JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO. 5, OCTOBER 2007 1185
RF MEMS Sequentially Reconfigurable Sierpinski
Antenna on a Flexible Organic Substrate
With Novel DC-Biasing Technique
Nickolas Kingsley, Member, IEEE, Dimitrios E. Anagnostou, Member, IEEE,
Manos Tentzeris, Senior Member, IEEE, and John Papapolymerou, Senior Member, IEEE
Abstract—Devices and systems that use RF microelectro-
mechanical systems (RF MEMS) switching elements typically use
one switch topology. The switch is designed to meet all of the
performance criteria. However, this can be limiting for highly dy-
namic applications that require a great deal of reconfigurability. In
this paper, three sets of RF MEMS switches with different actua-
tion voltages are used to sequentially activate and deactivate parts
of a multiband Sierpinski fractal antenna. The implementation
of such a concept allows for direct actuation of the electrostatic
MEMS switches through the RF signal feed, therefore eliminating
the need for individual switch dc bias lines. This reconfigurable
antenna was fabricated on liquid crystal polymer substrate and
operates at several different frequencies between 2.4 and 18 GHz
while maintaining its radiation characteristics. It is the first in-
tegrated RF MEMS reconfigurable antenna on a flexible organic
polymer substrate for multiband antenna applications. Simulation
and measurement results are presented in this paper to validate
the proposed concept. [2007-0013]
Index Terms—Liquid crystal polymer (LCP), multiband,
reconfigurable antenna, RF microelectromechanical systems
(RF MEMS), Sierpinski fractal antenna.
I. INTRODUCTION
T
HE RF microelectromechanical systems (RF MEMS)
switches are quickly becoming a popular switching ele-
ment among microwave designers. Their low loss, small size,
excellent isolation, and low distortion make them attractive for
a wide range of applications. They have already been integrated
into filters [1], [2], antennas [3]–[5], phase shifters [6], and
many other RF devices.
Designers typically optimize the MEMS geometry to meet
a given specification. Switches can be made wide and short
or narrow and long to meet a specific size requirement. The
materials can be tailored to meet a desired actuation voltage.
Manuscript received January 22, 2007; revised April 27, 2007. This work
was supported by the National Science Foundation (NSF) under Grant
ECS0500860. Subject Editor S. Lucyszyn.
N. Kingsley was with the School of Electrical and Computer Engineering,
Georgia Institute of Technology, Atlanta, GA 30308 USA. He is now with
the Modeling and Design Group, Auriga Measurement Systems, Lowell, MA
01854 USA (e-mail: kingsley@gatech.edu).
D. E. Anagnostou is with the Electrical and Computer Engineering De-
partment, South Dakota School of Mines and Technology, Rapid City,
SD 57701 USA.
M. Tentzeris and J. Papapolymerou are with the School of Electrical
and Computer Engineering, Georgia Institute of Technology, Atlanta, GA
30308 USA.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JMEMS.2007.902462
The height of the membrane can be tuned to give a certain
level of isolation when in the “
OFF” position. The inductive
and capacitive regions can be designed to work best at a given
frequency. For each application, and in all existing literature,
there is typically one switch membrane geometry, and it is used
exclusively throughout the system.
However, using only one switch membrane geometry can be
limiting if the device needs maximum reconfigurability. It could
be beneficial to utilize multiple switch membrane geometries
to add an additional level of reconfigurability. For example, to
provide the lowest possible loss from the switching element,
several different switches could be used in parallel which are
tuned for different operating frequencies. As the frequency is
changed, the switch that works best at that frequency is used.
Since MEMS switches offer excellent “
OFF”-state isolation
(usually better than 30 dB), the presence of the additional
switches would have a negligible effect on the device. This
same technique could be used to select switches of different
impedances, switching speed, isolation, capacitance, etc. A sys-
tem could also select between ohmic and capacitive switches to
operate from dc to very high frequencies. This technique is ideal
for applications that need maximum reconfigurability and can
tolerate the slight size increase from the additional switches.
To demonstrate the effectiveness of using multiple switch
geometries in a working system, this paper presents a three-
iteration coplanar waveguide (CPW)-fed Sierpinski gasket
monopole antenna that is reconfigured by turning on various
RF MEMS switches. Different areas of the antenna geometry
are sequentially activated and deactivated by changing the dc
voltage present at the RF feed. This method eliminates the need
for dc bias lines at each MEMS switch, which will improve the
radiation characteristics of the antenna.
In the past decade, fractal or prefractal shapes have been
introduced in antenna and array designs. Several of these de-
signs have been extensively studied, including Koch [7]–[9],
Hilbert, Peano, Minkowski, and Sierpinski [10], [11] geomet-
rical shapes or array arrangements [10], demonstrating both
compactness and multiband behavior. A comprehensive review
[12] describes in detail, among others, the multiband function
of the Sierpinski gasket monopole and dipole antennas.
In the majority of the published literature, integration has
been accomplished on rigid and nonflexible semiconducting or
organic polymer substrates such as silicon and FR-4. The idea
of integrating RF MEMS switches into a multiband self-similar
antenna was first implemented in [3], where the entire system,
1057-7157/$25.00 © 2007 IEEE
1186 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO. 5, OCTOBER 2007
Fig. 1. Illustration of a MEMS reconfigurable Sierpinski antenna. The center-
line of the CPW feed provides the RF input and dc voltage for MEMS switch
actuation. Switches S1 and S1’ actuate at a low voltage, switches S2 and S2’
actuate at a medium voltage, and switches S3 and S3’ actuate at a high voltage.
Fig. 2. Four different reconfigurable antenna states: State 1 has no voltage
applied, state 2 has a low voltage applied, state 3 has a medium voltage applied,
and state 4 has a high voltage applied. The activated (radiating) part of the
antenna is darkened.
including the RF MEMS, the planar self-similar antenna, and
the CPW–coplanar strip transition, was fabricated on silicon. In
this paper, the integration is achieved on a very thin and flexible
liquid crystal polymer (LCP) substrate.
Since all of the switches share a common dc feed, this tech-
nique provides reconfigurability without the need for additional
bias lines. This is advantageous since dc bias lines take up
space, add loss, and reduce the bandwidth of a device. This
technology is particularly useful for antennas where bias lines
can have a pronounced effect on radiation patterns. In this
paper, simulation and measurement results are presented with
good agreement.
II. I
MPLEMENTATION OF RECONFIGURABILITY
The implementation of a sequentially activated antenna is
shown in Fig. 1. All of the MEMS switches used are single-
supported (cantilever type) and ohmic. Regardless of the ap-
plied voltage, the triangular element that is closest to the RF/dc
input is always active (Fig. 2, state 1). When no dc voltage is
applied, the antenna radiates at its highest frequency.
When a low dc voltage is applied to the signal line, the
first set of MEMS switches (S1 and S1’) actuate, and this
activates the second level of triangular elements (Fig. 2,
state 2). The antenna now radiates at a lower frequency. Since
all of the switches are ohmic, the low voltage is also present
at the membrane of the next set of switches (S2 and S2’).
However, these switches are designed to actuate at a higher
voltage so they are unaffected by the voltage present.
When a higher dc voltage is applied, the first set of MEMS
switches (S1 and S1’) remains in the “
ON” position while the
second set of switches (S2 and S2’) actuates (Fig. 2, state 3).
This activates the next iteration, consisting of six additional
radiating elements. Again, this higher voltage is present at the
next set of switch membranes (S3 and S3’), but the electrostatic
force created is not sufficient for actuation.
Finally, when the voltage is increased to its highest value, the
first two sets of switches (S1 and S1’ and S2 and S2’) remain
in the “
ON” position while the remaining set of the switches
(S3 and S3’) actuates (Fig. 2, state 4). In a way, the voltage
cascades from one state to the next like a sequence of overflow-
ing buckets. This technique could not be used if the switches
were capacitive since they do not pass dc voltage. The four
different states are illustrated in Fig. 2, where all of the activated
regions for different voltages are dark in color.
This biasing technique allows for direct actuation of the
electrostatic MEMS switches through the RF feed structure.
Since only the RF feed is dc grounded, the switches actuate with
the use of a floating ground. That is, the signal pin of the CPW
feed is connected to the dc cathode, and the ground pins are
corrected to the anode of a bias tee. The electrostatic charges
that are created during switch actuation can dissipate through
the substrate and be removed by the dc-grounded RF feed
when in the “
OFF” state. This method has been successfully
documented in [6].
The reduction or elimination of bias lines is highly advanta-
geous because they can significantly distort the radiation pat-
terns and they can introduce additional unwanted resonances.
III. RF MEMS S
WITCH DESIGN PROCEDURE
To change the actuation voltage of a MEMS switch, there are
four parameters that can be changed.
1) Membrane material: Switch membranes are almost al-
ways made of metal. This is due to their pliable nature.
Stiffer metals (that is, those with a high Young’s Modulus
E) will have a higher actuation voltage than those with a
lower Young’s Modulus.
2) Bridge thickness: The thicker the bridge, the stiffer the
membrane. This gives a higher actuation voltage.
3) Bridge height: The higher the bridge, the larger the gap
between the metal layers. This decreases the electrostatic
force and increases the actuation voltage.
4) Membrane geometry: Springs can be designed into the
shape of the membrane to lower the actuation voltage.
Of these parameters, only the fourth one does not require any
fabrication changes. Making changes to a fabrication process
can be a costly endeavor and may add additional variables.
For example, it can be more challenging to precisely control
the membrane height or thickness. For these reasons, we chose
to alter the membrane geometry. By carefully controlling the
spring constant (κ) of the switch membrane, the actuation
voltage can be tailored to a desired value.
A. RF MEMS Switch Design and Simulation Results
An accurate method for determining the actuation voltage for
a given switch geometry was published in [13]. In this method,
a switch is simulated using the static structural mechanics
module from FEMLAB 3.0 [14]. FEMLAB, by Comsol, is
KINGSLEY et al.: RF MEMS SEQUENTIALLY RECONFIGURABLE SIERPINSKI ANTENNA ON A SUBSTRATE 1187
Fig. 3. Procedure for determining the optimal force needed to deflect the
switch membrane. If the force is too high, the deflection is more than the
membrane height. If the force is too low, the deflection does not reach
the substrate. The optimal force is determined when the deflection matches
the membrane height. The ground symbols denote the location of the floating
ground.
Fig. 4. Varieties of switch geometries are shown. The stationary posts are
shown in black. The dotted areas show the electrostatic regions. The switches
are listed from lowest actuation voltage (lowest spring constant) to the highest
actuation voltage (highest spring constant). All dimensions are labeled.
a multiphysics simulator that uses the finite-element method.
Any mechanical simulator that can perform a force–deflection
analysis can use this method. Once the geometry and material
specifications have been loaded into the software, a force can
be applied to the beam over the electrostatic area, and the
deflection can be determined. The force is changed until the
deflection matches the desired bridge height. This procedure is
demonstrated in Fig. 3.
The actuation voltage V can then be calculated using
V =
2g
2
F
(1)
where g is the gap (membrane height), F is the force per
area, and is the free-space permittivity. In [13], the measured
voltage was within 5 V of the expected voltage.
For the mechanical simulations, it was assumed that an
aluminum bridge with a thickness of 1.5 µm was used that
was suspended 5.0 µm above the substrate. Aluminum has a
Young’s Modulus (E) of 70 GPa, a Poisson’s Ratio (ν) of
0.33, and a density (ρ) of 2700 kg/m
3
. Single-supported (can-
tilever) ohmic switches were chosen although this technique
could be used with other topologies. The switch geometries
shown in Fig. 4 were loaded into FEMLAB with the mechan-
ical and material properties stated before. These geometries
were chosen because they have a wide variety of spring con-
stants. They were also tuned to give a convenient ratio to the
Switch 1 actuation voltage. That is, Switch 2 has an activation
voltage that is 1.5 times higher than that of Switch 1, and
TAB LE I
S
IMULATED PULL-DOWN FORCE AND THE CALCULATED
PULL-DOWN VOLTAGE FROM (1)
Fig. 5. Simulated deflections for the switch geometries from Fig. 4 are shown.
The darkest areas represent the location of the posts where there is no deflection.
Switch 3 has an activation voltage that is two times higher than
that of Switch 1, etc.
The simulated force that resulted in a 5-µm deflection for
each of the geometries is given in Table I. The deflection is
shown in Fig. 5. These values were entered into (1) to calculate
the pull-down voltage. These values are also given in Table I.
Careful considerations were made to ensure complete and
symmetric actuation of the switch membranes. All the voltages
presented in this paper are for the full pull-down state of the
switch membrane. This was verified visually by increasing
the pull-down voltage until the “
ON”-state resistance and RF
insertion loss values converged. Ohmic switches that are par-
tially actuated will have a higher “
ON”-state resistance and RF
insertion loss than a fully actuated switch.
B. RF MEMS Switch Measurement Results
The switches shown in Figs. 4 and 5 were fabricated on
LCP substrate. However, this technique would work for most
substrates. The fabricated switches are shown in Fig. 6. They all
have a measured resistance of 1.7 Ω in the “
ON” position and
a measured capacitance of approximately 35 fF in the “
OFF”
position. This low capacitance provides excellent isolation in
the “
OFF” position.
The minimum voltage was measured by starting at 0 V and
increasing by 2 V every second. This increment was chosen
because it is important to actuate the switch before substan-
tial dielectric charging occurs. When the switch actuated, S-
parameter measurements were taken. These results are shown
in Figs. 7 and 8. The measured pull-down voltages agreed well
with the expected values. These results are given in Table II.
IV. A
NTENNA DESIGN PROCEDURE
To date, Sierpinski gasket antennas have been fabricated
on many different rigid substrates with low permittivity (such
1188 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO. 5, OCTOBER 2007
Fig. 6. Fabricated MEMS switches modeled after the designs shown in
Figs. 4 and 5.
Fig. 7. S-parameter measurement results when the switch membrane is in the
up position (not actuated).
Fig. 8. S-parameter measurement results when the switch membrane is in the
down position (actuated).
TAB LE II
C
OMPARISON OF CALCULATED AND MEASURED PULL-DOWN VOLTAGES.
A
LL MEASURED VOLTAGES ARE WITHIN 2VOF THE ACTUAL
MINIMUM VALUE DUE TO THE 2-V/s INCREMENT
Fig. 9. Photograph of the fabricated Sierpinski antenna with MEMS switches
shown. The design parameters are labeled on the plot.
as CuClad) and high permittivity (such as silicon). LCP was
chosen as the substrate for its numerous advantages. LCP is a
thin (100 µm) flexible low-loss (tan δ ≈ 0.004) low-moisture-
absorbing material with low permittivity (ε
r
≈ 3) [15]. Since
the material is a polymer, there are additional packaging and
cost advantages. All of these characteristics make it an ideal
substrate for antennas, particularly at high frequencies.
With respect to the geometry, the antenna elements have
a60
◦
flare angle and maintain the resonant structure’s self-
similarity with a log-periodicity of δ =2. The antenna is fed
through a 6-mm-long CPW transmission line with a 50-µm
gap, a 1.3-mm signal conductor width, and a 1.5-µm-thick
aluminum layer. Switch geometries 1–3 from Figs. 4–6 were
used to implement switches S1, S2, and S3, respectively. A
picture of the fabricated antenna is shown in Fig. 9. The overall
size of the antenna, including the feed, is 20 × 25 mm.
The CPW feed was chosen to facilitate the measurement
setup. This reconfigurable antenna operates at four different
principle frequencies. For each of these frequencies, the an-
tenna maintains its multiband performance.
The antenna was simulated using IE3D,
1
a method-of-
moments electromagnetic solver. The simulated return loss is
shown in Fig. 10. The switches were modeled in two ways.
First, they were simplified to a 200 × 400-µm gap in the “
OFF”
position and by a metal pad of the same size in the “
ON”
position. Those results were compared to a simulation that
included the MEMS geometry in the “
OFF” and “ON” positions.
The difference in the results between the two simulations was
minor, which indicates that the isolation provided by the MEMS
was adequate.
It was verified that the antenna has a different first reso-
nant frequency for each of the four states. Since the antenna
is self-similar with a log-periodicity of two, each time the
antenna transitions to the next state, the frequency should
be halved. That is, the resonant frequency of state 2 should
be half that of state 1. The simulated E-plane copolarization
(zy-plane, ϕ =90
◦
) patterns for the four states are shown in
Fig. 11. These patterns are as expected for a monopole antenna.
The simulated radiation pattern for the H-plane copolarization
1
IE3D is a trademark of Zeland Software.
KINGSLEY et al.: RF MEMS SEQUENTIALLY RECONFIGURABLE SIERPINSKI ANTENNA ON A SUBSTRATE 1189
Fig. 10. Simulated return loss for all four states of the designed reconfigurable
antenna.
Fig. 11. Simulated radiation pattern for the E-total copolarization (zy-plane,
ϕ =90
◦
) plane for all four states of the designed reconfigurable antenna at the
first resonant frequency. It is clear that the MEMS have a minimal effect on the
antenna patterns, as it maintains its broadside characteristics.
(xz-plane, ϕ =0
◦
) is not presented for brevity since it shows
an omnidirectional pattern in that plane.
V. MEMS S
WITCH INTEGRATION
The placement of the RF MEMS switches was illustrated in
Fig. 1 and shown in Fig. 9. In order to bias the ohmic switches
for electrostatic actuation, the MEMS need to have an applied
voltage. A metal pad beneath the switch should be present to
attract the charged metal. The metal pad must be placed under a
thin dielectric material (such as silicon nitride) to prevent direct
metal bridge to metal pad contact. Otherwise, no charge will
develop, and the switch will not actuate.
Traditionally, the actuation voltage is applied via a dc bias
line. However, in order to prevent RF leakage into the dc
path, careful attention needs to be given to the dc bias lines
themselves. This can be implemented in different ways.
1) By using a quarter-wavelength transmission line con-
nected to a quarter-wavelength open-circuit radial stub.
Alternatively, a half-wavelength transmission line with-
out a radial stub can be used with a reduced bandwidth.
Each MEMS switch would require a different dc bias line
and, for this antenna, that would require six lengthy metal
lines being added. This would have a pronounced effect
on the antenna performance. Therefore, this solution is
not advisable.
2) High-resistance lines have been investigated to provide a
wider bandwidth alternative [16]. Aluminum-doped zinc
oxide is one example, which is used for biasing in [3].
Thin films of this kind are generally deposited using
combustion chemical vapor deposition, which uses very
high temperatures. This is not a problem for materials
like silicon, but it is much higher than the melting point
of the organic substrate (≈315
◦
C) used in this paper. At
the moment, very high-resistivity materials that can be
deposited at low temperature are not widely available but
are under investigation [17].
The proposed alternative to these approaches is to eliminate
the need for individual switch dc bias lines. Instead, the biasing
is handled through the antenna structure itself. Here, the dc
voltage and the RF signal are both applied to the antenna
through the same signal conductor of the CPW feed line. The
antenna reconfigurability is made possible by using MEMS
switches of varying actuation voltages.
Like all RF MEMS devices, self-actuation of the switches
can be an issue for the antenna. If the RF signal ever becomes
large enough to actuate the switches, then the antenna will
remain in state 4. This will occur because these switches
have an actuation time of approximately 40 µs. This is almost
100 000 times slower than the period of the wavelength at
2.4 GHz (the lowest operating frequency of the antenna). Ef-
fectively, all of the switches will remain in the “
ON” position
(state 4). This antenna should only be used at normal RF MEMS
switch power levels (micro- to milliwatt range).
VI. A
NTENNA AND MEMS FABRICATION
Fabrication and MEMS integration was performed in six
general steps. First, the LCP material was polished using an
alumina slurry until the surface roughness was approximately
10 nm. This roughness is comparable to that of a polished
silicon wafer. Therefore, the original polymer roughness has
no effect on the switch or the antenna performance. Second,
the bottom seed layer was electron-beam deposited. Third,
a silicon nitride layer was deposited using plasma-enhanced
chemical vapor deposition, patterned, and etched using a
reactive ion etch. Fourth, a sacrificial photoresist layer was
patterned to define the switch height. Fifth, a 1.5-µm-thick
