High-Sensitivity Bi-Directional Flow Sensor Based on ... - Chang Liu
High-Sensitivity Bi-Directional Flow Sensor Based on Biological Inspiration of Animal Haircell Sensors Craig Tucker, Nannan Chen, Jonathan Engel, Yingchen Yang, Saunvit Pandya and Chang Liu Micro and Nanotechnology Laboratory University of Illinois Urbana-Champaign Urbana, IL, USA [email protected] Abstract—Artificial Haircell (AHC) sensor is presented for highly sensitive flow-field measurements. Design considerations and MEMS process flows are given. Oscillating flow field measurements show sensitivity down to 0.6mm/s flow rates, steady state flow fields detected down to 0.1mm/s.
I.
INTRODUCTION
Flow imaging systems based around hair cell sensors are found in nature to be very sensitive compared to traditional solutions in engineering. Spiders [1], crickets, fish [2], and bats all have haircells which can detect sub-nanometer deflections, allowing them to locate prey, avoid predators, and characterize flow conditions around the organism in real time. Such systems would be useful in underwater autonomous vehicles (UAVs), self-stabilizing micro air vehicles (MAVs), as well as in traditional flow-mapping situations. The first step to mimicking a biological flow tracking system is demonstration of highly sensitive flow detection using micro-machined sensors. Here we present the continuing work in characterizing and optimizing an AHC sensor. II.
Figure 1. SEM micrograph of high sensitivity artificial haircell sensor
DEVICE DESIGN
Drawing from biological inspiration, the device consists of a cilium-like polymer hair attached to a piezoresistive Silicon cantilever [3]. The cilium transfers its moment load to the cantilever, exerting a strain at the base. Analytical models of the beam-cilium system with consideration given to fabrication constraints leads to a beam thickness of 2µm, width of 40µm, and length of 200µm, with hair diameter of 80µm and height of 600µm leading to a sensitivity of about 2.5 micro-strain, or under 1mm/s flow rate sensitivity. The semi-empirical model takes into consideration the noise level of our detection circuitry as well as parasitic resistances. The chosen beam dimensions also lead to a resonant frequency on the order of several KHz, far out of the regime of intended use. Ion-implanted strain gauges using boron dopant in light n-type substrate were designed
This research funded in part by DARPA and AFOSR grants.
Figure 2. Schematic drawing of AHC sensor. Cutaway shows cavity beneath cantilever
to be 1/3 of the beam thickness in depth with a doping concentration of 1x1020cm-3 to optimize gauge factor. Wheatstone bridge resistors were also fabricated on die to eliminate thermal noise and signal drift.
AHC S ensor Underw ater AC Flow V elocity S w eep
Sensor Output (V)
0.1
0 .01
0.0 01
0.00 01 0 .0 00 1
0 .0 01
0 .0 1
Flo w Velocity (m /s)
Figure 4. Response of artificial haircell sensor to oscillating flow stimulus. Figure 3. Fabrication process for artificial haircell sensor. (a) SOI wafer doped using ion implantation; (b) oxide insulator grown and patterned; (c) metal contacts and wiring deposited and patterned; (d) cantilever defined by DRIE etch; (e) cavities created through backside DRIE; (f) SU-8 hair created; (g) buried oxide etched to release sensor
DEVICE FABRICATION
Fabrication starts with a Silicon-on-Insulator (SOI) wafer, with 2µm device layer, 2µm buried oxide, and 300350µm handle. Ion implantation with Boron is performed to create bridge resistors and the paddle piezoresistor (Fig. 3a). Implantation energy is chosen to maximize gauge factor in p-type piezoresistor. Thermal oxide is grown to serve as electrical isolation for the metal leads and also to drive in the implanted ions to the proper depth (Fig 3b). Contact windows are opened in the oxide with a wet BHF etch to allow electrical connection to the doped regions. 500 Å titanium and 5000 Å gold are e-beam evaporated and patterned by liftoff to form the wiring (Fig 3c). A DRIE etch defines the cantilever and isolation trenches between bridge resistors (Fig 3d). This is a Bosch-process etch designed to have good selectivity against oxide, which must remain intact for later steps. Polyimide (HD4000) is photodefined to cover everything but the wirebonding areas, cantilevers and previously defined paddle etch region. Curing temperature is chosen to minimize diffusion of gold in the doped contact regions. The handle silicon is then etched from the backside by Bosch-process DRIE (Fig 3e). AZ4620 resist is used as an etch mask. After exposure using a Quintel IR backside aligner, the resist is hard-baked to improve its resistance to the etching species. SU-8 is spun and patterned to create the high-aspect ratio cilia (Fig 3f). This process uses only one coat of SU-8 2075, exposed with a 320nm optical filter to prevent t-topping effects. Finally, a wet oxide etch releases the sensor (Fig 3g). Due to the presence of residues from PRs and Bosch-process deposited carbon on the frontside oxide and the hydrophobicity of the backside silicon cavities, often a high-power oxygen plasma etch is performed on the back side of the sample to assist in the HF etch. Vapor-phase etching was also investigated to help with titanium undercutting, but this only added to the problem.
16.0 14.0
RMS Sensor Output (mV)
III.
AHC Sensor Underwater DC Flow Velocity Sweep
12.0 10.0 8.0 6.0 4.0 2.0 0.0 0.01 -2.0
0.1
1
10
100
Flow Velocity (mm/s)
Figure 5. Response of artificial haircell sensor to steady flow
IV.
EXPERIMENT AND RESULTS
Fabricated sensors were conformally coated with 2µm of Parylene after wirebonding and cable soldering to provide extra environmental protection. The sensors were then tested for sensitivity under steady state (DC) and oscillating (AC) flow. A. AC Flow Characterization To determine the sensitivity under oscillation flow, a packaged sensor was rigidly affixed in a water tank containing a vibrating dipole sphere source. The dipole was a small plastic sphere 6.35mm in diameter attached to a thin but rigid metal rod, actuated by a commercially available one-dimensional membrane based shaker to vibrate at 50 Hz. A commercial accelerometer was attached to the shaker. Given the dipole flow field equation [4]
v flow (r ,θ ) = (a 3U 0
a 3U 0 sin θ cos θ ˆ e + ) ( )eˆθ (1) r 2 r3 r3
Where a is the sphere radius, U0 is the RMS dipole velocity, and the coordinate pair (r,θ) define the position of the sensor with respect to the dipole. Placing the sensor such that the
theta component drops out, we wrote a LabView function to calculate the RMS flow at the tip of our cilia as a function of our accelerometer signal. After setting the distance between the sensor and dipole, the LabView program sweeps through a predefined list of shaker control voltages. A National Instruments DAQ system interfaced with our sensor amplifiers reads in the sensor voltage timetraces. Taking the fast fourier transform (FFT) and choosing the peak corresponding to the dipole frequency allows us to generate a graph showing the sensor signal and its relation to the noise floor, or the sensor output under zero flow condition. The experimental results show the sensor acting in a clear linear regime down to about 0.6mm/s flow rates, where the signal becomes on the order of the electronic and intrinsic sensor noise. Differential sensitivity in the linear region, calculated using the worst-case standard deviation in measurement comes to 0.2mm/s. B. DC Flow Characterization Since flow sensors are often used to determine steady state flow, we calibrated the AHCs under DC flow conditions. In this setup, a packaged sensor was mounted rigidly to a stepper motor-based computer-controlled linear stage system. The stages were controlled by a LabViewbased program to traverse a fluid tank containing still water. Sensor was adjusted to have a 5˚ angle of attack to the direction of traversal in order to help shield from flow separation and vortex shedding effects on the leading edge of the sensor die. Mean voltage output measurements were taken using a National Instruments 12-bit DAQ system during sensor traversal and compared to a baseline measurement taken under zero motion to eliminate DC offsets in bridge tuning. When measuring stead flow, DC level drifting presents problems not seen in oscillating flow. First, the Wheatstone bridge is tuned using a DC voltage offset controller by a PCB-mountable multi-turn potentiometer. These potentiometers are prone to electrical noise introduced by the mechanical wipers inside, and their resistance can shift over time due to thermal effects. Second, electrical power dissipated in the sensor brings about drifts caused by convection and fluid cooling. In our case an excitation voltage of 1.2V is used with a typical sensor resistance of 600-800Ω, giving a dissipated power of about 2mW. This heating and subsequent cooling causes the resistances to be transient, adding both low and high-speed drifts in the signal. This can be combated by lowering the excitation voltage and thus dropping the power quadratically, but one has to keep in mind the fact of the unamplified signal being proportional to this voltage. Finally, given the fluid media itself is conductive, issues arise due to charging and current leakage when the polyimide and especially Parylene layers are damaged or incomplete. Correcting for the slower DC signal drifts in biology is done through neural adaptation, where a persistent signal is eventually tuned out by inhibitory nerve processes. For
calibration experiments, this was simply achieved by taking a baseline mean value moments before starting stage traversal. This is then subtracted from the mean value found while the sensor is experiencing flow, thus zeroing the signal perfectly around the condition of still water. The experiment shows flow detection down to 0.1mm/s. Beyond this range, thermal and electronic drifts dominate the DC signal. V.
DISCUSSION AND CONCLUSION
The sensor presented in this paper is very sensitive compared to the best commercial flow sensors, without some of the problems normally associated with them. Hotwire or hot-film anemometers obtained commercially were tested down to around 1.0mm/s in our lab under similar conditions to our flow sensors, however they suffer from AC signal rectification problems, poor directionality rejection, thermal convection and bubble formation issues. By utilizing a system which is in near thermal equilibrium with the fluid mass, many of the hot-wire’s pitfalls are avoided. Though we report high levels of sensitivity, these devices are still a couple of orders of magnitude off from estimates of threshold sensitivity in the biological systems this sensor takes inspiration from. However, sensitivity of the AHC can be improved further. Improving electrical noise from the simple op-amp circuit could potentially drive down the noise floor. Optimization of the metallization process would yield higher gauge factors and less parasitics in the resistance measurement. Finally, modified cilia geometries to better mimic complex arthropod hair structures would yield better fluid coupling. This is all in addition to previously presented ideas of hydrogel cupulas and fluid canal structures. The work presented here demonstrates a highly sensitive artificial hair cell with the ability to detect both steady state and oscillating flows in the sub-mm/s range. The AHC is judged to be a suitable replacement for hot-wire anemometers, both micromachined and conventional, as well as being a useful tool for the further study of biological sense organs. REFERENCES [1]
[2] [3]
[4]
J. T. Albert, O. C. Friedrich, H. E. Denchant, F. G. Barth, “Arthropod touch reception: spider hair sensilla as rapid touch detectors,” Journal of Comparative Physiology A, vol. 187, pp. 303-312, 2001. S. Dijkgraaf, “The functioning and significance of the lateral-line organs,” Biology Review, vol. 38, pp 51-105, 1962. N. Chen, J. Chen, J. Engel, C. Tucker, C. Liu, “Development and characterization of high-sensitivity bioinspired artificial haircell sensor”, Hilton Head 2006, Hilton Head, South Carolina, June 4-8, 2006. J. Chen, J. Engel, N. Chen, S. Pandya, S. Coombs, C. Liu, “Artificial lateral line and hydrodynamic object tracking,” MEMS 2006 Conference, Istanbul, Turkey, Jan. 22-26, 2006
I.
INTRODUCTION
Flow imaging systems based around hair cell sensors are found in nature to be very sensitive compared to traditional solutions in engineering. Spiders [1], crickets, fish [2], and bats all have haircells which can detect sub-nanometer deflections, allowing them to locate prey, avoid predators, and characterize flow conditions around the organism in real time. Such systems would be useful in underwater autonomous vehicles (UAVs), self-stabilizing micro air vehicles (MAVs), as well as in traditional flow-mapping situations. The first step to mimicking a biological flow tracking system is demonstration of highly sensitive flow detection using micro-machined sensors. Here we present the continuing work in characterizing and optimizing an AHC sensor. II.
Figure 1. SEM micrograph of high sensitivity artificial haircell sensor
DEVICE DESIGN
Drawing from biological inspiration, the device consists of a cilium-like polymer hair attached to a piezoresistive Silicon cantilever [3]. The cilium transfers its moment load to the cantilever, exerting a strain at the base. Analytical models of the beam-cilium system with consideration given to fabrication constraints leads to a beam thickness of 2µm, width of 40µm, and length of 200µm, with hair diameter of 80µm and height of 600µm leading to a sensitivity of about 2.5 micro-strain, or under 1mm/s flow rate sensitivity. The semi-empirical model takes into consideration the noise level of our detection circuitry as well as parasitic resistances. The chosen beam dimensions also lead to a resonant frequency on the order of several KHz, far out of the regime of intended use. Ion-implanted strain gauges using boron dopant in light n-type substrate were designed
This research funded in part by DARPA and AFOSR grants.
Figure 2. Schematic drawing of AHC sensor. Cutaway shows cavity beneath cantilever
to be 1/3 of the beam thickness in depth with a doping concentration of 1x1020cm-3 to optimize gauge factor. Wheatstone bridge resistors were also fabricated on die to eliminate thermal noise and signal drift.
AHC S ensor Underw ater AC Flow V elocity S w eep
Sensor Output (V)
0.1
0 .01
0.0 01
0.00 01 0 .0 00 1
0 .0 01
0 .0 1
Flo w Velocity (m /s)
Figure 4. Response of artificial haircell sensor to oscillating flow stimulus. Figure 3. Fabrication process for artificial haircell sensor. (a) SOI wafer doped using ion implantation; (b) oxide insulator grown and patterned; (c) metal contacts and wiring deposited and patterned; (d) cantilever defined by DRIE etch; (e) cavities created through backside DRIE; (f) SU-8 hair created; (g) buried oxide etched to release sensor
DEVICE FABRICATION
Fabrication starts with a Silicon-on-Insulator (SOI) wafer, with 2µm device layer, 2µm buried oxide, and 300350µm handle. Ion implantation with Boron is performed to create bridge resistors and the paddle piezoresistor (Fig. 3a). Implantation energy is chosen to maximize gauge factor in p-type piezoresistor. Thermal oxide is grown to serve as electrical isolation for the metal leads and also to drive in the implanted ions to the proper depth (Fig 3b). Contact windows are opened in the oxide with a wet BHF etch to allow electrical connection to the doped regions. 500 Å titanium and 5000 Å gold are e-beam evaporated and patterned by liftoff to form the wiring (Fig 3c). A DRIE etch defines the cantilever and isolation trenches between bridge resistors (Fig 3d). This is a Bosch-process etch designed to have good selectivity against oxide, which must remain intact for later steps. Polyimide (HD4000) is photodefined to cover everything but the wirebonding areas, cantilevers and previously defined paddle etch region. Curing temperature is chosen to minimize diffusion of gold in the doped contact regions. The handle silicon is then etched from the backside by Bosch-process DRIE (Fig 3e). AZ4620 resist is used as an etch mask. After exposure using a Quintel IR backside aligner, the resist is hard-baked to improve its resistance to the etching species. SU-8 is spun and patterned to create the high-aspect ratio cilia (Fig 3f). This process uses only one coat of SU-8 2075, exposed with a 320nm optical filter to prevent t-topping effects. Finally, a wet oxide etch releases the sensor (Fig 3g). Due to the presence of residues from PRs and Bosch-process deposited carbon on the frontside oxide and the hydrophobicity of the backside silicon cavities, often a high-power oxygen plasma etch is performed on the back side of the sample to assist in the HF etch. Vapor-phase etching was also investigated to help with titanium undercutting, but this only added to the problem.
16.0 14.0
RMS Sensor Output (mV)
III.
AHC Sensor Underwater DC Flow Velocity Sweep
12.0 10.0 8.0 6.0 4.0 2.0 0.0 0.01 -2.0
0.1
1
10
100
Flow Velocity (mm/s)
Figure 5. Response of artificial haircell sensor to steady flow
IV.
EXPERIMENT AND RESULTS
Fabricated sensors were conformally coated with 2µm of Parylene after wirebonding and cable soldering to provide extra environmental protection. The sensors were then tested for sensitivity under steady state (DC) and oscillating (AC) flow. A. AC Flow Characterization To determine the sensitivity under oscillation flow, a packaged sensor was rigidly affixed in a water tank containing a vibrating dipole sphere source. The dipole was a small plastic sphere 6.35mm in diameter attached to a thin but rigid metal rod, actuated by a commercially available one-dimensional membrane based shaker to vibrate at 50 Hz. A commercial accelerometer was attached to the shaker. Given the dipole flow field equation [4]
v flow (r ,θ ) = (a 3U 0
a 3U 0 sin θ cos θ ˆ e + ) ( )eˆθ (1) r 2 r3 r3
Where a is the sphere radius, U0 is the RMS dipole velocity, and the coordinate pair (r,θ) define the position of the sensor with respect to the dipole. Placing the sensor such that the
theta component drops out, we wrote a LabView function to calculate the RMS flow at the tip of our cilia as a function of our accelerometer signal. After setting the distance between the sensor and dipole, the LabView program sweeps through a predefined list of shaker control voltages. A National Instruments DAQ system interfaced with our sensor amplifiers reads in the sensor voltage timetraces. Taking the fast fourier transform (FFT) and choosing the peak corresponding to the dipole frequency allows us to generate a graph showing the sensor signal and its relation to the noise floor, or the sensor output under zero flow condition. The experimental results show the sensor acting in a clear linear regime down to about 0.6mm/s flow rates, where the signal becomes on the order of the electronic and intrinsic sensor noise. Differential sensitivity in the linear region, calculated using the worst-case standard deviation in measurement comes to 0.2mm/s. B. DC Flow Characterization Since flow sensors are often used to determine steady state flow, we calibrated the AHCs under DC flow conditions. In this setup, a packaged sensor was mounted rigidly to a stepper motor-based computer-controlled linear stage system. The stages were controlled by a LabViewbased program to traverse a fluid tank containing still water. Sensor was adjusted to have a 5˚ angle of attack to the direction of traversal in order to help shield from flow separation and vortex shedding effects on the leading edge of the sensor die. Mean voltage output measurements were taken using a National Instruments 12-bit DAQ system during sensor traversal and compared to a baseline measurement taken under zero motion to eliminate DC offsets in bridge tuning. When measuring stead flow, DC level drifting presents problems not seen in oscillating flow. First, the Wheatstone bridge is tuned using a DC voltage offset controller by a PCB-mountable multi-turn potentiometer. These potentiometers are prone to electrical noise introduced by the mechanical wipers inside, and their resistance can shift over time due to thermal effects. Second, electrical power dissipated in the sensor brings about drifts caused by convection and fluid cooling. In our case an excitation voltage of 1.2V is used with a typical sensor resistance of 600-800Ω, giving a dissipated power of about 2mW. This heating and subsequent cooling causes the resistances to be transient, adding both low and high-speed drifts in the signal. This can be combated by lowering the excitation voltage and thus dropping the power quadratically, but one has to keep in mind the fact of the unamplified signal being proportional to this voltage. Finally, given the fluid media itself is conductive, issues arise due to charging and current leakage when the polyimide and especially Parylene layers are damaged or incomplete. Correcting for the slower DC signal drifts in biology is done through neural adaptation, where a persistent signal is eventually tuned out by inhibitory nerve processes. For
calibration experiments, this was simply achieved by taking a baseline mean value moments before starting stage traversal. This is then subtracted from the mean value found while the sensor is experiencing flow, thus zeroing the signal perfectly around the condition of still water. The experiment shows flow detection down to 0.1mm/s. Beyond this range, thermal and electronic drifts dominate the DC signal. V.
DISCUSSION AND CONCLUSION
The sensor presented in this paper is very sensitive compared to the best commercial flow sensors, without some of the problems normally associated with them. Hotwire or hot-film anemometers obtained commercially were tested down to around 1.0mm/s in our lab under similar conditions to our flow sensors, however they suffer from AC signal rectification problems, poor directionality rejection, thermal convection and bubble formation issues. By utilizing a system which is in near thermal equilibrium with the fluid mass, many of the hot-wire’s pitfalls are avoided. Though we report high levels of sensitivity, these devices are still a couple of orders of magnitude off from estimates of threshold sensitivity in the biological systems this sensor takes inspiration from. However, sensitivity of the AHC can be improved further. Improving electrical noise from the simple op-amp circuit could potentially drive down the noise floor. Optimization of the metallization process would yield higher gauge factors and less parasitics in the resistance measurement. Finally, modified cilia geometries to better mimic complex arthropod hair structures would yield better fluid coupling. This is all in addition to previously presented ideas of hydrogel cupulas and fluid canal structures. The work presented here demonstrates a highly sensitive artificial hair cell with the ability to detect both steady state and oscillating flows in the sub-mm/s range. The AHC is judged to be a suitable replacement for hot-wire anemometers, both micromachined and conventional, as well as being a useful tool for the further study of biological sense organs. REFERENCES [1]
[2] [3]
[4]
J. T. Albert, O. C. Friedrich, H. E. Denchant, F. G. Barth, “Arthropod touch reception: spider hair sensilla as rapid touch detectors,” Journal of Comparative Physiology A, vol. 187, pp. 303-312, 2001. S. Dijkgraaf, “The functioning and significance of the lateral-line organs,” Biology Review, vol. 38, pp 51-105, 1962. N. Chen, J. Chen, J. Engel, C. Tucker, C. Liu, “Development and characterization of high-sensitivity bioinspired artificial haircell sensor”, Hilton Head 2006, Hilton Head, South Carolina, June 4-8, 2006. J. Chen, J. Engel, N. Chen, S. Pandya, S. Coombs, C. Liu, “Artificial lateral line and hydrodynamic object tracking,” MEMS 2006 Conference, Istanbul, Turkey, Jan. 22-26, 2006
