow the noncylindrical geometry and smaller common practice of collapsing the experimental hot-wire data in the form (4) and are the temperature in the sensor and fluid, where is the resistance of the hot-wire sensor. The respectively, and term is the overall heat transfer coefficient (5) where is a fitting factor that accounts for the smaller aspect ratio and wire geometry [9], [12], [19]. The heat transfer coefficient accounts for the conduction end loss as well as natural is the forced convection term. convection, and
984
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
Fig. 5. Electroplating in a bath under microscope to reinforce hinge: (a) Device placed in bath and external magnetic field applied to bring the structure to 90 , (b) power supplied to the hinge and plating started at 2 mA for 5 min, and (c) once plating is completed, sample is removed from the bath and resist is stripped.
(a) (a)
(b) Fig. 6. SEM micrographs of (a) an HWA immediately after sacrificial release and PDMA assembly, and (b) an HWA after postrelease strengthening using electroplating.
(b) Fig. 7. SEM graphs of the plastically deformed hinge (a) without postrelease Ni plating and (b) with postrelease plating.
A. Constant Current Response In the constant current (CC) mode, the response can be derived from (5) by holding the current constant. A calibration equation for constant current mode can be written as (6a) where (6b) (6c)
The terms and are velocity independent constants, and is the change in voltage drop across the hot-wire. The is defined as the resistance overheat ratio term at velocity . A CC driving circuitry is shown in Fig. 9(a), using an LM334 current source in conjunction with a diode to reduce temperature dependence of the circuit. Due to a positive TCR, the voltage across the hot wire decreases as the flow rate increases. The sensitivity of output voltage with respect to air velocity increases with increasing overheat ratio because of the increase in temperature difference between hot wire and the fluid.
CHEN AND LIU: SURFACE MICROMACHINED, OUT-OF-PLANE HOT-WIRE ANEMOMETER
985
Fig. 8. Photo of the wind tunnel measurement setup. An anemometer chip is placed on a glass holder that is positioned on a stage located in the test section. The finished package is mounted onto the test fixture. The yaw and pitch angle of the chip can be controlled from outside of the wind tunnel.
Fig. 10. The response of a 400-m-high and 200-m-long, hot-wire anemometer at various overheat ratio (defined at U = 0 m=s) operating in the constant current mode. Solid lines are the curve fit using (6a)–(6c). The air velocity U is calibrated using the Traceable hot-wire anemometer.
(a)
(b)
Fig. 9. (a) Constant current circuit, LM334 is an IC current mirror. (b) Constant temperature circuit. Bridge gain G R =R + 1, R R ,V and R not used during steady-state operation.
=
The output response of a representative hot wire that is long and 400 height is shown in Fig. 10. The three 150 . The current-overheat curves correspond to three different correspondence is determined by voltage-current division at every data point with an Agilent 4155 parameter analyzer, is obtained at 5 mV. Curve fitting with (6a) verifies where good correspondence between the measured data and the theoretical relation at velocity above 3 m/s, with equals 1. At low velocity, because a larger portion of the heat transfer is due to natural convection, the deviation between the data and curve fit is greater. As expected from (6a)–(6c), the voltage drop is larger at a high overheat, but flattens out in all three overheats at high velocity. Velocity response of HWA at various lengths operating at is show in Fig. 11, curve fit with . A comparratio is more revealing. From (5), is not a ison of the function of flow velocity, but of the wire geometry. The higher the B/A ratio, the greater the heat loss attributed to forced con, and it has vection. The longest micromachined HWA is 200
Fig. 11. The response of 400-m-tall hot-wires of various lengths operating in the constant current mode. The devices operate at resistance overheat a = 0:3. The heat transfer ratio B=A defined in (5) is determined from B=A = D =(1 + a ).
a ratio of 0.11, compared to the shortest HWA of 100 , which has a B/A ratio of 0.063. This comparison is in agreement ratio will increase with length with (3), where because there is more loss from forced convection. Commercial HWA has a B/A of around 0.15, which is higher than the micromachined ones made here [19]. This is because most com, while maintaining mercial HWA has a length of over 1000 a wire diameter of a few microns. However, commercial HWA has a geometric exponential factor of 0.5, which means more degradation at higher velocity, compare to the micromachined HWA presented here. For all HWA of various lengths tested, the ratio of the same hot-wire operating at various does not change significantly. B. Constant Temperature Response The steady-state equation for constant temperature (CT) mode operation can be solved in a similar way as the CC mode. is placed on one leg of a Whetstone The hot-wire sensor is kept constant by the feedback bridge. In CT mode, op-amp circuit shown in Fig. 9(b).
986
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
m
m
Fig. 12. Response of a 400- -high, 200- -long, hot-wire anemometer operating in the constant temperature mode. The bridge ratio R =R of the CT circuit is 10, and overheat ratio a is set at 0.1, 0.2, and 0.3.
The bridge circuit is balanced when . The reis a variable resister used to set the hot-wire temperasistor ture (7) is the resistance overheat The term at all air velocity , and is the resistance of the hot-wire at room temperature. Rewriting (4), it can be shown that the voltage output will vary with flow speed according to the relation below:
m
Fig. 13. The response of 400- -tall hot-wires of various lengths operating in the constant temperature mode. The devices operates at resistance overheat a = 0:3. The heat transfer ratio B=A = D =C .
of the hot wire and the conductive thermal resistance. Commercial HWA typically have bandwidth of a few hundred hertz in CC mode, and can approach tens of kilohertz in CT mode. Frequency response for CT operation also has two methods of supply. electrical testing, sine wave or square wave using the The schematic diagram of the constant temperature circuit is shown in Fig. 9(b). Assuming the HWA has only one time constant and the op-amp has a single pole at , the output voltage is derived [2], [20] as a second-order transfer function
(8)
(10a)
where with (9) is the voltage at The terms and are constants from (5), , and is the bridge gain defined as . At low fluid velocity, the calibration curve does not correspond well with the data because of free convection. The CT output response of the same HWA in the CC test (Fig. 10) with curve fitting from (8) is shown in Fig. 12. The fitted lines are the theoretical curves, and they agree well with the measured data except at very low velocity. Curve fitting gives a value in the CT mode, compare to reported of by Jiang et al. [2]. The velocity response of HWA with length to 200is plotted in Fig. 13 operating varying from 100. To compare the three HWAs, the at an overheat ratio between the heat transfer coefficient of conductive loss and force convection is used. The B/A ratio increases from 0.063 for the 100- -long HWA to 0.13 for the 200- -long HWA. V. TRANSIENT RESPONSES When there is a sudden change in the fluid velocity, the output signal will lag behind. It is important to understand the valid frequency range of the HWA, especially for high Re flow measurement. The time constant is determined by the thermal mass
(10b) (10c) is the dc voltage gain, is the velocity sensitivity, The term is the quality factor, is the time constant of the hot-wire, is and are the op-amp gain and pole, and overheat ratio, and are the small signal input voltage and velocity. The small and velocity disturbance are signal voltage disturbance decoupled in the numerator of (10a). Therefore, by analyzing in the frequency domain, we can the HWA response due to obtain the HWA frequency limit with velocity fluctuation. This is beneficial because voltage disturbance is simpler to generate than velocity disturbance. The responses of the CT circuit with a 200- -long HWA to a small signal voltage at constant velocity are shown in Fig. 14. is flat up to , then increases at 20 The value of at the peak, dB/decade until the it reaches a approximately beyond which it asymptotes at 20 dB/decade. measured at the peak corresponds to the The second pole corner frequency of a voltage output with a velocity fluctuation
CHEN AND LIU: SURFACE MICROMACHINED, OUT-OF-PLANE HOT-WIRE ANEMOMETER
987
(a)
Fig. 15. Yaw angular response of a 150-m-long, 600-m-high HWA, the velocity reading is based on wind tunnel calibration using (8). The measured data points are placed along side solid line fitted using the cosine relation (10).
(b)
(c)
m
Fig. 14. Frequency response of a 200- -long hot-wire in CT mode with a sine wave disturbance under the following conditions: (a) R =R = 10, = 0:3, U = 2 m=s; (b) R =R = 10, = 0:3, U = 20 m=s; (c) R =R = 1, a = 0:3, U = 20 m=s. Time constant measured with square wave disturbance is embedded in each figure and corresponds to f = 1=(1:3 1 ).
, making the sine wave response a useful measurement for the cutoff frequency (for velocity fluctuation). Beyond , the output due to will drop off at 40 dB/decade. In order to ensure a flat response, the HWA in CT mode should not operate in turbulence near . Using the frequency response plotted in Fig. 14(a) and (b), we can see that increases slightly from 4 dekHz to 5 kHz as the velocity is increased. As the creases [see Fig. 14(c)], increases to 9 kHz. However, this is at the expense of the velocity sensitivity. Another way to measure the cutoff frequency is by examwith a square wave . The square wave test assumes ining varying the heating current can represent the heating and cooling of the HWA with velocity fluctuation. For example, at the onset of a voltage increase, the feedback circuit will try to balance the bridge by decreasing the voltage output (and hence the heating current), this is similar to a sudden decrease in fluid velocity. to settle back to 3% of peak value is The time it takes for defined as , and can be related to by [21] (11) The corresponding square wave response for each sine wave test clearly marked. The is imbedded in the bode plot with the responses from square wave testing verify the cutoff frequency found in all three cases with the sine wave test. This cutoff frequency of new HWA sensors is comparable to that of a commercial anemometer.
Fig. 16. Micrograph of an array of two-component HWA sensor fabricated on a sheet of Kapton thin film.
VI. ANGULAR RESPONSE Provided a sufficiently large aspect ratio for the HWA filament, the yaw response should follow the cosine relation (12) The yaw response is tested using a rotating stage setup shown in Fig. 8. Constant temperature mode is used because of the higher flow sensitivity associated with this mode. In the measurement, and 50 at 5 HWA output is measured at angle between intervals. This voltage output is then converted to velocity using (8) and then plotted in Fig. 15, along with a fitted line based on the cosine relation. The yaw response of the 150- -long HWA seems to fit the cosine relation fairly well. Because the thermal element is made of a composite film, the response to reverse flow is not expected to be symmetrical. This problem can be circumvented by device calibration, by using an array of sensors, or by sandwiching the metal in between two polyimide layers. VII. COST ESTIMATE The main difference between the bulk-micromachined HWA and the surface micromachined HWA approach is the expected cost. The new HWA promises to reduce the costs for realizing arrayed flow sensors. Conventional Silicon micromachined sensors are associated with greater substrate costs and longer time for processing and assembly. Some commercial hot-wire arrays are made by depositing a high TCR material on a thermal insulating Kapton film. As a comparison, we show in Fig. 16 a two-dimensional array of
988
JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 12, NO. 6, DECEMBER 2003
HWA fabricated on a flexible Kapton film. The ability to fabricate large arrays on flexible Kapton film, coupled with the low temperature nature of the process, means this type of HWA can potentially be fabricated using roll-to-roll, high throughput processing. VIII. CONCLUSION We have designed and developed an efficient microfabrication process for realizing a novel HWA. The HWA can be fabricated using nonsilicon materials and hence can be potentially produced with low costs and high fabrication efficiency. The HWA can potentially be realized in large array format for distributed flow sensing. If necessary, the sensor can be made on flexible substrates for conformal coating of fluid dynamic surfaces of interests. The sensor has been systematically characterized under both constant current and constant temperature modes. The time constant of a sensor is 10 kHz under constant temperature mode. The main objective of this work is to establish the design and fabrication process, and to characterize the performance to establish the proof-of-concept and performance merits. We have developed a new HWA sensor that offers unique functional advantages over existing hot wire sensors. Comprehensive optimization of design, materials, and associated fabrication methods are quite involved and are not exhaustively addressed in this work. Such issues will be systematically explored to incrementally improve the performance, ease of fabrication, and reliability.
[8] J. O. Hinze, Turbulence: And Introduction to Its Mechanism and Theory. New York: McGraw Hill, 1959, p. 78. [9] C. G. Lomas, Fundamentals of Hot-Wire Anemometry. Cambridge, U.K.: Cambridge University Press, 1986. [10] A. E. Perry, Hot-Wire Anemometry. Oxford, U.K.: Clarendon, 1982. [11] R. J. Emrich, Ed., Methods of Experimental Physics: Fluid Dynamics Part A. New York: Academic Press, 1981, pp. 259–314. [12] H. H. Brunn, Hot-Wire Anemometry: Principles and Signal Analysis. Oxford, U.K.: Oxford University Press, 1995. [13] A. F. Mayadas, M. Shatzkes, and J. F. Janak, “Electrical resistivity model for polycrystalline films: The case of specular reflection at external surfaces,” Applied Physics Letters, vol. 14, p. 343, 1969. [14] S. Sedky et al., “Experimental determination of the maximum postprocess annealing temperature for standard CMOS wafers,” IEEE Trans. Electron Devices, vol. 48, pp. 377–385, Feb. 2001. [15] F. Jiang, Y. C. Tai, C. M. Ho, and W. J. Li, “A micromachined polysilicon hot-wire anemometer,” in Technical Digest, Solid-State Sensor and Actuator Worlshop, Hilton Head Island, SC, 1994, pp. 264–267. [16] J. Chen, J. Zou, and C. Liu, “A surface micromachined, out-of-plane anemometer,” in Proceedings MEMS, Las Vegas, 2002, pp. 332–335. [17] S. A. Chambers and K. K. Chakravorty, “Oxidation at the polyimide/CU interface,” J. Vacuum Sci. Technol. A, vol. 6, no. 5, pp. 3008–3011, Sept. 1988. [18] Y. Nakamura, Y. Suzuki, and Y. Watanabe, “Effect of oxygen plasma etching on adhesion between polyimide films and metal,” Thin Solid Films, vol. 290, pp. 367–369, 1996. [19] F. Jiang, “Silicon Micromachined Flow Sensors,” Ph.D. dissertation, California Inst. Technol., Pasadena, CA, 1998. [20] P. D. Weidman and F. K. Browand, “Analysis of a simple circuit for constant temperature anemometry,” J. Phys. E—Sci. Instrum., vol. 8, no. 2, pp. 553–560, July 1975. [21] P. Freymuth, “Frequency response and electronic testing for constant temperature hot-wire anemometers,” J. Phys. E—Sci. Instrum., vol. 10, pp. 705–709, 1977. [22] G. T. A. Kovacs, Micromachined Transducer Sourcebook. New York: McGraw-Hill, 1998. [23] Pyralin Polyimide Coating for Electronics: PI2610 Series—Product Information and Process Guidelines, HD Microsystems, 1998.
ACKNOWLEDGMENT The authors wish to acknowledge the technical support of J. Hughes, H. Romans, and R. Blaney for maintaining processing equipment at the Micro and Nanotechnology Laboratory of the University of Illinois, and the technical support of A. Broeren at the Department of Aeronautical and Astronautical Engineering at University of Illinois. REFERENCES [1] R. J. Goldstein, Ed., Fluid Mechanics Measurements. Washington, DC: Hemisphere, 1983, pp. 99–145. [2] F. Jiang, Y. C. Tai, C. M. Ho, R. Karan, and M. Garstenauer, “Theoretical and experimental studies of micromachined hot-wire anemometers,” in Technical Digest, International Electron Devices Meeting(IEDM), San Francisco, CA, 1994, pp. 139–142. [3] T. Ebefors, E. Kalvesten, and G. Stemme, “Three dimensional silicon triple hot-wire anemometer based on polyimide joints,” in Proc. MEMS, New York, 1998, pp. 93–98. [4] T. Neda, K. Nakamura, and T. Takumi, “A polysilicon flow sensor for gas flow meters,” Sens. Actuators A, Phys., vol. A54, no. 1–3, pp. 621–631, June 1996. [5] D. Dittmann, R. Ahrens, Z. Rummler, K. Schlote-Holubek, and W. K. Schomberg, “Low-cost flow transducer fabricated with the AMANDAprocess,” in Proc. Transducer, Munich, Germany, 2001, pp. 1472–1475. [6] J. Zou, J. Chen, C. Liu, and J. E. Schutt-Aine, “Plastic deformation magnetic assembly (PDMA) of out-of-plane microstructures: Technology and application,” J. Microelectromech. Syst., vol. 10, no. 2, pp. 302–309, June 2001. [7] K. S. J. Pister, M. W. Judy, S. R. Burgett, and R. S. Fearing, “Microfabricated hinges,” Sensor and Actuators A-Physical, vol. A33, no. 3, 1992, pp. 249–256.
Jack Chen (M’02) received the B.S. degrees in mechanical engineering and electrical engineering from the University of Illinois at Urbana-Champaign, IL, in May 2000. As an undergraduate, his research focused on fabricating microdischarge devices in silicon. He received the M.S. degree in electrical engineering in August 2002. He is currently pursuing the Ph.D. degree at the Micro and Nanotechnology Laboratory at the University of Illinois. His current research focuses on low-cost polymerbased sensors.
Chang Liu (S’92–M’00–SM’01) received the M.S. and Ph.D. degrees from the California Institute of Technology (Caltech), Pasadena, in 1991 and 1996, respectively. In January 1997, he became an Assistant Professor with major appointment in the Electrical and Computer Engineering Department and minor appointment in the Mechanical and Industrial Engineering Department. In 2003, he was promoted to Associate Professor with tenure. His research interests cover micro sensors, microfluidic lab-on-a-chip systems, and applications of MEMS for nanotechnology. He has 13 years of research experience in the MEMS area and has published 100 technical papers. He teaches undergraduate and graduate courses covering the areas of MEMS, solid-state electronics, and heat transfer. Prof. Liu received the NSF CAREER award in 1998 and is currently an Associate Editor of the IEEE SENSORS JOURNAL. In 2002, he has been elected to the “Inventor Wall of Fame” by the Office of Technology Management of the University of Illinois.
