INSTITUTE OF PHYSICS PUBLISHING
J. Micromech. Microeng. 12 (2002) 655–661
Design and fabrication of artificial lateral
line flow sensors
Zhifang Fan1, Jack Chen1, Jun Zou1, David Bullen1,
Chang Liu1,3 and Fred Delcomyn2
1
Micro Actuators, Sensors and Systems Group, Micro and Nanotechnology Laboratory,
University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
2
Department of Entomology, University of Illinois at Urbana-Champaign, Urbana,
IL 61801, USA
E-mail: changliu@uiuc.edu
Received 4 March 2002, in final form 23 April 2002
Published 24 June 2002
Online at stacks.iop.org/JMM/12/655
Abstract
Underwater flow sensing is important for many robotics and military
applications, including underwater robots and vessels. We report the
development of micromachined, distributed flow sensors based on a
biological inspiration, the fish lateral line sensors. Design and fabrication
processes for realizing individual lateral line sensor nodes are discussed in
this paper, along with preliminary characterization results.
1. Underwater flow sensing
The fish uses lateral line sensors to monitor surrounding flow
fields for maneuvering and survival under water [1]. A
lateral line system, shown in figure 1, consists of an array
of distributed sensor nodes (the so-called neuromasts) that
span the length of the fish body. Each sensor node, in turn,
consists of a cluster of hair cells embedded in protective, gel-
like domes. An individual biological hair cell, a fundamental
mechanoreceptor, consists of a vertical cilium attached to a
neuron. If the cilium of the hair cell is bent by the local fluid
flow, the neuron attached to the cilium stretches and produces
action potentials. In certain fish species, the lateral line sensor
nodes may be directly exposed at the surface of the skin. In
others, the sensor nodes may be hidden in subdermal channels
in order to minimize wear and damage.
Underwater vehicles and robots require reliable flow
sensing methods. Engineered flow sensors have been
developed in the past based on a number of sensing
principles, including thermal (hot-wire) anemometry [2],
Doppler frequency shift and indirect inference from pressure
differences [3, 4]. Hot-wire anemometers use fine heating
elements that double as temperature sensors. The local flow
rate is inferred from the extent of forced convective heat
transfer from the hot wire. The majority of existing hot-
wire anemometers are conventionally made using macroscopic
3Present address: 313 Micro and Nanotechnology Laboratory, 208 North
Wright Street, Urbana, IL 61801, USA.
0960-1317/02/050655+07$30.00 © 2002 IOP Publishing Ltd Printed in the UK
JOURNAL OF MICROMECHANICS AND MICROENGINEERING
PII: S0960-1317(02)34497-8
machining methods. Micromachined hot-wire anemometers
have also been developed in recent years [5–8]. Sensors based
on Doppler frequency shifts consist of an acoustic launcher
and a receiver. The overall size of the device is generally
large. Existing flow sensors based on pressure distribution
measurements also have large sizes and are generally not
suitable for forming distributed arrays.
In the past two decades, several research groups have
developed micromachined flow sensors that are based on a
variety of sensing principles including the three principles
mentioned above. Microfabrication offers the benefits of
high spatial resolution, fast time response, integrated signal
processing and potentially low costs. Microsensors [9] based
on various principles (including thermal transfer [10–12],
torque transfer [13–16] and pressure distribution [17–19]) have
been demonstrated. In addition to flow sensors, boundary-
layer shear stress sensors have been realized using floating
element methods [20] and thermal transfer principles [21].
However, shear stress sensors are located directly on the fluid-
solid boundary and do not provide adequate information about
mean stream flow velocity.
This work offers an alternative method for underwater
flow sensing. The sensors can have small footprint and high
integration density, if necessary. An array of sensors can be
integrated monolithically. The sensor does not involve thin
fragile wires as in hot-wire anemometry. A proposed system-
level implementation of artificial lateral line sensors is shown
in figure 2. In this paper, we focus on the development of
individual lateral line sensor nodes. The principle of sensing,
655
Z Fan et al
(a )
(b )
(c )
Figure 1. Schematic diagram of fish lateral line systems:
(a) location of lateral line on the surface of a fish; (b) enlarged view
of a segment of the fish lateral line, showing distributed sensors
nodes; (c) schematic diagram of an individual neuron node, called a
neuromast, consisting of a cluster of individual hair cells.
Figure 2. Schematic diagram of distributed artificial lateral line
sensors along the surface of an underwater vehicle or robot.
the fabrication technique and test results are discussed in the
following.
2. Principle of sensing
The schematic diagram of the artificial lateral line sensor
is shown in figure 3. Each sensor consists of an in-plane
fixed-free cantilever with a vertical artificial cilium attached
at the distal, free end. External flow parallel to the sensor
substrate imparts upon the vertical cilium. Due to rigid
connection between the in-plane cantilever and the vertical
656
Figure 3. Schematic diagram of a single artificial hair cell sensor
consisting of a horizontal cantilever with a vertical cilium attached
at the free end. We sense the bending of the vertical cilium using the
strain sensor located at the base of the horizontal cantilever.
cilium, a mechanical bending moment is transferred to the
horizontal cantilever beam, inducing strain at the base of the
cantilever beam. The magnitude of the induced strain can
be sensed by many means, for example, by using integrated
piezoresistive sensors. The overall sensor system may use an
array of individual sensory nodes with systematically varying
positions, height of cilia and orientation. Such a distributed
sensor array can provide information about flow structure
evolution and increase the accuracy of flow field interpretation.
In this paper, we focus on the development and
characterization of individual hair-cell-like sensors.
A comprehensive understanding of the relation between
the output (relative change of resistance) and the mean flow
velocity must take into consideration many possible flow
conditions. For complex situations, it may be necessary
to use the finite-element numerical method to estimate the
sensor output precisely. A comprehensive treatment on the
fluid mechanics aspect, which focuses on the discussion of
new design and fabrication methods, is beyond the scope of
this paper. Nonetheless, the relationship can be analytically
expressed if the flow condition is simple and well defined [22].
We denote the mean-stream flow rate as u0 and the overall
length of the cilium as h. The flow velocity is constant beyond
the boundary layer thickness. The accepted formulae for
estimating the boundary layer thickness (δ) are
5
δ=√ x (1)
Rex
for laminar flow (see page 428 of [22]) or
0.16
δ= x (2)
(Rex )1/7
for turbulent flow (e.g., Rex
> 106), with
 the term Rex being
the local Reynolds number Rex = uν0 x . Supposing a sensor
is located 1 cm away from the leading edge (x = 0.01 m),
the boundary layer thickness is approximately 0.94 mm at
u0 = 0.5 m s−1. In our designs the height of the cilium is
typically less than 1 mm. Hence, for many sensing applications
it is possible that the artificial cilium is entirely immersed
 Design and fabrication of artificial lateral line flow sensors

(a ) (b )

Figure 4. Schematic diagram of the PDMA process: (a) a representative cantilever with electroplated magnetic material attached;
(b) magnetic actuation causes the cantilever beam to rotate off the substrate plane. Plastic deformation occurs near the anchor region.

within the boundary layer. However, complete immersion in
a uniform flow field is not a necessary condition for the sensor
to work properly.
The velocity distribution within the boundary layer is a
function of the vertical distance y. For example, a generally
accepted relationship for low Re conditions (laminar boundary
layer) is
 
2y y2
uy = f (y) = u0 − 2 . (3)
δ δ
If the cilium is entirely embedded within the viscous
√ sublayer
of a turbulent region (which extends up to y+ = y τw /ρ/ν =
10–15 units), the velocity is linearly proportional to y. The
terms τw , ρ and ν are the shear stress at the wall, the fluid
density and the kinematic viscosity, respectively.
To calculate the drag force we divide the vertical cilium
into slices, each with height dy. We calculate the drag force
based on the assumption that the flow velocity is parallel to
the in-plane cantilever. The drag force acting on each slice is


Fy = CD 12 ρu2y w dy (4)
where the term CD is the local drag coefficient, and w is the
width of the vertical cilium in the direction facing the flow.
There are empirical formulae for CD, which is dependent on
Re based on the hydraulic diameter and local flow rate.
The moment applied to the horizontal beam is obtained by
performing finite integration through the length of the vertical
cilium, yielding
 y=h
M= Fy · y. (5)
y=0
The magnitude of induced strain at the base of the cantilever
beam is approximated, under a small bending assumption, as
Mt
ε= (6)
2EI
where E is the Young’s modulus and I is the momentum of
inertia of the in-plane cantilever. If the horizontal cantilever
has a rectangular cross-section with width wl and thickness
t, the expression for the moment of inertia is I = w12l t . The
3
lattice and by the doping level. By assuming that the velocity
profile is given by equation (3), we obtain the following
formula for the sensor output:
 4 
h h6 4h5
R 3GwC D ρ δ 2 + 6δ 4 − 5δ 3 u20
= . (8)
R Ewl t 2
The term w is the width of the vertical cilium. Note that w
and wl are not necessarily equal.
The most important conclusion at this stage is that, as a
first-order approximation, the magnitude of the output signal
is proportional to u20 .
3. Fabrication process
Flow sensors based on flow moment transfer have been
reported in the past. The previously reported devices fall into
two categories. The first type utilizes monolithic cantilever
beams fabricated within the plane of the wafer by using
bulk micromachining methods [13, 14]. Such a sensor is
not applicable for large array formation, as each sensor must
be discretely packaged. The second type utilizes a vertical,
high-aspect-ratio cilium that is connected to micromachined
force-sensing components [23, 24]. The overall yield and
repeatability of the sensors are generally low because the
fabrication process is not monolithic. In this study, we leverage
a recently developed, efficient three-dimensional assembly
process to realize the vertical cilium, a key component of
the hair cell.
3.1. The PDMA three-dimensional assembly process
The vertical cilium is realized using a three-dimensional
assembly technique called plastic deformation magnetic
assembly (PDMA) [25]. The schematic of the process is
shown in figure 4. To illustrate the principle we use a simple
configuration, which consists of a cantilever beam with a
piece of electroplated magnetic material attached to it. The
structure can be realized by using surface micromachining

techniques with one layer of sacrificial material at the

relative change of resistance is given by
bottom. The cantilever beam is released by using sacrificial
R Mt etching. Subsequently, an external magnetic field is applied
=G (7)
R 2EI from underneath the substrate. Under the applied magnetic
where G is the lateral gauge factor of the doped piezoresistor. field, the cantilever beam, together with the magnetic piece,
The magnitude of G is influenced by the orientation of the may be bent out-of-plane. The magnetic field is applied
horizontal cantilever beam with respect to the silicon crystal globally underneath the wafer—many structures on a wafer

657
Z Fan et al

are activated in parallel. The magnetic field can be provided

by an electromagnetic field or a piece of permanent magnet. (a )
If the amount of bending is above a certain threshold, the
cantilever will be plastically deformed, resulting in permanent
displacement. The bent form is retained even after the
magnetic field is removed. (b)
The PDMA process offers a number of advantages. It
requires simple, low-temperature process steps and therefore
is compatible with many materials. The assembly of many
(c )
vertical structures can be achieved simultaneously using a
global magnetic field, hence increasing the efficiency. The
footprint of the vertical structure is small since the vertical
structure does not require external structures to maintain its (d )
upright position.

3.2. Hair cell fabrication process

The overall fabrication process of the artificial hair cell
is discussed in the following and illustrated in figure 5.
The process begins with deposition of a silicon dioxide
layer by thermal oxidation. The oxide layer is patterned
and selectively removed to provide openings for forming
piezoresistive sensors by the diffusion doping process. Boron
pre-deposition is performed at 950 ◦ C for 15 min. It is followed
by oxidation drive-in at 1100 ◦ C for 2 h, reaching an oxidation
thickness of 500 nm. The targeted doping concentration is
1.3 × 10−19 cm−3 and the sheet resistivity of the doped resistor
is approximately 100 /
. The gauge factor of such a doped
piezoresistive sensor is a function of the doping concentration
and the substrate orientation [26, 27].
Structures that are pertinent to the PDMA process are
then deposited and patterned. We deposit a layer of thin-film
copper (300 nm thick) and pattern it photolithographically. A
600 nm thick gold thin film is deposited by using the lift-off
technique. This ensures good step coverage over the boundary
of the sacrificial material and minimizes stress concentration.
The thickness of the subsequently electroplated Permalloy
material is 10 µm. The wafer is etched from the backside
using wet anisotropic etching or deep reactive ion etching
until a silicon membrane with desired thickness is achieved.
Both methods have been implemented successfully although
the deep reactive ion etching is preferred because it allows
easy control of the thickness of the diaphragm. The thickness
of the silicon diaphragm corresponds to the thickness and the
moment of inertia of the in-plane cantilever.
The wafer is then etched from the front side to define the
cantilever beams. Following this step, the wafer is immersed
in diluted HCl solutions to remove the copper sacrificial layer
and to free the gold cantilever beam. We then perform the
PDMA process by using a piece of permanent magnet that
has a magnetic flux intensity of 1470 G. The wafer is slowly
lowered on top of the magnet to permanently deform the metal
cantilevers. The entire wafer is then air-dried. The SEM
micrograph of a representative device is shown in figure 6.
The length, width and thickness of the in-plane cantilever are
1100 µm, 180 µm and 17 µm, respectively. The height (h),
width (w) and thickness of the vertical cilium are 820 µm,
100 µm and 10 µm, respectively.
Improvement has been made to the original PDMA
process to increase the yield of assembly. The cross-sectional
(e )
(f )
(g )
(h )
Figure 5. Fabrication process of an artificial lateral line sensor with
integrated strain gauges. At the final step, the vertical cilium is
assembled into three-dimensional position. It can be further
protected and insulated by uniform, conformal deposition of
Parylene thin film (figure 9).
view of the bending region according to the original PDMA
assembly method is shown in figure 7. Two problems
have been observed. First, the beam has been prone to
breaking during the assembly process due to excessive stress
concentration at the step caused by the edge of the sacrificial
layer. Secondly, the exact location of bending is uncertain,
being between points A and B. This affects the uniformity of
the angle of assembled hair-cell cilia.
A structural modification has been proposed and validated
(figure 8). An extended patch made of 10 µm thick
electroplated Permalloy is electroplated to overlap with the
edge of the sacrificial layer. This produces two notable
improvements. First, the exact location of bending is much
more easily controlled. We confirmed that the bending occurs
at point C. Secondly, bending will not involve the step-
coverage region and therefore the stress-induced fracture is
removed.
The robustness of the sensor and compatibility with the
underwater environment are important issues. Under laminar
flow, the sensor could fail by the fracture of the horizontal

658
 Design and fabrication of artificial lateral line flow sensors

Figure 6. SEM of a single artificial hair cell sensor. The cilium is 820 µm tall. The strain gauge has a nominal resistance of 3 k. The
strain gauge is 5 µm wide and effectively 150 µm long.

magnetic piece magnetic piece

B piece C
A electroplated

(a ) Without Enhancement (b ) With enhancement

Figure 7. Schematic diagram of improved PDMA process: (a) cross-sectional view of bending region according to the original PDMA
process; (b) cross-sectional view of the improved bending region.

beyond the scope of this paper. However, we have conducted

Figure 8. SEM of the plastically bent region at the base of the
vertical cilium.
beam or the vertical cilium, or the breaking of the joint. Under
turbulent flow, the sensor could fail by excessive vibration. The
sensors must also survive for sufficient time period in seawater.
A full experimental and analytical treatment of this subject is
preliminary tasks to strengthen and insulate the sensors.
We have demonstrated that it is possible to strengthen
the plastically deformed joint of the vertical cilium by using
localized electroless plating methods. First, strain gauges
and wire leads are protected from shorting by an organic
insulator such as the photoresist. A thin layer of electrolessly-
plated gold is grown onto the exposed vertical cilium at
95 ◦ C for 10 min with the growth rate being approximately
0.3 µm min−1.
The fabricated device can be further coated with 300 nm
thick uniform Parylene deposition for providing electrical
insulation and increasing strength (figure 9). The Parylene
film also covers electrical bonding wires. This allows sensors
to operate in conducting fluids without electrolysis or shorting.
The deposited material, however, increases the stiffness of the
cantilever beam (figure 10). This is briefly discussed in the
following. The effective force constant of a cantilever beam
made of a homogeneous material with a Young’s modulus E
is approximated by
wl t 3
k= . (9)
3EI
The terms wl and t are the width and thickness of the cantilever,
respectively. As more Parylene is deposited, the value of k

659
Z Fan et al
Figure 9. Schematic diagram of structural strengthening by using
electroless plated metal and Parylene. The thickness of the layers is
not drawn to scale. The thickness of the strengthening layers is
exaggerated intentionally.
Figure 10. View of the bent region after 1 mm Parylene coating.
Figure 11. Schematic diagram of packaged sensor in a fluid flow
environment.
will increase due to increased overall thickness. In summary,
the thickness of the Parylene coating must be optimized with
respect to two competing concerns, electrochemical protection
and sensitivity.
4. Results and discussion
The sensors are mounted on a thin glass plate, which is placed
within a laminar flow water tunnel (figure 11). One edge of the
glass is polished to present a sharp profile facing the flow. The
sensor is located 1 mm from the leading (sharp) edge. The flow
rate of the water tunnel is varied manually. Laminar flow with
660
2670
R = 62.563u2 + 8.2756u + 2594
2660
R2 = 0.9989
Resistance (Ohms)
2650
2640
2630
2620
2610
2600
2590
0 0.2 0.4 0.6 0.8 1
Mean flow rate (m/s)
Figure 12. Measurement results showing output current versus flow
rate of a representative sensor under constant voltage biasing.
u0 ranging from 0 to 1 m s−1 passes the sensor element. The
flow impacts the cilium at its broad side, i.e., parallel to the
long axis of the substrate cantilever. The sensor is biased under
constant voltage (1 V) and the output current is monitored.
The output response is shown in figure 12. The best-fit curve
follows a second-order polynomial expression, as expected
from earlier analysis.
We are currently developing arrayed sensors to
characterize the response of the sensors with respect to the
orientation of the flow and to the height of the artificial cilium.
It is believed that in order to increase the sensitivity, one can
increase the height of the cilium or reduce the thickness of
the in-plane cantilever. However, the vertical cilium should
not be excessively tall as this may increase the chance of
damage. For long cilia and higher flow rates, the interpretation
of flow is more challenging as multiple flow regimes (laminar,
transition or turbulent) may be applied through the length of
the cilium. Understanding the exact nature and profile of
local flow would require multiple-sensor clusters much like
the biological sensors.
5. Conclusions
We have developed the design and fabrication process of an
individual micromachined artificial lateral line sensor for flow
sensing applications. The sensor is realized using combined
bulk micromachining methods and an efficient three-
dimensional assembly method. Sensors can be strengthened
using electroless plating and Parylene conformal coating.
They can be realized monolithically using combined bulk
micromachining and three-dimensional assembly methods.
Acknowledgments
This work is conducted with funding provided by the US Air
Force Office of Scientific Research (AFOSR), the National
Science Foundation (NSF) and NASA.
References
[1] Bond C E 1996 Biology of Fishes 2nd edn (Philadelphia, PA:
Saunders)
[2] Perry A E 1982 Hot Wire Anemometry (Oxford: Clarendon)

[3] Richter M, Wackerle M, Woias P and Hillerich B 1999 A novel
flow sensor with high time resolution based on differential
pressure principle Proc. 12th Int. Conf. on Micro Electro
Mechanical Systems (Orlando, FL) pp 118–23
[4] Rediniotis O K and Vijayagopal R 1999 Miniature multihole
pressure probes and their neural-network-based calibration
AIAA J. 37 666–74
[5] Jiang F, Tai Y C, Ho C M, Rainer K and Garstenauer M 1994
Theoretical and experimental studies of micromachined
hot-wire anemometer Digest IEEE Int. Electron Devices
Meetings (IEDM) (San Francisco) pp 139–42
[6] van Baar J J, Wiegerink R J, lammerink T S J, Krijnen G J M
and Elwenspoek M 2001 Micromachined structures for
thermal measurements of fluid and flow parameters
J. Micromech. Microeng. 11 311–8
[7] Ebefors T, Kalvesten E and Stemme G 1998 Three
dimensional silicon triple-hot-wire anemometer based on
polyimide joints Proc. 11th Annual Int. Workshop on Micro
Electro Mechanical Systems: An Investigation of Micro
Structures, Sensors, Actuators, Machines and Systems pp
93–8
[8] van der Wiel A J, Linder C, Rooij de N F and Bezinge A 1993
A liquid velocity sensor based on the hot-wire principle
Sensors Actuators A A37–A38 693
[9] Kovacs G T A 1998 Micromachined Transducers Sourcebook
(New York: McGraw-Hill)
[10] van Honschoten J W, Krijnen G J M, Svetovoy V B,
de Bree H-E and Elwenspoek M C 2001 Optimization of a
two wire thermal sensor for flow and sound measurements
Proc. 14th Int. Conf. Micro Electro Mechanical Systems
(MEMS’2001) pp 523–6
[11] de Bree H-E, Jansen H V, Lammerink T S J, Krijnen G J M and
Elwenspoek M 1999 Bi-directional fast flow sensor with a
large dynamic range J. Micromech. Microeng. 9 186–9
[12] Lofdahl L, Stemme G and Johansson B 1992 Silicon based
flow sensors used for mean velocity and turbulence
measurements Exp. Fluids 12 270–6
[13] Svedin N, Stemme E and Stemme G 2001 A static turbine flow
meter with a micromachined silicon torque sensor Technical
Digest MEMS 2001: 14th IEEE Int. Conf. on Micro Electro
Mechanical Systems (Interlaken, Switzerland) pp 208–11
[14] van der Wiel A J, Boillat M A and de Rooij N F 1995 A
bi-directional silicon orifice flow sensor characterised for
fluid temperature and pressure 8th Int. Conf. on Solid-State
Sensors and Actuators and Eurosensors IX vol 2 pp 420–3
[15] Svedin N, Kalvesten E, Stemme E and Stemme G 1998 A new
silicon gas-flow sensor based on lift force
J. Microelectromech. Syst. 7 303–8
Design and fabrication of artificial lateral line flow sensors
[16] Enoksson P, Stemme G and Stemme E 1996 A Coriolis mass
flow sensor structure in silicon Proc. 9th Annual Int.
Workshop on Micro Electro Mechanical Systems: An
Investigation of Micro Structures, Sensors, Actuators,
Machines and Systems pp 156–61
[17] Boillat M A, van der Wiel A J, Hoogerwerf A C and
de Rooij N F 1995 A differential pressure liquid flow sensor
for flow regulation and dosing systems Proc. IEEE Micro
Electro Mechanical Systems pp 350–2
[18] Lofdahl L, Kalvesten E, Hadzianagnostakis T and Stemme G
1996 An integrated silicon based wall pressure-shear stress
sensor for measurements in turbulent flows Proc. 1996 Int.
Mechanical Engineering Congress and Exposition
(New York, NY, USA) pp 245–51
[19] Kalvesten E, Vieider C, Lofdahl L and Stemme G 1996 An
integrated pressure-flow sensor for correlation
measurements in turbulent gas flows Sensors Actuators
A 52 51–8
[20] Padmanabhan A, Goldberg H, Breuer K D and Schmidt M A
1996 A wafer-bonded floating-element shear stress
microsensor with optical position sensing by photodiodes
J. Microelectromech. Syst. 5 307–15
[21] Xu Y, Jiang F, Lin Q, Clendenen J, Tung S and Tai Y C 2002
Under water shear stress sensor MEMS’02: 15th IEEE Int.
Conf. on Micro Electro Mechanical Systems (Las Vegas,
USA, 20–24 Jan.)
[22] White F M 1999 Fluid Mechanics 4th edn (New York:
McGraw-Hill)
[23] Ayers J, Zavracky P M, McGruer N, Massa D, Vorus V,
Mukherjee R and Currie S 1998 A modular
behavioral-based architecture for biomimetic autonomous
underwater robots Proc. Autonomous Vehicles in Mine
Countermeasures Symp. (Naval Postgraduate School)
CD ROM, webpage http://www.cis.plym.ac.uk/cis/
InsectRobotics/biomimetics.htm
[24] Ozaki Y, Ohyama T, Yasuda T and Shimoyama I 2000 An air
flow sensor modeled on wind receptor hairs of insects Proc.
MEMS 2000 (Miyazaki, Japan) pp 531–6
[25] Zou Jun Chen J and Liu C 2001 Plastic deformation magnetic
assembly (PDMA) of out-of-plane microstructures:
technology and application IEEE/ASME J. MEMS 10
302–9
[26] Pfann W G and Thurston R N 1961 Semiconducting stress
transducers utilizing the transverse and shear
piezoresistance effects J. Appl. Phys. 32 2008–19
[27] Middelhoek S and Audet S A 1989 Microelectronics and
Signal Processing: Silicon Sensors (New York:
Academic)
661