0% found this document useful (0 votes)
166 views174 pages

RF MEMS Switches and Switch Circuits: Shimul Chandra Saha

RF-MEMS SWITCHES

Uploaded by

Raj
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
166 views174 pages

RF MEMS Switches and Switch Circuits: Shimul Chandra Saha

RF-MEMS SWITCHES

Uploaded by

Raj
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 174

Shimul Chandra Saha

RF MEMS Switches and


Switch Circuits
Modeling of RF MEMS switches and development of RF
MEMS capacitive switches and MEMS tunable filters

Thesis for the degree of Dr. Ing

Trondheim, August 2008

Norwegian University of Science and Technology


Faculty of Information Technology, Mathematics and
Electrical Engineering
Department of Electronics and Telecommunications
NTNU
Norwegian University of Science and Technology

Thesis for the degree of Dr. Ing.

Faculty of Information Technology, Mathematics and Electrical Engineering


Department of Electronics and Telecommunications

© Shimul Chandra Saha

ISBN 978-82-471-1150-5 (printed ver.)


ISBN 978-82-471-1151-2 (electronic ver.)
ISSN 1503-8181

Doctoral theses at NTNU, 2008:230

Printed by NTNU-trykk
Preface

The research work presented in the thesis was carried out as a partial fulfillment of the
requirements to obtain the degree of “Doctor Ingeniør” (PhD/Dr. Ing.) at the department
of Department of Electronics and Telecommunications (IET), Faculty of Information
Technology, Mathematics and Electrical Engineering (IME), Norwegian University of
Science and Technology (NTNU), Trondheim, Norway. I was employed as a research
fellow at the department within the project Smart Microsystem for Diagnostic Imaging
in Medicine (SMiDA). The funding for the project was provided by Norwegian research
council. I would like to thank Norwegian Research Council for their financial support to
my work.

I would like to express my sincere thanks to my principle supervisor Prof. Trond


Sæther, to give me the opportunity to work on RF MEMS, an exciting and new area of
RF and Microwave technology. I also like to thank him for his continuous support,
guidance and encouragement during my last 3 and half years study and work. I would
also like to express my gratitude to my supervisor Prof. Tor Fjeldly for his
encouragement during my work and his help in articles preparation and course work
completion. I am also grateful to my both supervisors for helping me in proof reading
and their advice on my thesis writing.

I would like to thank Prof. Trond Ytterdal, SMiDA project leader to introduce me to
Prof. Trond Sæther initially and his continuous friendly and encouraging attitude during
my stay at the department. I would like to thank all SMiDA family members, circuit and
system group members and colleagues at the IET department for their friendly and
helping minded behavior during my stay at the department. Special thanks to Sigrid
Berg for her help in proof reading of the thesis.

I would like to express my deepest gratitude to Dr. Geir Uri Jensen from SINTEF MiNa
lab, to introduce me to the MiNa lab at Oslo for the fabrication of the switches and
filters and his help during my work in the lab. I would also like to thank Dr. Håkon
Sagberg for his continuous help during process work inside clean room and his valuable

i
advice on various issues regarding process development. He was very encouraging and
friendly all the time to me. I like to thank Dr. Erik Poppe for his help in the various
process developments, especially metallization, nitride deposition and various dry
etching. I would also like to express my gratitude to all the process engineers and
scientists at MiNa lab for their help in various process developments and their friendly
behavior.

Special thanks to Assoc. Prof. Ulrik Hanke, HIVE (previously at SINTEF ICT,
Trondheim), for his valuable help in my thesis work. He was always very patience and
encouraging to me when I went to him for any kind of discussion. He provided guidance
in modeling and simulation work, also in preparing articles for submission to various
conferences and journals. I would also like to thanks Dr. Anton M. Bøifot, Dr. Kirsten
Husby and Dr. Bengt Holter for their help in characterization of the switches and other
issues.

I like to express my deepest thank to my family members, especially my parents,


brother, sister and uncle for their ever lasting support as well as their encouragement to
my work. They were very kind and responsive to me always. Warm regards from my
heart to my wife Sharmistha Saha for her lovely, supportive and encouraging attitude to
my work. Also special thanks to my parents in law. I would also like to thank my
friends from school, college and universities and all my relatives for their support
during all these years.

Trondheim, August 2008

Shimul Chandra Saha.

ii
Abstract

This thesis presents the design and fabrication of Radio Frequency


Microelectromechanical Systems (RF MEMS) switches and switch circuits. The work
primarily concerns novel MEMS tunable low pass and band pass filters. A novel model
of spring constant and pull-down voltage for non-uniform bridge and cantilever is also
reported in this work. It also presents RF MEMS capacitive switches, first ever to be
produced in Norway. The reliability of the switch is very good compared to the present
state of art works presented elsewhere. Modeling and design of high speed switches and
high capacitance ratio switches are performed. Design and simulation of tunable low-
pass and band-pass filters are also presented. A model for pull down voltage of a non-
uniform bridge and cantilever is developed. Fabrication of the capacitive switches and
MEMS tunable filters is conducted in the SINTEF MiNa lab. Measurements of pull
down voltage and characterization of RF performance for the switches and filters are
also performed.

A high speed switch (short switching time) is required to switch the capacitive
micromachined ultrasonic transducer (CMUT) array for medical diagnostic application.
A switching time of less than 5 µs is required to switch the same CMUT from the
transmitter to the receiver electronics. A study is performed on different beam
structures, like bridges and cantilevers, different beam materials and damping
coefficients. The simulations are performed in CoventorWare to verify and compare the
results. A high capacitance ratio shunt switch is designed and simulated for RF front
end technology. The switch has to work from 10-32 GHz, with a certain specified
insertion and return loss in upstate and isolation in downstate. A capacitance ratio of
180 is specified for these requirements. A switch is designed and simulated, with a
reasonably low pull-down voltage. The switch performances are verified using
simulation in MathCad, ADS (Advance design system) and HFSS (High Frequency
Structure Simulator).

Tunable low-pass and band-pass filters at C to X band are designed for transmitter and
receiver blocks and RF front end circuits. The filters are designed using both

iii
transmission line and MEMS variable capacitors (a novel approach). Stepped
impedance transmission line low-pass filters are used as basic building blocks. The low
impedance is replaced by shunt capacitors. The band-pass filter is obtained using the
basic low-pass filter and series capacitors. The cut-off frequency for the low-pass filter
and the center frequency for the band pass filter can be tuned by shunt capacitors. The
bandwidth of the band-pass filter can be tuned by the series capacitors. The design and
simulation of the filters are performed in ADS with verification in HFSS.

A novel model for spring constant and pull-down voltage of non-uniform bridges and
cantilevers is developed through basic force deflection calculation. The non-uniform
beam, which is wider at the electrode area and narrower at the anchor area, can be used
to reduce the pull-down voltage. To our knowledge, very few reports exist on accurate
modeling of non-uniform beams to date. We have developed a model for the spring
constant and the pull-down voltage of non-uniform beams, considering the distributed
force, using basic force-moment calculation. The model includes residual stress (bridge)
and stress gradient (cantilever). The model pull-down voltage is verified by
CoventorWare simulations, and the results agree well with simulation results.

The fabrication of RF MEMS shunt capacitive switches and MEMS tunable filters is
performed at SINTEF, Micro and Nano Lab (MiNa) Lab, Oslo, Norway. Several
experiments were performed to develop the various process steps. An experiment on the
hard-baking parameters was conducted to tune the sacrificial resist layer angle at the
anchor position. An elaborate experiment was performed on DC sputtering to develop a
recipe for low tensile stress gold for the suspended bridge. Also an experiment was
performed on release methods to release the RF MEMS switch successfully without any
residue left.

A range of capacitive shunt switches are fabricated with various widths and lengths. The
fabricated switches have slightly different dimensions than the simulated high
capacitive ratio switches. This is done in order to facilitate the fabrication and to obtain
a successful prototype in the first round. A reduced initial gap of 2.5 µm is used for the
first prototype. A thicker dielectric layer of ~200 nm is also used instead of the 100 nm

iv
used in the simulations. The switch RF performance is measured using a vector network
analyzer at Norwegian University of Science and Technology (NTNU). The switch has
reasonably low insertion loss and low return loss in the upstate. A higher upstate
capacitance is found than the simulated value due to a reduced initial gap and a wider
bridge. A comparatively lower isolation is measured in the downstate due to reduced
capacitance. A reduction in the downstate capacitance occurred, due to a thicker
dielectric. A lower downstate capacitance can also occur, due to surface roughness and
non-planarity of the bridge at the downstate.

Low-pass and band-pass filters are also fabricated, using traditional parallel plate
capacitances. The filter performance is also measured using a vector network analyzer.
Tuning of both center and cut-off frequency and also bandwidth is obtained by actuating
different combination of shunt and series capacitors. The measured results agree very
well with the simulations results.

v
Contents:

Preface i
Abstract iii
Contents vi

1. Introduction 1
1.1: A brief history of the MEMS switch 1
1.2: RF MEMS technology 3
1.3: RF MEMS Applications areas 3
1.4: Performance of RF MEMS switches 6
1.5: Limitation of RF MEMS switches 7
1.6: RF MEMS devices 8
1.7: Present status of the RF MEMS 11

2. Motivation 17
2.1: Smart Microsystem for Diagnostic Imaging in Medicine (SMiDA) 17
2.2: Integrated reconfigurable Radio Front-end Technology (IRRFT) 22
2.3: Combined work for the above mentioned projects 23

3. Thesis overview 25
3.1: Modeling and simulation of various RF MEMS switches and switch circuits 25
3.2: Development of various process steps for fabrication of the RF MEMS
switches and switch circuits 29
3. 3: Measurement of the fabricated RF MEMS switches and switch circuits 36

4. Discussion 39

5. Conclusions 47
Correction list 49

Appended papers

vi
1. Introduction to RF MEMS

1.1: A brief history of the MEMS switch

Radio Frequency Microelectromechanical Systems (RF MEMS) are becoming a popular


technology for many RF applications. Although MEMS have been developed since the
1970s for sensors, accelerometers, gas chromatographs, and other sensor devices, it did
not get much attention from the RF and Microwave frequency community for a long
time. MEMS switches use the same principle as the simple mechanical moving switch,
like basic single pole single through (SPST). The beam moves mechanically either to
make an open or short circuit. The difference is that it is miniaturized and works at RF
and Microwave frequencies. Figure 1.1 shows a basic MEMS cantilever switch with a
separate actuation electrode. When a DC voltage is applied between the actuation
electrode and the suspended beam, an electrostatic force will develop and pull the beam
down. This will create a contact between the signal in and signal out electrodes.
Depending on the configuration the contact can be two types: DC contact and capacitive
contact.

Figure 1.1: A basic MEMS switch

Under the support of DARPA (Defense Advanced Research Projects Agency), Dr. Lary
Larson at the Hughs research labs in Malibu, California, developed the first MEMS
switch (a varactor) in 1990-91 [1.1]. The switch was specifically designed for
microwave applications. Although the switch was not mature and reliable, it
demonstrated excellent RF performance up to 50 GHz. The initial results of this
research were very outstanding and stirred the interest of several groups in the U.S.

1
government. By 1995, Rockwell Science Center and Texas Instruments both had
developed outstanding RF MEMS switches [1.1]. The Rockwell switch was a DC
(metal-to-metal) contact type, suitable for DC-60 GHz applications [1.1]. Texas
Instrument developed a capacitive contact switch suitable for 10-120 GHz applications.
From that time on many universities, laboratories and companies have been actively
pursuing the research on RF MEMS devices. Nowadays there are also many companies
working in this area, such as Motorola, Analog Devices, Samsung, Omron, NEC, and
ST-Microelectronics [1.1]. The potential of RF MEMS in high frequency applications is
also presented in a paper by Linda P. B. Katehi et al. [1.2].

Today there is a great interest and activity in research and development of RF MEMS
due to their huge potential in both commercial and defense application. At the same
time, significant advancements are taking place in the semiconductor industry, in areas
such as HEMTs (high electron mobility transistors), MOSFETs (metal oxide
semiconductor field effect transistors), etc. The gate lengths are becoming smaller and
the power consumption is reduced for CMOS (complementary metal oxide
semiconductor). But the advancement in the semiconductor switching technology, for
example, on p-i-n diodes has not advanced in the same rate. The cut-off frequency of p-
i-n diode switches is not advancing as the same rate as other technologies. The cut-off
frequency of a device is the frequency, at which the on and off impedance becomes
equal. For satellite and short-range communication applications, the operating
frequency is increasing and they are approaching to the THz range. So a new switching
technology is required to meet the demands for higher cut-off frequencies. RF MEMS
device can fulfill these demands at a lower loss. The cut-off frequency of a typical RF
MEMS switch can go up to 40 THz, which is truly outstanding. RF MEMS are
nowadays becoming very popular due to their very low insertion loss, high isolation and
high operating frequency. The bias circuitry is very simple compared with that of p-i-n
diodes and FET switches. The power consumption of the RF MEMS switch is in the
micro watt range which is very low. This will make them very suitable for portable
device applications, like mobile handsets and satellite communication.

2
1.2: RF MEMS technology

There are several RF MEMS technology areas where the research is going on
nowadays. They are briefly mentioned below [1.1].
1) RF MEMS switches, varactors and inductors are nowadays mature enough for use
in practical application. They can operate from DC to several tens of GHz with very
good RF performances. In the RF MEMS switch, the suspended beam can move up to
several micrometers during actuation without any problem.
2) Acoustic vibrations can be used in FBAR (thin film bulk acoustic resonator) and
filters. They have very excellent performance with very high quality factor (Q) up to
several GHz. They are widely used for wireless applications, particularly in cellular
telephones. The interdigital surface acoustic wave (SAW) filters are usually large at low
frequencies but they can be miniaturized at higher frequency.
3) Lots of research is taking place in the field of mechanical resonators. The mechanical
movement of the beam is very small, in the range of nanometers. The structure can be a
bridge, a cantilever and a radial disk. The mechanical resonance frequency of the bridge
or cantilever depends on the dimensions and material of the beam. These resonant
elements can be used in filters, oscillators (for reference clock), etc. Usually they have
very high Q. Both bulk and surface micromachining technique can be used for the
processing of such devices. The resonance frequency is not very high at present. As the
process technology is advancing, smaller devices can be fabricated with better control in
resolution and higher resonance frequency.

1.3: RF MEMS Applications areas

RF MEMS have very good RF performance in the RF and Microwave region. DC


contact switches provide very good insertion loss and isolation at low frequency from
DC to several tens of GHz. Capacitive contact switches provide very good insertion loss
and isolation at higher frequency from 3-4 GHz to 100 GHz. The power consumption of
RF MEMS switches is very low or close to zero, so it can be used in portable systems
where the battery life time is very important. The cut-off frequency of RF MEMS

3
switches is 30 to 50 times better than their counterparts. They can be used in phase
shifters and tunable circuits (matching networks, filters, etc.).

The prominent applications for RF MEMS lie in the following systems such as wireless
handsets, base stations, space applications, test and instrumentation, RF
instrumentation, aerospace and defense, etc [1.3, 1.4, 1.5, 1.6, 1.7]. For RF radio
convergence toward a single multi band radio, two clear conditions should be met,
reduced size and power consumption. RF MEMS ohmic contact switches and variable
capacitors can be very efficient in meeting these requirements [1.5]. Reconfigurability is
a key issue in base stations in terms of various air interface standards, like Global
System for Mobile communications (GSM), Universal Mobile Communications System
(UMTS), and Code Division Multiple Access (CDMA). Also each system has various
frequency bands where the system can operate. RF MEMS technology can perform this
reconfigurability with great flexibility [1.5].

Satellites need to manage the home, ground, mobile and space information with limited
available power generation and storage capabilities. Thus RF MEMS are a very suitable
candidate as they consume very little power during transmission and also in the standby
period [1.5]. They are also very lightweight, which is an advantage in space
applications.

Figure 1.2: Potential RF MEMS based system applications [1.6, *].

4
A typical wireless connectivity for a satellite application is shown in Figure 1.2. The
prospects of a RF MEMS mass market are promising according to [1.3, 1.6, and 1.7].
The application areas and the lifetime of RF MEMS devices and number of cycles
required for a typical system are shown in Table 1.1 [1.1, p7]. The subsystems and
circuits that can be benefited from RF MEMS technology are also mentioned in the
Table 1.2 [1.1, p7].

Table 1.1: Application areas of MEMS switches, varactors, and High-Q Inductors [1.1]

Area System Number of cycles Years


(Billions)
Phased arrays Communications systems Ground 1-10 2-10
Space 10-100 2-10
Airborne 10-100 2-10
Phased arrays Radar systems Ground 10-100 5-10
Space 10-100 5-10
missile 0.1-10 1-5
Airborne 1-100 5-10
Automotive 1-10 5-10
Switching and Wireles portable 0.01-4 2-3
reconfigurable s Base station 0.1-100 5-10
networks Commu
nication
s
Satellite (communications 0.1-1 2-10
and radar)
Airborne (communication 0.1-10 2-10
and radar)
Instrumentation 10-100 10
Low-power Wireless communications 0.1 2-3
oscillators and (portable)

5
amplifiers (varactors, Satellite (communication 0.1-1 2-10
inductors) and radar)
Airborne (communication 0.1-10 2-10
and radar)

Table 1.2: Subsystems and circuits that can benefit from RF MEMS elements [1.1]

RF MEMS Elements Switch, Varactor, Inductor


Switching networks
Transmit/receive switches
Very high isolation switches (instrumentation)
Programmable attenuators
Phase shifters (digital and analog)
Reconfigurable antennas
Reconfigurable matching (or impedance) networks
Reconfigurable Butler matrices for multi-beam systems
Tunable filters
Switched filter banks
Miniature
Switched diversity antennas, oscillators, amplifiers
Low phase-noise oscillators (fixed and tunable)
High-efficiency networks (low-power systems)

1.4: Performance of RF MEMS switches:

RF MEMS switches have extremely low insertion loss during the on-state and very high
isolation during the-off state. They are very suitable for operation at RF and Microwave
frequency region. They are extremely linear and have very low intermodulation
products. They consume very little power during actuation and can provide a very high
capacitance ratio compared to their counterparts. The performance of RF MEMS

6
switches compared to their counter parts, PIN diode and FET switches, is shown in table
1.3 [1.1, p5]:

Table 1.3: Performance comparison of FETs, PIN Diodes, and RF MEMS [1.1]

Parameter RF MEMS PIN FET


Voltage (V) 10-80 r3-5 3-5
Current (mA) 0 3-20 0
Power consumption 0.05-0.1 5-100 0.05-0.1
(mW)
Switching time 1-300 Ps 1-100 ns 1-100 ns
Cup (series) fF 1-6 40-80 70-140
Rs (series) : 0.5-2 2-4 4-6
Capacitance ratio 40-500 10 n/a
Cutoff frequency (THz) 20-80 1-4 0.5-2
Isolation (1-10 GHz) Very high High Medium
Isolation (10-40 GHz) Very high Medium Low
Isolation (60-100 GHz) High Medium None
Loss (1-100 GHz) (dB) 0.05-0.2 0.3-1.2 0.4-2.5
Power handling (W) <1 <10 <10
Third-order intercept +66-80 +27-45 +27-45
point (dBm)

1.5: Limitation of RF MEMS switches

Although RF MEMS switches have very good RF performance they have some
limitations also. Firstly they require comparatively high actuation voltage. The
switching time is in the range of micro seconds, slower than their counterparts.
Reliability is one of the major issues in commercialization of RF MEMS switches. The
lifetime of a switch can be affected by factors like environment, contact degradation,
dielectric stiction, and mechanical failures [1.1, 1.8].

7
RF MEMS switches are very sensitive to moisture or out-gassing. A slight amount of
moisture may cause stiction in the switches and make them useless. Contact degradation
occurs mainly in ohmic or DC contact switches. The contact resistance may change or
micro welding may happen after a certain time or a certain numbers of cycles. Dielectric
stiction occurs in the capacitive switch. This happens due to electric charge storage in
the dielectric layer. The charge can lead the switch to stay at upstate or stick to
downstate; also it may cause a drift in the actuation voltage. Usually the mechanical
failure is minimal compared with the other failure mechanisms. A ductile and less
wearable metal is required for the suspended beam to have a better mechanical
performance.

In order to reduce the stiction due to environment, RF MEMS switches need hermetic
packaging. This will increase the cost of the switch. Cost is still a major issue for
commercialization of RF MEMS switches. In order to penetrate the mass mobile
market, RF MEMS have to be in a price range, which is lower than or similar to their
counterparts. The RF power handling capability of MEMS switches is still lower than
in their counterparts. There is a lot of research going on to address these limitations.

1.6: RF MEMS devices

Various kinds of basic RF MEMS switches are shown in Figures 1.3 -1.6. From an
electrical contact point of view the MEMS switches can be two types: a) Capacitive
contact, and b) DC contact. A dielectric is deposited on top of the bottom electrode of
the capacitive contact switch as shown in Figure 1.3 and 1.5. When the suspended beam
is in the upstate the capacitance is in the range of fF. When a DC actuation voltage is
applied between the actuation electrode and the suspended beam, the suspended beam
will move downward and collapse on the bottom electrode. This will increase the
capacitance in the range of pF, 20-100 times higher than the upstate capacitance. The
upstate capacitance depends mainly on the initial gap. The downstate capacitance
depends on the dielectric thickness, dielectric constant etc. SiO2, Si3N4, TiO2 can be
used as dielectric for RF MEMS switches. The capacitive contact switch is suitable for

8
high frequency application. At low frequency (DC), the impedance always will be very
high for a capacitive switch, whatever the capacitance value is.

Figure 1.3: A capacitive contact shunt bridge

For DC contact switches, a metal to metal contact is created between the suspended
beam and the bottom electrode or transmission line, as shown in Figures 1.4 and 1.6.
When the beam is in the upstate the resistance is very high, in the range of Mega Ohm.
When the beam is actuated and in the downstate, a metal to metal contact is formed and
the resistance becomes very small, in the range of 0.1-1 Ohm. The DC contact switch is
suitable for low frequency application (from DC to several tens of GHz). For a DC
contact switch a separate actuation electrode is always required. For capacitive contact
switches, the same electrode can be used for the electrostatic actuation and the
capacitive contact, as shown on Figure 1.3. A separate actuation mechanism can also be
used for capacitive contact switches, as shown in Figure 1.5.

Figure 1.4: A DC contact shunt bridge.

9
Depending on the electrical configuration, RF MEMS switches can be two types: a)
shunt, b) series. For a shunt switch, as shown in Figures 1.3 and 1.4, the switch will stay
on electrically, when the beam in upstate, and off when the beam is in downstate. For a
series switch as shown in Figures 1.5 and 1.6, the switch will stay on when the beam is
in downstate and off when the beam is in upstate.

Figure 1.5: A capacitive contact series cantilever


In terms of the mechanical structure the switches fall in two types: a) fixed-fixed bridge,
b) free end cantilever. Figures 1.3 and 1.4 shows the fixed-fixed bridges and Figures 1.5
and 1.6 shows the free end cantilever beams. For fixed-fixed bridge the beam is rigidly
anchored at both ends. This makes the beam very stiff and the spring constant very high.
For free end cantilever the beam is rigidly anchored at one end and the other end is
freely suspended. The cantilever is much less stiffer than the bridge [1.1]. For similar
dimensions of the bridge and the cantilever, the pull down voltage is higher for the
bridge than the cantilever.

Residual stress in the bridge and stress gradient in the cantilever may develop during the
fabrication process. The pull down voltage for a bridge will increase due to tensile
stress. The positive stress gradient in the cantilever will buckle the beam upwards and
increase the pull down voltage. A compressive stress in the bridge will buckle the
bridge upward or downward. A higher value of compressive stress may make the beam
useless. A negative stress gradient in the cantilever will bend the beam downward. A

10
higher value of stress gradient may bend the beam enough to touch bottom electrode
and lead to a useless switch.

Figure 1.6: A DC contact series cantilever.

For most of the MEMS switches, the beam movement is vertical as shown in Figures
1.3-1.6. A lateral movement can also be used, but they are not very common to date.
The actuation mechanism is predominantly electrostatic as discussed already. The other
actuation mechanisms are thermal, magneto-static, and piezoelectric. The sizes of these
kinds of switches are larger than for electrostatic actuation switches. For miniaturization
and low power consumption, electrostatic switches are mostly used for RF and
microwave applications.

1.7: Present status of the RF MEMS


Although RF MEMS started with great promises, they had suffered some
disappointment. This is due to reliability issues, packaging, price and inability to deliver
to mass market in time as promised. The RF MEMS hype curve is shown in Figure 1.7,
according to Jeremie Bouchaud and Bernardo Knoblich, [1.3, 1.7].

11
Figure 1.7: The hype curve of RF MEMS Switch [1.3, *]

From the Figure 1.7, it can be seen that the RF MEMS switches started with a great
expectation and had a sudden rise until 2003, when Teravicta and Magfusion had
announced their first sample. The expectation curve suffered a major blow after 2003.
This happened as they could not provide samples to all prospective customers in time
and the performance was not as good as specified. After that many companies and
research groups were discouraged and stopped their work on RF MEMS. Luckily some
patient and enthusiastic researchers and companies still continued their work on RF
MEMS. Now RF MEMS have left the ‘trough of disillusionment phase’ and emerged
shining into the ‘slope of enlightenment’. The present RF MEMS trend and their future
are also described in an interview [1.4].

At present there are a number of companies who have started to commercialize RF


MEMS switches. Among them Teravicta, Radant, Advantest and Matsushita are
prominent. They are only available in USA market yet. The price of a commercial RF
MEMS switch is still high compared with their counterparts, like FET and PIN diode
switches. Radant has the cheapest switch with a price range of $ 25 for a SPST and $ 40
for a SPDT switch [3]. Until 2006 the Market for RF MEMS switches was only $ 6
million. The forecast says that the market can be up to $ 210 million by 2011. The

12
market forecast for RF MEMS switch is shown in Figure 1.8. A commercial RF MEMS
SPST switch from Radant is shown in Figure 1.9.

Figure 1.8: The market forecast for RF MEMS switches [1.3,*].

Figure 1.9: A SPST hermetically-sealed MEM switch [1.6,*].

Figure 1.8 shows the market potential only for RF MEMS switches. There are many
other RF MEMS components presently in the market at component and system level.
They are tunable capacitors, BAW resonators, micro-mechanical resonator, MEMS
inductors, and cavity resonators. The expected implementation scheme for RF MEMS

13
components is shown in Figure 1.10. The total market for RF MEMS component in
2004 was US $ 126 million. The forecast for RF MEMS turnover by 2009 is shown in
figure 1.11.

Figure 1.10: Implementation of RF MEMS by applications, 2009 [1.7,*]

14
Figure 1.11: Total turnover forecasts for the RF MEMS market by applications, 2004-
2009 [1.7,*].

References:
[1.1] G. M. Rebeiz, RF MEMS Theory, Design and Technology, John Willey and Sons,
2003.
[1.2] Linda P. B. Katehi, F. Harvey, Elliott Brown, MEMS and Si Micromachined
Circuits for High-Frequency Applications, IEEE Transactions on Microwave Theory
and Techniques, Vol. 50, No 3, March 2002.
[1.3] J. Bouchaud and B. Knoblich, RF MEMS switches deliver early promises, MEMS,
Microfluidics, Microsystem Executive review, MEMS Investor Journal, October 11,
2007.
[1.4] RF MEMS: A brief history and future trends, MEMS, Microfluidics, Microsystem
Executive review, MEMS Investor Journal, October 05, 2006.

15
[1.5] Hector J. De Los Santos, Georg Fischer, Harrie A. C. Tilmans, and Joost T. M.
Van Beek, RF MEMS for Ubiquitous Wireless Connectivity, Part 2-Application, IEEE
Microwave Magazine, pp 50- 65, December 2004.
[1.6] Hector J. De Los Santos, Scott Rassoulian, and John Maciel, MEMS for future
microwave systems, 2005 IEEE MTT-S, 905-908.
[1.7] Jeremie Bouchaud, Bernardo Knoblich, Henning Witch, Will RF MEMS live up
their promise?, 36th European Micorwave Conference, Sept 2006, pp 1076-1079.
[1.8] Brandon Pillans, Gabriel Rebeiz, Jeong-Bong Lee, Advances in RF MEMS
Technology, GaAs IC Symposium, 2003, pages 17-20.

* Permission obtained from the respective authors through e-mail communications

16
2. Motivation of the work

MEMS technology is on the verge of revolutionizing high frequency applications.


Present and future RF systems require lower weight, volume, cost and power
consumption, especially in portable devices and satellite communications. The system
also demands increased functionality and frequency of operation. Reconfiguration is
another important issue for present RF system, as they work in different operating
standards and frequency bands. As discussed in introduction chapter 1, RF MEMS can
address these requirements very efficiently. RF MEMS consume very little power (µW).
They have very high cut-off frequency compared with their counter part PiN diode and
FET switches, and can operate up to 40 THz. They have very low insertion loss and
high isolation over a wide frequency range 0 to 100 GHz. The bias circuitry is also very
simple compared with the FET and PIN diode switches. RF MEMS switches are still
not commercially mature enough. At present very few commercial RF MEMS switches
are present at market. There are not many commercial foundries available to date, to
fabricate RF MEMS component. So research and development on design and
fabrication of RF MEMS switches and switch circuits are very prospective and essential
for the advancement of RF MEMS products.

This thesis is based on the work performed on two projects named: Smart Microsystem
for Diagnostic Imaging in Medicine (SMiDA) at the Department of Electronics and
Telecommunications (IET), Norwegian University of Science and Technology (NTNU)
[2.1] and Integrated Reconfigurable Radio Front-end Technology (IRRFT) at the
Department of Information and Communication Technology (ICT), SINTEF [2.2]. A
brief introduction of theses projects and the motivation to use RF MEMS components
are presented below.

2.1: Smart Microsystem for Diagnostic Imaging in Medicine (SMiDA)

It is believed that more than 80% of sudden heart attacks are caused by rupture of
vulnerable plaques, which lead to blood clots in the coronary arteries. The rupture
occurs in the unstable plaque. Figure 2.1 shows a coronary artery with possible plaque.

17
A smart and convenient way is required to detect the plaque in the coronary arteries and
also to characterize them as stable or unstable. The project aim is to develop the smart
microsystem for detection of plaque with characterization. Capacitive Micromachined
Ultrasonic transducer (CMUT) is proposed as a device to image the plaque in the
coronary artery. An array of CMUT will be placed on a catheter shown in Figure 2.2
with interface electronics inside the catheter. A brief discussion of CMUT and RF
MEMS involvement in the imaging system is mentioned in the following paragraphs.

Figure 2.1: A coronary with possible plaque [2.3]

CMUTs are an upcoming technology in the field of medical imaging which promises to
offer advanced automation possibilities. They will also offer, unlike the conventional
piezoelectric transducer probes, compatibility with the CMOS process and integration

18
of the signal processing circuitry on the probe itself. This will drastically reduce the
hard-wiring of the probe to the signal processing console. The possibility of signal
processing near the transducers opens new prospects of ‘Smart probes’ capable of
handling large amounts of data, required for example, for 3D ultrasound imaging.
Another imaging application of the CMUTs is in the detection of plaque inside the
coronary arteries for an early heart disease diagnosis.

Figure 2.2: A catheter with CMUT element in the blood vessel [2.4]

The active part of a the CMUT is usually a metal coated Si3N4 membrane. A highly
doped silicon substrate constitutes the bottom electrode. When a DC bias voltage is
applied between the top and bottom electrodes, the membrane is pulled towards the
substrate (electrostatic actuation). If an AC voltage is applied along the bias point, a
harmonic motion is obtained; thus the membrane will generate an ultrasonic wave into
the medium outside the transducer. As in piezoelectric ultrasound probes, there are two
modes of operation namely, transmit, and receive. CMUT elements require to be
switched ON and OFF at different times either to send the actuation signal, or to receive
the reflected signal. Hence it is necessary to have a switching matrix to switch the

19
elements from transmit mode to receive mode as shown in Figure 2.3. Also the
elementary mechanism for CMUT switching between transmitter and receiver, with
bridge and cantilever are shown in Figures 2.4 and 2.5. As the actuation signal is low, it
is necessary to have a low insertion-loss and very good isolation of the switching
matrix, for which RF MEMS switches are a good candidate. However as the vein radius
is small, the time of flight of the signal to travel to the vein wall and back to the catheter
surface is very short. Thus it is necessary to design the switch with very short switching
time. The release time is less important considering the switching configuration. For the
CMUT application a DC contact switch will be necessary as the operating frequency is
in the range of 30 MHz. A switch with a switching time (from upstate to downstate) of
less than 5 Ps is desired for the above application. Some works on reasonably fast
switch are published in [2.5, 2.6]. They are not fast enough to meet our requirement.

Figure 2.3: A CMUT array with transmit and receive block including switching control.

20
Figure 2.4: The possible switching mechanism for CMUT with MEMS cantilever

Figure 2.5: The possible switching mechanism for CMUT with MEMS bridge

RF front end circuits for the transmitter and receiver blocks, as shown in Figure 2.3 is
also required for the medical application. Power consumption is the key issues in such

21
kind of application. In terms of power saving, RF MEMS are a prominent technology,
as they consume very little power. So RF circuits, specially filter with MEMS varactor,
inductor are of prime importance for RF front end circuits. Also MEMS technology will
better facilitate integration. MEMS capacitances will also be very useful for RF
architectures and signal processing units, such as oscillators, LNAs, etc. Traditionally
the half wavelength length filters are larger in dimensions and also their tuning range is
not that high. In our work we proposed a stepped impedance transmission line filter,
which is compact. Some works on stepped impedance filters are published in [2.7, 2.8
and 2.9]. In our work, we considered a continuous of tuning frequency using a MEMS
varactor at relatively lower frequency.

2.2: Integrated reconfigurable Radio Front-end Technology (IRRFT)

This project involves design and implementation of different RF circuits for front end
technology. They includes antenna, efficient power amplifier, low noise amplifier, low
noise high frequency synthesizer etc. The antenna can be reconfigured using RF MEMS
switches and capacitances. The frequency synthesizer can be tuned using RF MEMS
capacitances. This will reduce the power consumption of the device. The possible use of
MEMS components in a typical transceiver application is shown in Figure 2.6.

Figure 2.6: Bottom-up approach: The green (grey on black-white copy)-colored sub-
systems in the conventional transceiver are replaced by MEMS based components [2.2].

22
An array of identical antenna elements is used in the antenna architectures. The
elements are separated by a half-wavelength of the resonant frequency of each
individual element. MEMS switch capacitors can be used to connect these antennas. By
tuning the MEMS capacitances the antenna resonant frequency and polarization can be
changed. The frequency synthesizer can be consisting of a phase lock loop (PLL) and a
voltage controlled oscillator (VCO). The capacitance tuning can be done using MEMS
varactors instead of PIN diodes. The capacitances can be changed using an actuation
voltage. An initial case study is performed on different kinds of RF MEMS switches, in
term of performances like isolation, insertion loss, frequency of operation and process
overview. According to SINTEF personnel, a capacitive shunt switch with high
capacitance ratio (~180) is preferable for their RF front end application [2.10].

2.3: Combined work for the above mentioned projects

This thesis is based on the work performed on the SMiDA and the IRRFT projects. In
order to reduce the pull-down voltage, for both the high-speed switch and high
capacitance ratio switch, a non-uniform beam is preferable. Although some works are
present [2.11, 2.12] on this kind of modeling, they are not accurate enough. So a
comprehensive model for spring constant and pull-down voltage of the non-uniform
bridge and the cantilever is developed. A fabrication foundry is required to fabricate RF
MEMS switches and switch circuits for the SMiDA projects. The IRRFT project from
SINTEF has a similar goal, to fabricate the RF MEMS capacitive switches. They
planned to fabricate their switch in SINTEF Micro and Nano Lab (MiNa Lab), Oslo,
Norway. So we collaborated with SINTEF to fabricate our switch together at MiNa lab.
Instead of using technology by others, our goal was to develop our own technology. In
terms of fabrication, we have chosen a capacitive switch for the first prototype.
Capacitive switches require less process steps than DC contact switches. Our preference
was to develop a working device rather than making a device with precise
specifications. For the first prototype, our goal was to fabricate capacitive shunt
switches and MEMS tunable filters, with similar process parameters.

23
References:
[2.1] Trond Ytterdal et al, Project description: Smart Microsystem for Diagnostic
Imaging in Medicine, May 2003.
[2.2] Ulrik Hanke et al, Project Description: Integrated Reconfigurable Radio Front-end
Technology, SINTEF memo.
[2.3] Scientific America, May 2002.
[2.4] Trond Ytterdal, A presentation on the project: Smart Microsystem for Diagnostic
Imaging in Medicine
[2.5] D. Peroulis, S. Pacheco , K. Sarabandi and L. P. B. Katehi, Electromechanical
considerations in developing low-voltage RF MEMS switches, IEEE Transactions on
MTT, Vol.51, No. 1 January 2003.
[2.6] N. Nishijima, J. Hung, G. M. Rebeiz, A Low-voltage high contacts force RF-
MEMS switch, IEEE MTT-S Digest, 2004.
[2.7] D. Peroulis, S. Pacheco , K. Sarabandi and L. P. B. Katehi, Tunable Lymped
Components with Applicatins to Reconfigurable MEMS Filters, Microwave
Symposium Digest, 2001 IEEE MTT-S, Volume 1, 20-25 May 2001 Page(s):341 - 344
vol.1
[2.8] S. Lee, J. Kim , J . Kim, Y. Kim. and Y. Kwon, Millimeter-Wave MEMS tunable
low pass filter with reconfigurable series inductor and capacitive shunt switches, IEEE
Microwave and Wireless Components letters, vol 15, No 10, October 2005. pp 691-693.
[2.9] J. Park, S. Lee, J. Kim., H. Kim, Y. Kim. and Y. Kwon, Reconfigurable
millimeter-wave filters using CPW-based periodic structures with novel multiple-
contact MEMS switches, IEEE Journal of Microelectromechanical Systems, Vol 14,
Issue 3, June 2005, pp 456-463.
[2.10] Ulrik Hanke, Anton M. Bøifot and Geir U. Jensen, SINTEF Memo on Integrated
Reconfigurable Radio Front-end Technology, Assessment and performance for various
capacitive switches. Version 1.3, July 2005.
[2.11] S. Afrang and G. Rezazadeh, Design and Simulation of Simple and Varying
Section Cantilever and Fixed-Fixed End Types MEMS Switches, Proceeding of ICSE
2004. pp. 593-596.
[2.12] L. Lv, Z. Deng, F. Zhao, Y. Liu, K. Han, Analysis and Simulation of RF MEMS
for Wireless Communication, Proceedings of ISCIT 2005, pp 1095-1098.

24
3. Thesis Overview

This thesis is based on the work performed on the SMiDA project from NTNU and the
IRRFT project from SINTEF. Modeling and simulation of various switches and filters
were performed initially. The fabrication of the switches is done in the SINTEF Micro
and Nano Lab (MiNa). We had to develop all the process steps by ourselves, as the
fabrication of RF MEMS switches had not been performed earlier at MiNa lab. The
thesis work is divided into three parts.

3.1 : Modeling and simulation of various RF MEMS switches and switch circuits
3.2: Development of various process steps for fabrication of the RF MEMS switches
and the switch circuits
3.3: Measurement of the fabricated RF MEMS switches and switch circuits

3.1: Modeling and simulation of various RF MEMS switches and switch circuits

At first the modeling and simulation of various types of switches, beam structures, and
MEMS tunable filters are presented. Design and simulation is performed on a high
speed switch and a high capacitance ratio switch. Novel electromechanical models of
spring constant and pull down voltage for non-uniform beam bridges and cantilevers are
developed. Design and simulation of MEMS tunable filters with a transmission line and
MEMS variable capacitors are also performed. The modeling and simulation work of
this thesis is based on the following papers. A brief description of the paper is also
added with the paper titles.

Paper 1: Shimul Chandra Saha, Tajeshwar Singh, Trond Sæther, “Design and
Simulation of RF MEMS Switches for High Switching Speed and Moderate Voltage
Operation”, In proceedings of IEEE PRIME 2005, pp 207-210, July 2005,
EPFL, Lausanne, Switzerland.

This paper describes the modeling and simulation of high switching speed RF MEMS
switch for CMUT switching application. In the simulation, a capacitive contact bridge

25
with comparatively large dimensions is considered for initial design. The targeted
switching time (from upstate to downstate) is less than 5 µs. The switching time for
different metals, like gold and aluminum is simulated and compared. The switching
time for various damping coefficient is also considered. The damping coefficient can be
controlled by pressure of the media and introducing holes in the suspended beam. The
switching time can also be reduced by increasing the actuation voltage above pull-down
voltage.
Paper 2: Shimul Chandra Saha, Tajeshwar Singh, Trond Sæther, “Design and
simulation of RF MEMS cantilever and bridge switches for high switching speed and
low voltage operation and their comparison”, In proceedings of IEEE ISSCS 2005, pp
131-134, July 2005, Iasi, Romania.

This paper also describes the modeling and simulation of high speed RF MEMS switch
for CMUT switching application. The targeted switching time (from upstate to down
state) is less than 5 µs. A comparison between the bridge and the cantilevers as beam
structures is performed. Different metals like Gold, Aluminum, and Nickel is considered
as beam material. The damping factor in term of pressure in the media and perforation
in the suspended beam is also considered. The switching speed can be increased by
increasing the actuation voltage, above pull down voltage. The simulations are
performed in CoventorWare and a suitable switch is proposed for the CMUT
application. In the simulation, capacitive switches with relatively larger dimensions are
considered for the initial design, considering fabrication issues. The smaller dimension
of DC contact switches can be rescaled and designed easily afterwards.

Paper 3: Shimul C. Saha1, Ulrik Hanke2, Geir U. Jensen3, Anton M. Bøifot4, Tor A.
Fjeldly1, Trond Sæther1, “Modeling and Simulation of High Capacitance Ratio RF
MEMS Capacitive Shunt Switch”, In proceedings of third IASTED international
conference on Circuits, Signals and Systems CSS 2005, pp 186-193, October 2005,
Marina Del Rey, CA, USA.

This paper describes the design, modeling and simulation of the RF MEMS switch for
IRRFT project. The switch will be used for RF front end applications, such as antennas,

26
frequency synthesizers, etc. The switch is simulated in term of RF performance like
insertion loss, isolation, and return loss at the frequency range 10-32 GHz. The
simulations also consider the pull-down voltage, switching time. A high capacitance
ratio (~180) switch is proposed, which meets the RF performance with reasonable pull-
down voltage.

Paper 4: Shimul Chandra Saha, Trond Sæther, “Modeling and Simulation of Low Pass
Filter using RF MEMS Capacitance and Transmission line”, Proceedings of IMAPS
Nordic conference 2005, pp 155-159, September 2005, Tønsberg, Norway.

This paper describes modeling, design and simulation of the MEMS tunable low-pass
filter at C band. The filter is based on the theory of stepped impedance transmission line
filters. The high impedance transmission line is equivalent to a series inductance. The
low impedance transmission line is equivalent to a shunt capacitor, where we propose to
use a MEMS shunt capacitor. The MEMS capacitors are combined in shunt with a high
impedance transmission line to obtain the low-pass filter. The MEMS shunt capacitor
can be used to tune the cut-off frequency.

Paper 5: Shimul C. Saha, Ulrik Hanke, Trond Sæther, “Modeling, Design and
Simulation of Tunable Band Pass Filter using RF MEMS Capacitance and
Transmission line”, In Proceedings of Microelectronics: Design, Technology, and
Packaging II, Vol. 6035, pp 60350-C1-C11 part of SPIE Symposium on
Microelectronics, MEMS, and Nanotechnology, December 2005, Brisbane, Australia.

This paper describes the work on modeling, design and simulation of the MEMS
tunable band-pass filter at C band. The filter uses a stepped impedance transmission line
low pass filter as discussed in paper 4, with series capacitances to obtain band-pass filter
characteristic. The filter has capability of tuning both the center frequency and the
bandwidth. High capacitance ratio MEMS shunt and series capacitances are proposed
for continuous tuning of the center frequency and the bandwidth. The filter is much
more compact than traditional half wavelength filters. The design is done in ADS and a
comparison is performed with high frequency structure simulator (HFSS) simulation.

27
Paper 6: Shimul Chandra Saha, Ulrik Hanke, Geir Uri Jensen and Trond Sæther,
“Modeling of Spring Constant and Pull-down Voltage of Non uniform RF MEMS
Cantilever”, In proceedings of 2006 IEEE International Behavioral Modeling and
Simulation Conference, pp 56-60, September 2006, San Jose, California, USA.

This paper presents a novel model for the spring constant and the pull-down voltage of
non-uniform cantilevers. The suspended beam is narrower close to the anchor and wider
at the actuation area. This will reduce the spring constant and the pull-down voltage
compared to a uniform beam. A comprehensive model for the spring constant is
developed using the basic force-deflection calculation. The modeled pull-down voltage
is compared with CoventorWare architect simulations and matches very well.

Articles submitted to journal:

Paper 7: Shimul Chandra Saha, Ulrik Hanke, Geir Uri Jensen and Trond Sæther,
“Modeling of Spring Constant and Pull-down Voltage of Non uniform RF MEMS
Cantilever Incorporating Stress Gradient”, has been submitted to Sensors and
Transducers Journal.

This paper is an extension of paper 6, and includes the effect of the stress gradient. The
presence of a stress gradient in cantilever beams is very common. For a positive stress
gradient the cantilever will buckle up and will have a gradual increase in the initial gap
from the actuation electrode. This will increase the pull-down voltage, as it strongly
depends on the initial gap. The model for the pull-down voltage including the stress
gradient is presented and verified with CoventorWare analyzer Finite Element Method
(FEM) simulation.

Paper 8: Shimul Chandra Saha, Ulrik Hanke, Geir Uri Jensen and Trond Sæther,
“Modelling of Spring Constant of a Non uniform RF MEMS Bridge” has been submitted
to International Journal of Modelling and Simulation (ACTA press/IASTED).

28
This paper presents a novel model for the spring constant and the pull-down voltage of
non-uniform bridge. The suspended beam is narrower at the two ends, close to the
anchor and wider at the actuation area. The model includes the residual stress and can
be used for more general cases. A tensile stress will increase the pull-down voltage and
its effect is dominant in the longer and thinner beams. A comprehensive model is
developed using basic force-deflection calculation. The model is verified and agrees
well with CoventorWare analyzer FEM simulations.

3.2: Development of various process steps for fabrication of the RF MEMS switches
and switch circuits
The fabrication of the RF MEMS switches and switch circuits is done in SINTEF MiNa
lab in Oslo, Norway. We had to develop all the process steps by ourselves, as RF
MEMS had not been fabricated at MiNa lab earlier. From the fabrication point of view,
a capacitive switch is easier to fabricate than a DC contact switch since it requires fewer
process steps. Our main goal was to fabricate a workable device and the SINTEF
IRRFT project also recommended capacitive shunt switches. So we planned to fabricate
the capacitive switches, as a first prototype. There are some deviations in the dimension
of the fabricated switches from the initially simulated switches, such as a reduced initial
gap, a shorter beam length and a wider bridge. This is done in order to facilitate a
successful fabrication of the switches and reduce the process complexity. A brief
overview of the fabrication is described below.

A 280 µm thick 4” silicon wafer with 4-8 k:-cm resistivity was used as process wafer.
A 0.5 µm thick oxide (SiO2) was grown on the wafer using wet oxidation. Tungsten was
used for the transmission line. Gold could also have been used, but due to possible
contamination problems we had to use tungsten. A significant amount of stress was
present in the sputtered tungsten (a1 GPa and compressive). This caused the tungsten to
peel off from some of the wafers after sputtering and in some cases during nitride
deposition. We therefore used a titanium adhesive layer to protect the tungsten from
peeling off. To increase the probability of success, some wafers were sputtered with

29
aluminum as transmission lines. Also some wafers were processed without the oxide
layer, as tungsten shows better adhesion to silicon than to the oxide. Plasma enhanced
chemical vapor deposition (PECVD) Si3N4, a thickness of approximately 200 nm was
used as the dielectric layer for the bridge capacitors. The measured thickness was
slightly higher, 220 nm. HiPR 6517 was used as a sacrificial layer for the switch. The
initial gap was 2.5 µm. A special baking method was used to tune the resist angle at the
edge of the rising anchor. Gold DC sputtering was used for the suspended bridge with a
thickness of approximately 1.1 µm. A thickness of 1.2 µm was also used for the second
round of the fabrication. An elaborate experiment was conducted to develop a recipe for
low tensile stress gold. A compressive stress will make the device useless and a high
tensile stress will increase the pull-down voltage significantly. So a low tensile stress is
preferable for low pull-down voltage and a reliable operation of the switch. After gold
patterning, the switch was released by dissolving the sacrificial layer in Microposit 1165
using a method developed as part of the project. Finally the switches were dried in
critical point dryer. A 3-D view of the capacitive switch and an illustration of the
process flow are shown in Figures 3.1 and 3.2.

Figure 3.1 : A 3D view of the proposed fabricated switch

30
31
32
33
Figure 3.2: The cross section side view of the process flow.

34
The fabrication work of the thesis is based on the following published articles.

Paper 9: Shimul Chandra Saha, Håkon Sagberg, Geir Uri Jensen, and Trond Sæther,
“Tuning of Resist slope for RF MEMS with hard baking parameters”, In proceedings of
19th international Microprocess and Nanotechnology Conference (MNC 2006), pp 314-
315, October 2006, Kanagawa, Japan.

This paper presents the works on tuning of the resist slope by hardbaking parameters. A
photoresist is used as a sacrificial layer in the fabrication of the RF MEMS switches. In
order to reduce the number of process steps, a climbing anchor is chosen over a vertical
anchor. In order to obtain a thick enough gold layer at the anchor positions during the
sputtering, the resist angle should be close to 45q. Hardbaking parameters like the
hotplate temperature and the baking time can be used to tune the resist angle. We have
shown that the baking temperature and time can be chosen in an optimum way to tune
the resist angle.

Paper 10: Shimul Chandra Saha, Håkon Sagberg, Erik Poppe, Geir Uri Jensen, Tor A
Fjeldly and Trond Sæther, Tuning of resist slope with hardbaking parameters and
release methods of extra hard photoresist for RF MEMS switches, Journal of Sensor
and Actuators A: Phys. Vol. 142 (2), pp 452-461, 2008.

This paper describes the work on tuning of the resist slope and the release methods for
RF MEMS switch. For a climbing anchor, a slope close to 45q in the sacrificial resist is
preferred. The resist angle can be tuned by the baking parameters. In order to release the
MEMS switches, the sacrificial resist needs to be dissolved completely. It is not easy to
dissolve the sacrificial resist by anordinary acetone dip. We have also used Microposit
1165, which is specially suited for dissolving the hard resist. It is found that some
residue is still left after the Micorposit 1165 dissolution and drying. We have developed
a special method to release the RF MEMS switches successfully without any residue
left. The resist can be dissolved by an initial short plasma strip and then Microposit
1165 with heating.

35
Paper 11: Shimul Chandra Saha, Håkon Sagberg, Erik Poppe, Geir Uri Jensen, and
Trond Sæther, Metallization scheme and release methods for fabrication of RF MEMS
switches, In proceedings of 33rd Micro- and Nano- Engineering conference (MNE
2007), September 2007, pp 773-774, Copenhagen, Denmark.

This paper presents the metallization scheme and release methods for the RF MEMS
switch. Residual stress in the suspended bridge plays a significant role in the reliability
and the pull-down voltage. If a reasonably high compressive stress occurs, the bridge
will be useless. A high tensile stress will significantly increase the pull-down voltage.
We have performed several sputtering experiments in terms of sputtering parameters to
develop a recipe for low tensile stress gold for the suspended beam.

3.3: Measurement of the fabricated RF MEMS switches and switch circuits

We have successfully fabricated the shunt capacitive switches and MEMS tunable
filters. The filters measurement results are submitted and in process of submission to
journal. The measurement work of this thesis is described in the following, submitted
and published papers and memo.

Paper 12: Shimul Chandra Saha, Ulrik Hanke, Håkon Sagberg, Tor A. Fjeldly, Trond
Sæther, Tuneable Low-Pass Filter from C to X band with RF MEMS Capacitance and
Transmission line, has been submitted to Microelectronics Journal.

The paper presents a tuneable low pass-filter. The filter is based on the stepped
impedance transmission line theory as presented in papers 4 and 5. A filter with high
capacitance ratio switch is proposed. For verification a filter with simple parallel plate
capacitors is fabricated. The tuning of the cut-off frequency can be obtained by
actuating different combinations of parallel shunt bridges. A comparison between the
simulations and measurement results is also performed.

36
Paper 13: Shimul Chandra Saha, Ulrik Hanke, Håkon Sagberg, Tor A. Fjeldly, Trond
Sæther, Tunable Band-Pass Filter at C band with RF MEMS Capacitance and
Transmission line, has been submitted to Journal of European Microwave Association
(EuMA).

The paper presents a tuneable band-pass filter. The filter is based on the stepped
impedance transmission line theory as presented in paper 5. The filter has the
possibility of both tuning the centre frequency and the band width. The tuning of the
centre frequency can be obtained by tuning the shunt capacitors. The tuning of
bandwidth can be obtained by tuning the series capacitors. For verification, a filter with
simple parallel plate capacitors is fabricated. The tuning of the centre frequency and
bandwidth can be obtained by actuating different combinations of parallel shunt bridges
and series cantilever. A comparison between simulations and measurement results is
also performed.

M 14: Memo: Measurement results for the capacitive shunt switches from fist round of
fabrication are presented. The pull-down voltage is measured for a range of switches
with different widths and lengths. Some SEM and optical images are included in the
memo. A brief description of reliability, like the number of successful operating cycles,
and the charge stiction, is also presented. An article describing the switch fabricated in
the 2nd round is accepted in MEMSWAVE 2008.

Paper 15: Bengt Holter, Karsten Husby, Håkon Sagberg, Geir Uri Jensen, Ulrik Hanke,
Shimul Chandra Saha, A capacitive shunt RF MEMS switch for coplanar waveguide
transmission lines, In press, MEMSWAVE 2008, 30th June -3rd July 2008, Crete,
Greece.

This paper describes measurement results of the capacitive shunt switch fabricated in
second round, for SINTEF IRRFT project. The process parameters are almost identical
to the first round except the thickness of the substrate and the suspended bridge. The
thickness of the substrate is 380 µm instead of 280 µm used in first round. The

37
thickness of the gold bridge is 1.2 µm. The paper presents the reliability test and life
cycles of the switch. RF performance of the switch like insertions loss and return loss in
upstate and isolation in downstate is also presented.

38
4. Discussion

This thesis presents design and fabrication works on RF MEMS switches and switch
circuits (MEMS tunable filters) for RF front end applications and CMUT switching.
Modeling, design and simulation of various kinds of RF MEMS switches and MEMS
tunable filters are performed. A high speed switch (switching time from upstate to
downstate), is considered for CMUT switching between the transmitter and receiver
block. For CMUT applications, a switch with dimensions 50 µm × 20 µm will be
convenient considering miniaturization and integration. A DC contact switch will be
appropriate for this kind of switching as the operating frequency for CMUT is in the
range of several tens of MHz (30 MHz). For the initial design, switches with larger
dimensions are simulated. The capacitive contact switch is simulated instead of a DC
contact switch. In the simulations, the switching time is considered primarily in terms of
beam material, packaging pressure and environment and beam structures like bridges or
cantilevers. The actuation voltage is also considered, which obviously reduce the
switching time with increasing value. The capacitive contact switch can be converted to
a DC contact switch, by introducing a separate contact point other than the actuation
electrode. This can be done both for bridges and cantilevers. Possible ways of designing
a DC contact switch from a capacitive contact switch are shown in Figures 4.1, 4.2 and
4.3.

Figure 4.1: A side view of a bridge DC contact bridge. A metal to metal contact point is
formed by introducing a gap in the electrical actuation electrodes.
From the figures it can be seen that the switch can be converted easily by modifying the
actuation mechanism slightly and introducing a DC contact point. The contact point can

39
be formed at the middle (bridge) or at the end (cantilever) of the beam. The switching
time will change slightly with actuation voltage due to the rearrangement in the
actuation electrode. The fabrication process for the DC contact switches will be more
complex and will require some more process steps than capacitive contact switches. In
order to design a miniaturized switch with a similar actuation voltage and switching
time, it is possible to use a reduced initial gap, a thinner beam and a non-uniform or
folded spring type suspended beam.

Figure 4.2: The top view of a bridge DC contact switch. The DC contact is formed
between the signal in and the signal out electrodes when the beam moves down.

Figure 4.3: The side view of a cantilever DC contact switch. The DC contact can be
formed by extending the suspended beam beyond the actuation electrode.

40
The spring constant ‘k’ for a typical bridge is mentioned in equation 4.1. In this
equation ‘t’ is the thickness of the beam and ‘L’ is the length of the beam. ‘E’ is the
Young’s modules and ‘w’ is width of the beam. It can be seen that the spring constant
significantly depends on beam length and thickness. The spring constant of the uniform
and shorter beam will increase significantly as it is inversely proportional to the 3rd
power of the length (L3). The pull-down voltage for a typical bridge or cantilever is
given by equation 4.2 and y=1 for uniform beam case. Here ‘g0’ is the initial gap and
‘Le’ is the length of the electrode. The spring constant and the pull-down voltage can be
reduced significantly by reducing the thickness of the beam. Also a reduced initial gap
will reduce the pull-down voltage. A non-uniform beam (narrower at the anchor and
wider at the actuation area) or a folded spring type beam will also reduce the pull-down
voltage.
1
K 32 Ew(t / L) 3 4.1
8( x / L)  20( x / L) 2  14( x / L)  1
3

8kg 03
Vp 4.2
27H 0 Le wy

The folded part of the beam can be simplified to a longer beam, which is narrower than
the actuation electrode. This beam can be modeled as a non-uniform beam. So the
model of the spring constant and the pull down-voltage of the non-uniform beam will be
a very useful tool for predicting the pull-down voltage. The model is verified with well
known commercial tools like CoventorWare and shows that the model is very accurate.
It would be better if we could have verified the model with experimental results.
However, no such data were available.

For the SINTEF IRRFT project front end circuits, a capacitive shunt switch with a high
capacitance ratio is desired. The operating frequency for these circuits is 10 GHz to 32
GHz. A shunt switch with an initial gap of 3.75 µm is designed. Si3N4 is used as a
dielectric with a thickness 100 nm. The beam length for a reasonable pull-down voltage

41
(15-30 V) is 600 µm with a thickness of 1 µm. From the simulation it is found that this
switch can provide a capacitance ratio of 180 which is desirable. A bridge with smaller
length and reduced initial gap is fabricated to increase the probability of success. The
reduced gap is used, to have a reliable anchor at the rising part of the beam. Also a
thicker dielectric is used to reduce the probability of breakdown during electrostatic
actuation.

Tunable low-pass and band-pass filters are designed at C to X band for RF front end
applications. Stepped impedance transmission line theory is used to design the filter.
High capacitance ratio MEMS shunt capacitors are proposed to tune the cut-off and
center frequency. This will facilitate a continuous tuning instead of stepwise tuning. For
fabrication, ordinary parallel plate capacitors are used as tuning element. This is done in
order to reduce the process complexity for the initial prototype and for verification of
the RF characteristic of the filters. Stepped impedance filters are more compact than
traditional half-wavelength band-pass filters.

A significant amount of the time was spent on process development and fabrication of
RF MEMS switches and switch circuits. As SINTEF IRRFT project required a
capacitive contact switch, our initial plan was to fabricate a capacitive contact switch.
From the process point of view, capacitive contact switches require fewer process steps
than DC contact switches. As the process for RF MEMS switches is new in the MiNa
lab, we had to develop all the process steps ourselves. It was not straightforward and all
the process steps had some peculiarity and difficulty to make them work properly.

We had to use tungsten as a transmission line. We could not use gold in the main clean
room because of a potential contamination problem. Although it looks very straight
forward to deposit tungsten on Si or SiO2, it was not that easy. The tungsten has a very
high melting point, requiring a relatively high power to eject atoms from the target. The
residual stress in the deposited tungsten is very high, in the range of 1.0 GPa and
compressive in nature. This high stress makes the wafer buckle significantly. This leads
to peeling of the tungsten from some of the wafers. Also it is found that the peeling is

42
higher in the wafer with oxide than in the bare silicon wafer. In order to avoid the
peeling, we had to use titanium (Ti) as an adhesive layer.

It is also found that even if the tungsten survives after sputtering, it may peel off during
the dry etching. In some wafers, when we deposited liquid phase chemical vapor
deposition (LPCVD) nitride on top of tungsten, most of the tungsten peeled off due to
high temperature. So we had to deposit plasma enhanced chemical vapor deposition
(PECVD) nitride, instead of LPCVD nitride as dielectric. The nitride deposition was
somehow less complicated, except the peeling off the tungsten. The etching of nitride
was rather complicated. The standard recipe for nitride etch contains SF6, which also
etches the tungsten. It is found that the underlying tungsten etches much faster than the
Si3N4. So we had to use a different recipe, with a gas mixture without SF6. In order to
increase the probability of successful fabrication, we also used aluminum in a few
wafers as transmission line. Aluminum is a soft metal and it has the problem of forming
bumps during subsequent high-temperature processes, like nitride deposition. This will
reduce the downstate capacitance significantly.

The tuning of the resist slope for the climbing anchor was an important part of this
fabrication work. Although it was somewhat time consuming, the experiment was quite
straightforward. First we had to choose a suitable exposure power and exposure gap.
Exposure parameters, especially the exposure gap has a significant effect on the resist
slope. It was not very convenient to change the lamp power and exposure gap, so we
chose the standard exposure parameters used by the MiNa lab. We used the hotplate
instead of an oven to tune the resist angle. The hotplate has much faster heating
capability and better controllability. To tune the resist angle we varied the baking
temperature and the baking time. After some experiments and inspection in SEM, it was
found that the resist angle can be tuned in an optimum way using baking temperature
and time.

After the sacrificial layer patterning, gold was sputtered to form the suspended beam. A
significant amount of experiments was performed on gold sputtering, to develop a

43
reliable recipe. A compressive stress will buckle the bridge upward or downward. For a
relatively high amount of compressive stress, the bridge will be useless. A small amount
of tensile stress is preferable for a reliable operation. The higher amount of tensile stress
will significantly increase the pull down voltage.

Initially we used a power of 1500 watts for the DC sputtering of gold. A significant
amount of heat was produced in the chamber which was enough to burn all the
sacrificial resist. So a high power was not possible to use, for our application. We used a
very low DC power (500 watts) for gold sputtering and it was not possible to vary the
power to tune the stress. We had to vary only the sputtering pressure to control the
residual stress in the sputtered gold. We have developed different test structures for
stress characterization. Bridges were used for characterizing the compressive stress,
Guckel rings were used for the tensile stress and cantilevers were used for the stress
gradient. A negative stress gradient was present in all cases. There were slight
uncertainties in the stress characterization due to the stress gradient. Using
CoventorWare and Guckel rings we characterized the stress. The experiment showed, a
tensile stress was present for all the cases and for higher pressure a lower tensile stress
were found. Later a strain gauge was also designed and fabricated, and showed the
presence of tensile stress in the sputtered gold.

After that the gold was patterned in the gold line of MiNa lab. Although all the tools are
old and manual they worked satisfactorily for the process. Gold was etched in potassium
iodide solution. The process was very convenient although we had to check the etching
rate. The release was somewhat more complicated. We used acetone in the beginning to
dissolve the resist. After drying we found that some threads and residue are left on the
substrate. Later we used Microposit 1165, which is specially used to dissolve hard
resist. We kept the samples in the solution for a longer time including overnight with
stirring. After drying we still found some resist residue left. So we had to develop our
own method. We used plasma strips to ash the initial skin, or tough parts of the resist.
Then we used Microposit 1165 with heating and stirring. After drying we found that the
switches were fully clean and no residues were left. For drying, we used a CO2 critical
point dryer. This is a very standard procedure to release RF MEMS switches without

44
stiction. Also I had some experience from my Master’s thesis on RF MEMS release
using critical point dryer, hence this method was very convenient to use for us.

As mentioned earlier, our initial goal was to fabricate a capacitive shunt switch. It was a
priority also, to realize a workable switch rather than to have a switch with precise
specifications. So there were some deviations in the fabricated switches from the switch
proposed in paper 3. The initial gap of the fabricated switch was 2.5 µm, instead of 3.75
µm used in the simulation. It is slightly complicated to obtain a thicker sacrificial layer.
Also a thicker sacrificial may cause a thinner bridge at the rising part of the anchor and
will reduce the reliability of the switch. The length of the bridge is also reduced
compared to the simulated bridge, although a range of different lengths are used for
fabrication. Although a shorter bridge will increase the pull-down voltage significantly,
the reduced initial gap will also reduce the pull-down voltage to some extent. A thicker
dielectric ~220 nm is deposited instead of 100 nm used in simulation. This was done in
order to reduce the probability of breakdown due to the higher actuation voltage. The
tensile stress was reasonably low (~ 30-40 MPa), so the pull-down voltage was not very
high. In fact the pull-down voltage was in the range of 15-20 volts for most of the
switches, even lower than the expected value.

The S parameters for the switches are measured in a vector network analyzer at NTNU.
The insertion loss and reflection coefficient were very good until 32 GHz. The isolation
was somewhat lower than expected. The main reason is the thicker nitride, which will
reduce the downstate capacitance. Also due to planarity and roughness issues, the
capacitance will be somewhat lower than expected. It is also found that the switch has a
stiction problem. A test is conducted with bipolar actuation, and it is found that the
switches operate until 300000-400000 cycles successfully before they stick to the
bottom electrode. For a thicker gold bridge (second round, 1.2 µm) the number of
successful operating cycles was more than 1.000.000 cycles, although a higher pull
down voltage (30 volt) was required for actuation. It is also found that if the pulse width
is reduced, the number of operation cycles increases. This gives us an indication of
charge stiction, which is very common in capacitive contact switches. The test is
conducted in an ordinary clean room, which may increase the probability of stiction due

45
to contamination and vapor. SINTEF is now conducting some more tests to identify and
solve the major issues responsible for the stiction. A test setup in a nitrogen chamber is
planned to make sure that vapor or other contamination may not cause the stiction. Also
an additional anchor on top of the gold bridge is planned for later fabrication, to reduce
the probability of stiction during actuation.

The tunable filters were fabricated with the same process parameters as the capacitive
shunt switches. A number of parallel bridges are used for shunt capacitors. Also a
number of parallel cantilevers were used for series capacitors. The change in the shunt
capacitance was obtained by actuation different numbers of bridges. As a negative stress
gradient was present in the cantilever, different sets of cantilevers were used for
different designs. If the cantilever is already downstate due to the stress gradient, it will
be not possible to change the capacitance by actuating different cantilevers. The first
round of measurements was very interesting and gave a proper validation of the
simulation results. The measured filters were with an aluminum transmission line.
Although the downstate capacitance is somewhat smaller due to bumps in the metal,
still very good results were obtained. We also got some measurement result with
tungsten transmission lines. The wafers with the tungsten transmission lines suffered
significant over-etch in the suspended beam and the capacitance value was reduced
drastically. The cut-off and center frequency shifted to some extent (2-3 times than
expected). The measurement results of the filter with aluminum transmission lines
showed very good match with simulation results considering the fabrication issues.

46
5. Conclusions

In this work, novel RF MEMS tunable low pass and band pass filters are designed and
fabricated. A new model for non uniform bridge and cantilever is also presented. We
have developed the first RF MEMS switch ever to be produced in Norway, making it
pioneering work in Norway. The switch reliability and RF performances are very
promising compared to the present state of art work published recently else where.
Some other research groups at different universities and research organization already
started working on RF MEMS in Norway. This work will give a valuable input for the
RF MEMS research community in Norway including NTNU and SINTEF and
international community.

This thesis presents design and fabrication of RF MEMS switches and MEMS tunable
filters. Modeling and simulation of high-speed switching and comparison between
bridge and cantilevers is performed. A high capacitance ratio shunt capacitive switch is
also designed and simulated. A switch is proposed for the desired specifications with
reasonable pull-down voltage. Design and simulation of tunable low-pass and band-pass
filters are also performed. The tuning of cut-off and center frequency and also
bandwidth can be obtained by actuation of different combinations of shunt and series
capacitances. A significant amount of work is performed on the process development.
As the process for RF MEMS switch is new in the MiNa lab, we had to develop all the
process steps by ourselves. The metallization of the transmission line and the suspended
beam required special efforts. After several experiments and evaluations, we developed
a successful method for tungsten sputtering for the transmission line. We also proposed
a recipe for low-stress planar gold for the suspended beam. A special method to release
the RF MEMS switches is also developed. The fabricated shunt switch has slightly
different dimension, specially reduced gap and thicker dielectric, than the simulated
switch. The S parameters of the switches are measured at NTNU using a vector network
analyzer. The measured results of the first prototype look very promising. The pull-
down voltage is also within a reasonable limit around 15-20 volts for most of the
switches. The pull down voltage for second version of the switch (1.2 µm thick
suspended beam) is around 30 volts. The filters are also measured in the vector network

47
analyzer. Although the cut-off and center frequency shifted slightly toward higher
frequencies, the measured result shows very good agreement with theory. The
comparison of measurements results with simulations, considering the process
constraints, agrees quite well.

48
Correction lists

1) Considering the scope and editorial board advice Paper no 7 is submitted to


Sensors and Transducers Journal
2) Considering the scope and editorial board advice Paper no 8 is submitted to
International journal of modeling and simulation (ACTA press).
3) The Final version of Paper no 15 (4 pages) is modified and updated according to
conference review committee.

49
DESIGN AND SIMULATION OF RF MEMS SWITCHES
FOR HIGH SWITCHING SPEED AND MODERATE
VOLTAGE OPERATION

Shimul Chandra Saha, Tajeshwar Singh, Trond Sæther

Department of Electronics and Telecommunications, Norwegian University of Science and


Technology (NTNU), 7491 Trondheim, Norway
E-mail: [email protected]

In proceedings of IEEE PRIME 2005, pp 207-210, July 2005, EPFL, Lausanne,


Switzerland.
DESIGN AND SIMULATION OF RF MEMS
SWITCHES FOR HIGH SWITCHING SPEED AND
MODERATE VOLTAGE OPERATION
Shimul Chandra Saha, Tajeshwar Singh, Trond Sæther

Department of Electronics and Telecommunications, Norwegian University of Science and


Technology (NTNU), 7491 Trondheim, Norway
E-mail: [email protected]

ABSTRACT imaging application of the CMUTs is in the


detection of plaque inside the coronary for an early
In this paper we present the design and simulations heart disease diagnosis.
of an RF MEMS switch with regard to its switching
speed and pull-down voltage. The switch is required The active part of a CMUT is usually a metal coated
to fulfill the requirements of high speed and a Si3N4 which forms the top membrane/electrode. A
moderate pull down voltage for the switching of highly doped silicon substrate constitutes the bottom
CMUT (Capacitive Micromachined Ultrasonic electrode. When a DC bias voltage is applied
Transducer) elements used for ultrasound imaging between the top and bottom electrodes, the
membrane is pulled towards the substrate
1. INTRODUCTION (electrostatic actuation). If an AC voltage is applied
along the bias point, a harmonic motion is obtained;
The objective of this paper is to present the design of thus the membrane will generate AC detection
a high speed RF MEMS switch. MEMS are currents. As in piezoelectric ultrasound probes, there
nowadays becoming increasingly popular in RF are two modes of operation namely, transmit, and
applications due to their attractive advantages such receive. CMUT elements require to be switched ON
as high isolation and low insertion loss. Depending and OFF at different times either to send the
on the switch type (DC contact or capacitive), the actuation signal, or to receive the reflected signal.
MEMS switches can show very good RF Hence it is necessary to have a switching matrix to
performance from DC to several tens of GHz. switch the elements from transmit mode to receive
However, there are two sides for this coin too; there mode. As the actuation signal is low, it is necessary
are also some disadvantages of the RF MEMS to have a low insertion-loss and very good isolation
switches, as compared to their solid-state of the switching matrix, for which an RF MEMS
counterparts, which include high pull-down voltage switch is a good candidate. However as the vein
and slow switching speed. In this paper it is shown radius is small, the time of flight of the signal to
that by a careful design, a compact switch with a travel to the vein wall and back to the catheter
moderate applied voltage can be optimized for the surface is very short. Thus it is necessary to design
switching speed (<5μs). the switch with very short switching time. For our
design we target for a switching time, (to switch
2. MOTIVATION from the up-state to down-state) <5μs which is 3-6
times lower compared to 15-30μs shown by the
CMUTs are an upcoming technology in the field of present designs, for example [2][3] operating at
Medical Imaging which promises to offer advanced similar actuation voltages (15-25V) .
automation possibilities. It will also offer, unlike the
conventional piezoelectric transducer probes, 3. DESIGN AND SIMULATION
compatibility with the CMOS process and
integration of the signal processing circuitry on the We consider a very simple RF MEMS capacitive
probe itself, thus drastically reducing the hard- switch for our design. The structure is a Fixed-Fixed
wiring of the probe to the signal processing console. beam using electrostatic actuation. The beam
The possibility of signal processing near the material is chosen to be Aluminum because it has a
transducers opens new prospects of ‘Smart probes’ moderate Young’s modulus which favors lower
capable of handling large amount of data, required actuation voltage at a smaller switch size. Selecting
for example, for 3D ultrasound imaging. Another a metal membrane also means simple fabrication
process. Preliminary simulations for switching time
& pull-down voltage were done with Mathematica®.
The pull-down voltage V p and switching time ts of
the bridge are given by [1][3]:
8 ⋅ k ⋅ g3
Vp = (1)
27 ⋅ ε 0 ⋅ W ⋅ w
Vp 27
ts = (2)
ω 0 ⋅vS 2 Figure 2: Switching time simulations for Al beam with
10μs pulse delay and Q≈1

Figure 1: The capacitive bridge

The actuation voltage Vs is selected slightly higher


than the pull-down voltage calculated in (1), to
increase the switching speed. The proposed switch is
shown in Fig. 1. Figure 3: Pull down voltages for Al and Au beams.

The dimensions of the beam 3.1 The damping and the Quality factor
are 250 μ m × 40 μ m × 1.5 μ m , the bottom electrode
The damping (dominantly squeeze film damping)
is 200 μ m × 40 μ m , and initial gap g 0 = 1μ m . From the plays an important role in the switching time of
simulations in Mathematica®, a pull-down voltage of MEMS switches (Equation 3). Within small gap
12V is obtained. Hence, with an actuation voltage of heights, as in MEMS devices, the gaseous medium
15V, which is of the order of the present state-of- acts like a viscous liquid. During switch actuation
the-art (10-15V), pull down time of ts ≈ 2.9 μ s is when the beam moves down, the gaseous medium
obtained, which is around 10 times lower than present within the gap must be pushed out and the
reported results [2]. Final device simulations were molecules undergo several collisions between the
done in Coventorware®. The switching time and beam and electrode [3]. The viscosity of the medium
actuation voltage is shown on the Table 1. For and in turn the quality factor can be controlled by
comparison, the Gold beam is also simulated and as reducing the pressure of the switch medium through
we can clearly see that the switching time is more packaging. The equations (4-7) for effective
than 2.5 orders better for Al than for Au for the same viscosity are shown below [3].
actuation voltage. The switching time (transient) and d 2z dz
m +b + kz = F total (3)
pull down voltage simulations are shown in Figure 2
dt 2 dt
and Figure 3 respectively.
p0
Table 1: The switching time for the switch with Q≈1 λa = λ0 (4)
pa
Material Pull down Actuation Switching
voltage voltage(V) time (μs)
λ
(V)
Kn = (5)
Aluminum 11 12 4.90 g
Aluminum 11 15 2.95
Aluminum 11 20 1.85 μ (6)
μe =
Aluminum 11 25 1.40 1 + 9 . 638 K 1 . 159
n

Gold 12 15 7.55 Where λa is the mean free path at pressure pa . g is


the gap height and K n is the Knudsen number. K n is
a measure of viscosity of the gas under MEMS
beam. μe is the effective viscosity, a function of the at very low Q (high viscosity), the switching time is
Knudsen number. A high Knudsen number means very large. Note from Figure 4 that very high Q also
low effective viscosity and the gas experiences few leads to a higher settling (release) time.
collisions and the flow is not viscous anymore. So
by reducing the pressure of the medium (vacuum)
the viscosity can be reduced and the switching speed
increased.
The effective damping coefficient can also be
reduced by having perforations in the moving plate.
The equations for damping without, and with
perforations on the beam are given in Equation (7)
and Equation (8) respectively [3, 4].
3 μA 2 (7)
b =
2 π g 03 Figure 4: The switching and release time for Va =15 V for
different values of Q.
12 μ A 2 p p2 ln( p ) 3 (8)
bh = ( − − − )
Nπ g0 3
2 8 4 8

Here, N is the number of holes; p is the fraction of


open area in the plate. From Equation (8) it can be
seen that by having perforations in the membrane,
the damping coefficient can be reduced significantly.
This option widely used because even by having the
perforations, the upstate capacitance is not affected
because the fringing fields ‘fill’ the empty space.
The beam was simulated with different values of Figure 5: The switching time for actuation voltage 25 V
for different Q.
Quality Factor ‘Q’ (Equation 9) which depends on
damping i.e. mainly gas damping and modal At a higher Q the switching (pull down) time improves but
damping ‘D’ (D=a⋅M+b⋅K, where ‘M’ is the mass of increasing Q beyond 1 or 2 does not reduce the switching
the beam and ‘K’ is the beam spring constant, ‘a’ time much as compared to the reducing medium viscosity.
and ‘b’ depend on the damping medium and the So Q≈1-2 may be a fair choice for high speed switching. It
beam structure). Q was varied from 0.05 to 10 and can also be seen that at substantially higher Q≈10 and
for different actuation voltage. actuation voltage of 25V, the switching time is 1.25μs,
which is very fast compared to the present technology [2,
k (9) 3].
Q =
ω 0b
3.2 The residual stress consideration
Thus, Q can be easily controlled by adjusting the
damping coefficient which in turn decides the One of the main concerns in the Fixed-Fixed beams
switching speed. As the gap height is low, the is the residual stress. It can result in changes in the
obtained Q is very low ≈0.05 under normal pull down voltage and in extreme cases render the
condition (atmospheric pressure, no perforations). structure useless. The residual stress usually arises
form the process conditions and can be minimised
Figure 9 shows the 3-D view of a beam with 80 by controlling the process parameters. It is possible
holes (4×4 μm2) for which Q≈1 was easily obtained to see the effect of residual stress on the switching
at normal pressure. The same value of Q can be time. When residual stress increases, it adds to the
obtained without perforations by reducing the spring constant, thus the resonant frequency
pressure equal to ≈1/30th of the atmospheric increases (ω0=k/m) which reduces the switching
pressure. For further increasing the Q, both options time. Figure 6 shows the comparative study of
can be combined (perforations and low pressure), a resonant frequency with residual stress for our beam.
Q≈5 was thus obtained with perforations and at 1/7th
atmospheric pressure. Figures 4 and 5 show the As seen, when the residual stress increases, the
switching time for actuation voltages 15V and 25V resonant frequency increases, which reduces the
respectively, for different Q. It can thus be seen that switching time. Thus care must be taken that the
residual stress does not exceed the limits because if frequency. The isolation can be improved by using
the pull down voltage exceeds the desired actuation high dielectric constant materials as shown below.
voltage then the beam cannot be switched.
1,10E+06
1,05E+06
1,00E+06
Resonant frequency

9,50E+05
9,00E+05
8,50E+05
8,00E+05
7,50E+05
7,00E+05
6,50E+05
6,00E+05
0,00E+ 5,00E+ 1,00E+ 1,50E+ 2,00E+ 2,50E+ 3,00E+ 3,50E+ Figure 8: The isolation and the insertion loss in dB.
00 06 07 07 07 07 07 07
Re s idu al s tr e s s

4. CONCLUSIONS
Figure 6: The resonant frequency vs. residual stress.
We have shown the design of an RF MEMS switch
3.3 Different beam lengths with a high switching speed at moderate actuation
voltage. Switching speed increases with actuation
The effect of varying the beam length is
voltage but we have a limitation on the maximum
investigated. For longer beams, the spring constant k
actuation voltage. Damping affects the switching
decreases thus the resonant frequency and the
speed and can be controlled by perforations on the
actuation voltage are both reduced. This has a mixed
beam and by regulating the medium and pressure.
effect on the switching time. Figure 7 shows the
Detailed effect of Q and damping was shown in
switching times for different beam length.
section 3.1. The switches will be fabricated in the
SINTEF, MiNaLab, Oslo.

Figure 7: The switching time for actuation voltage 20 V


for different length and Q≈1. Figure 9: A 3-D view of the beam with perforation.
We choose an actuation voltage of 20V as the pull
down voltage for the 200μm switch is more >15V. 5. ACKNOWLEDGEMENT
From Figure 7 it can be seen that from 200μm to Authors are grateful to Norwegian Research Council
250μm there is a considerable improvement in the for sponsoring the work through SMiDA project (No
switching time but from 250μm to 280μm the 159559/130).
switching time does not improve much. So it is not
worth increasing the beam length further, to 6. REFERENCES
maintain the compactness.
[1] J.B. Mauldavin, Nonlinear Electro-Mechanical
3.4 The S-parameters Modeling of MEMS Switches, IEEE MTT-S Digest,
2001.
The simulation for the S- parameters is shown in [2] D.Peroulis, Electromechanical Considerations in
Figure 8. The switch transmits the signal (ON state) Developing Low-Voltage RF MEMS Switches. IEEE
Transactions on MTT, Vol.51, No. 1 January 2003.
while in upstate and the signal is capacitively
[3] G.M. Rebeiz, RF MEMS Theory, Design and
grounded (OFF state) when the beam is in the Technology, 1st ed. New York: Wiley, 2003, Ch 3,5.
downstate. It can be seen that we get an insertion [4] J. Bergqvist, A Silicon Condenser Microphone with a
loss less than 1dB in upstate position and have more highly perforated backplate, In international
than 15 dB isolation in the down state position for conference on Solid-State Sensor and Actuators
typical operating frequency range (>10 GHz) for a Digest, New York, 1991, pp 266-269.
capacitive switch and more than 25 dB at higher
Design and simulation of RF MEMS cantilever and bridge switches
for high switching speed and low voltage operation and their
comparison.

Shimul Chandra Saha, Tajeshwar Singh, Trond Sæther

Department of Electronics and Telecommunications, Norwegian University of Science and


Technology (NTNU), N-7491 Trondheim, Norway.
[email protected]

In proceedings of IEEE ISSCS 2005, pp 131-134, July 2005, Iasi, Romania


MODELING AND SIMULATION OF HIGH CAPACITANCE
RATIO RF MEMS SHUNT SWITCH

Shimul C. Saha1, Ulrik Hanke2, Geir U. Jensen3, Anton M. Bøifot4, Tor A. Fjeldly1, Trond
Sæther1.

1: Department of Electronics and Telecommunications, Norwegian University of Science and


technology, 7491 Trondheim, Norway.
2: SINTEF ICT, NO-7465 Trondheim, Norway
3: SINTEF ICT, Gaustadalleen 23, 0371 Oslo, Norway
4: MicroConsult, Hallsetreina 10, 7027 Trondheim, Norway.
E-mail: [email protected], [email protected], [email protected], [email protected]
[email protected], [email protected]

In proceedings of third IASTED international conference on Circuits, Signals and


Systems CSS 2005, pp 186-193, October 2005, Marina Del Rey, CA, USA.
Is not included due to copyright
Modeling and Simulation of Low Pass Filter using RF MEMS
Capacitance and Transmission line

Shimul Chandra Saha, Trond Sæther

Department of Electronics and Telecommunications, Norwegian University of Science and


Technology (NTNU), 7491 Trondheim, Norway
E-mail: [email protected]

Proceedings of IMAPS Nordic conference 2005, pp 155-159, September 2005,


Tønsberg, Norway.
Modeling and Simulation of Low Pass Filter using RF MEMS
Capacitance and Transmission line
Shimul Chandra Saha, Trond Sæther

Department of Electronics and Telecommunications, Norwegian University of Science and Technology


(NTNU), 7491 Trondheim, Norway
E-mail: [email protected]

Abstract

Radio Frequency Micro electromechanical systems (RF MEMS) technology offers an attractive
capability for RF systems, particularly in support of switching and tuning functions. One such
component is a micro electromechanical voltage tunable capacitor, which can enable wide tuning
range and high quality (Q) factors. In this paper we present a design of 5.9 GHz 3rd and 5th order low
pass filter using the theory of stepped impedance transmission line filter. In a traditional lumped
element low pass filter the series element is an inductor and the shunt element is a capacitor. In the
stepped impedance transmission line filter, the high impedance transmission line is used to replace
the inductor; the capacitance is replaced by low impedance line. This topology does not give the
flexibility of tuning the cut off frequency. So we replace the low impedance line or the shunt
capacitance by RF MEMS capacitance which will provide the tuning capability.

In this paper we have shown the design of a low actuation voltage RF MEMS switch and modeling
of the switch as variable capacitance. The capacitance can be varied as a function of the actuation
voltage. The capacitance is used as a lumped component in a conventional low-pass LCR filter. The
design uses a novel approach to implement both distributed transmission line and RF MEMS
capacitance together to replace the lumped elements with having tuning capability of the cut-off
frequency.

Key words: RF MEMS Capacitance, Low pass filter, Stepped impedance filter, Tuning.

Introduction compared to the distributed elements. This is a


plus point for filter stop band characteristic and
Filters are an essential part of any high they are attractive for tuning and switching
frequency circuit. It plays an important role in action [2].
signal processing like reducing intermodulation
product, removing image frequency. It is also
an essential part for channel preselect block. Theory of Stepped-impedance Filter
Conventionally low pass filter are
lumped element at low frequency. It is easier to According to transmission line theory
produce, and the theory is straight forward to [3] the Z parameters of a transmission line of
develop. So it can be produced at low cost. But length " and characteristic impedance of Z0
it has some disadvantage like it does not have can be represented as shown if figure 1.
automatic frequency tuning capability without
switching different elements. For high frequency
usually lumped elements are bigger so they are
not suitable for miniaturization. At higher
frequency usually distributed element is used
for filter application. But for distributed element
filter it is difficult to tune the cut-off frequency.
In this paper we develop a filter which uses
both the transmission line topology and RF Figure 1: T equivalent circuit for a transmission line.
MEMS capacitance [1]. The RF MEMS
capacitance can be tuned by actuation voltage The element of the T section in the
so the cut-off frequency can be changed equivalent circuit for a transmission line is given
accordingly. This will give both miniaturization by following equation [3].
and frequency tuning capability for the filter.
The RF MEMS capacitance has high Q
X E" During upstate the capacitance is in the range
Z 0 tan( ) of fF depending on the width of the beam.
2 2 When a DC voltage is applied between beam
1 and bottom electrode the gap will change and
B sin E" the capacitance also changes accordingly. So
Z0
by changing the actuation voltage the
If the length of the transmission line is short and capacitance can be changed which can tune
the characteristic impedance is higher the the cut-off frequency of the filter. When the
equivalent circuit can be reduced to as shown voltage is equal to or greater than the pull down
in figure 2 and the parameter is given by voltage the beam collapses to the downstate
XL Z 0 E" position. The capacitance in the down state
B|0 position is in the range of pF. This capacitance
can be used as high value lump element
XL Z 0 E" capacitance. The dimension of the propose
switch is, LB=500 Pm, g0=1.5 Pm, tB=1.5Pm and
B B

wb=50Pm/100Pm. The electrode length le and


the electrode width we are varied depending on
Figure 2: Equivalent circuit for small length of the requirement of capacitance and
transmission line and high characteristic impedance transmission line parameter. A 0.1 Pm Si3N4 is
used as dielectric. The chosen beam material is
So a high characteristic impedance and short Al but Au can be a good option. The
length of transmission line can be represented transmission line is made of Gold. A
by series inductor. Although it is preferred to comparative study for pull down voltage and the
have very high impedance and short line length variation of capacitance with beam width
(E " <S/4) we choose medium impedance too 100Pm and le=224Pm and we=400Pm is shown
avoid narrowing. For narrow line width it will be below in figure 5. As the electrode (beam) width
difficult to obtain the desired MEMS become more, two beams can be used parallel
capacitance. So the line was chosen bit longer. for convenience.
In the lumped elements filter model, the shunt
capacitor will be replaced by RF MEMS
capacitance which will give the tuning
capability.

Capacitance model of MEMS switch

A low voltage RF MEMS capacitive


shunt switch is designed for the tuning
capacitor. The side view and top view are Figure 5: Pull down voltage and capacitance
shown in figure 3 and 4. The bridge can be simulation in Coventor.
modeled as a capacitance between the bottom Form the figure above it can be seen
electrode and the beam. The capacitance that the pull down voltage is very low 3.6 volt. It
modeling and simulation is done in may increase due to residual stress. The
Coventorware®. residual stress can be controlled by the process
parameter. The upstate capacitance varies from
524 fF to 725 fF before the beam collapse to
downstate. This is almost 40% more than that
of the zero bias voltage. So we can have a
large tuning range capacitance.

A third order filter design and simulation

Figure 3: The side view of RF MEMS bridge A third order low pass filter design is
shown in figure 6. The basic filter element is
extracted from the maximally flat low pass
butterworth filter prototype [3].

Figure 4: The top view of RF MEMS bridge

Figure 6: A third order low pass filter topology


The basic filter elements are given by The filter characteristic with variable
g1=1.00, g2=2.00, g3=1.00 and g4=1.00 is the capacitance is shown in figure 9. From the
load impedance. Using the maximum figure it can be seen that the cut-off frequency
transmission line impedance of 90: the length can be varied from 5.55 GHz to 6.4 GHz for
of the transmission line becomes 63° and two actuation voltage from 3.5V (725 fF) to 0V (525
nominal capacitances value is 550 fF. For fF). So the filter gives a good flexibility of tuning
transmission line we used coplanar waveguide. cut-off frequency with bias voltage.
This makes it easier to have the ground on the
same plane. The MEMS beam will easily form
the capacitance between center conductor and
6.4 GHz
ground plane. The filter topology is shown with
transmission line and MEMS capacitance in 5.55 GHz
figure 7 below. The substrate parameters used
for simulation are substrate thickness of 500
um, with relative permeability of 10 and
transmission line thickness of 1 um. For 90:
transmission line the dimension is
200Pm/55Pm/200Pm (G/W/G) and the length is
3.5 mm at frequency 5.9 GHz. The 50: line
dimension is 100Pm /224Pm/ 100Pm. If the Figure 9: The cut-off frequency variation for the
higher impedance is more than 100: then the MEMS filter.
signal line will be very narrow and it will be
difficult to obtain the required variable upstate A fifth order filter design and simulation
capacitance. The required beam dimension is
equal to the MEMS switch model described A fifth order low pass filter design is
earlier in the capacitance model of MEMS shown in figure 10. The basic filter element is
switch section. extracted from the maximally flat low pass
butterwoth filter prototype [3].

Figure 10: A third order low pass filter topology

The basic filter element is given by


g1=0.6180, g2=1.6180, g3=2.00, g4=1.6180,
Figure 7: A 3rd order low pass filter with transmission g5=0.6180. Using the maximum transmission
line MEMS variable capacitor line impedance of 90: the length of the both
transmission line becomes 51.5° (3.14mm) and
The RF simulation is done in ADS® from the nominal capacitance value is obtained
Agilent with the parameter mentioned using the C1=C5=365fF and C3=1.06 pF. The C3 is used
capacitance model from Coventor®. The as down state capacitance and C1 and C5 is
simulation result for the designed low pass filter used as upstate capacitance. For transmission
with a comparison of filter performance with line it was used the same topology as 3rd order.
lumped elements is shown in figure 8. The filter topology with transmission line is
shown in figure 11 below. The substrate
parameters are same as 3rd order described
MEMS earlier.

Lumped
element

th
Figure 11: A 5 order low pass filter with
Figure 8: The MEMS filter and Lumped element filter transmission line MEMS variable capacitor
performance.
The frequency tuning behavior is
shown in figure 12 below. The C1 and C5 are
varied from 285fF to 395fF. The cutoff
frequency varies from 5.77 GHz to 5.95 GHz.
The tuning range is not so high. Because the
dominant capacitor is C3 and it can control the
frequency more efficiently. The dimension of
the electrode and beam width is given in the
table 1 below keeping other parameters as the
beam parameters mentioned earlier. th
Figure 13: A 5 order low pass filter with
transmission line MEMS variable capacitor
Table 1: The beam parameters with other
parameters as mentioned earlier
The dimensions of one of the beam are
Capacitance Wb(Pm) we(Pm) le(Pm) state
le=100 Pm, we=360 Pm and wb=100 Pm with
C3 25 28 55 Down
C1/C5 50 200 224 Up other parameter as mentioned earlier. The
capacitance is obtained from combination of 4
equal upstate capacitances. The upstate
capacitance variation of one beam is shown in
figure 14 with actuation voltage. The pull down
voltage may be bit higher due to residual stress.

Figure 12: The cut-off frequency variation for the


MEMS filter.

Fifth order filter design with high tuning


range
Figure 14: A capacitance simulation of one beam.
From the above figure 12 it can be
The frequency variation of the designed filter
seen that the cut-of frequency does not vary
with capacitance is shown in figure 14. The C1
much with tuning voltage. So in order to have
and C5 are kept constant and their dimensions
high tuning range the 3rd element of the 5th
are same as mentioned in previous section.
order filter or C3 have to be tuned. And we need
From the figure it can be seen that the cut-off
to design the C3 as an upstate capacitance.
frequency varies form 5.55 GHz to 6.51 GHz at
The transmission line width of the high
actuation voltage 4.4 V to 0 volt. So it gives
impedance becomes very narrow. So we are
more tuning range than 3rd order filter.
going to choose a bit lower impedance 75:.
For lower impedance the width will be higher
and it will be easier to obtain the desired
upstate capacitance. The transmission line will 6.51 GHz
be bit longer compare to the higher impedance.
The lumped element topology is same as figure 5.55 GHz
10 and the transmission line topology is shown
in figure 13. Here four parallel beams are used
as the beam width is too big. It can also be
used several no of parallel beams. For 75:
transmission line the dimension is
185Pm/100Pm/185Pm (G/W/G) and the length
is 3.8 mm (optimized) at frequency of 5.9 GHz.
Figure 15: The cut-off frequency variation of the
MEMS filter with C3 variation
For comparison the simulation is also done with also gives tuning capability. The tuning voltage
one capacitance altogether. The comparison is comparatively very small from 0 to 5V in most
between 4 parallel beams and one beam cases and always less than 10 V. Due to
capacitance is shown in figure 15. From the residual stress the pull down voltage can be
figure 15 it can be seen that when one beam is increased to some extent. For higher upstate
used the frequency is much sharper. So it is capacitance the initial gap height can be
preferable. And it is necessary to consider the reduced, this will also reduce the pull down
process point of view. So if it is possible to voltage. In the simulation we did not include the
fabricate wider beam then it is preferable. inductance from RF MEMS switch. Because
below 20 GHz the inductances do not play
much role so they can be ignored. The
inductances are in the range of tens of pH.
4 parallel beam These can be made negligible having small gap
combination between signal and ground line and also
increasing the beam width. A 3-D view of
MEMS capacitance is shown in figure 16.
One beam
capacitance Acknowledgment

Authors are grateful to Norwegian


Research Council for sponsoring the work
through Smart Microsystems for Diagnostic
Figure 15: The magnitude response for one beam Imaging in Medicine (SMiDA) project (No
and 4 beam parallel 159559/130).
Conclusion
References
We have shown a design of tunable
and miniature MEMS low pass filter. The use of >1@ A. Abbaspour-Tamijani, A high performance
transmission line at higher frequency is easier MEMS miniature tunable band-pass filter, IEEE
and favorable as the dimensions become MTT-S Digest, 2003
smaller. >2@ R. L. Borwick, Variable MEMS capacitors
For a distributed filter the Q factor is usually implemented into RF filter systems, IEEE
lower than for lumped element. In this filter we Transaction on MTT, Vol.51, No.1,01/03
use both distributed element and MEMS [3] David M. Pozar, Microwave Engineering, Ch
capacitance which has high Q. So we can have 8, Second Edition, John Willey & Sons, 1998.
both advantages. The miniature MEMS filter

Figure 16: 3-D view of one of the proposed MEMS capacitance in CPW configuration.
Modeling, Design and Simulation of Tunable Band Pass Filter using
RF MEMS Capacitance and Transmission line

Shimul Chandra Saha *a, Ulrik Hanke b, Trond Sæther a


a
Department of Electronics and Telecommunications, Norwegian University of Science and
Technology (NTNU), 7491 Trondheim, Norway
b
SINTEF ICT, NO-7465 Trondheim, Norway

In Proceedings of Microelectronics: Design, Technology, and Packaging II, Vol. 6035, pp


60350-C1-C11 part of SPIE Symposium on Microelectronics, MEMS, and
Nanotechnology, December 2005, Brisbane, Australia.
Is not included due to copyright
Modeling of Spring Constant and Pull-down Voltage of Non uniform
RF MEMS Cantilever

Shimul Chandra Saha1, Ulrik Hanke2, Geir Uri Jensen3, Trond Sæther1

1
Department of Electronics and Telecommunications (IET), Norwegian University of Science and
Technology (NTNU), 7491 Trondheim, Norway
2
Institute of Microsystems Technology, Vestfold University College, 3103 Tønsberg, Norway
3
SINTEF ICT, Gaustadalleen 23, 0371 Oslo, Norway

In proceedings of 2006 IEEE International Behavioral Modeling and Simulation


Conference, pp 56-60, September 2006, San Jose, California, USA
Modeling of Spring Constant and Pull-down Voltage of Non-uniform
RF MEMS Cantilever Incorporating Stress Gradient

Shimul Chandra Saha1, Ulrik Hanke2, Geir Uri Jensen3, Trond Sæther1

1
Department of Electronics and Telecommunications (IET), Norwegian University of Science and
Technology (NTNU), 7491 Trondheim, Norway
2
Institute of Microsystems Technology, Vestfold University College, 3103 Tønsberg, Norway
3
SINTEF ICT, Gaustadalleen 23, 0371 Oslo, Norway

Has been submitted to Sensors and Transducers Journal.


Is not included due to copyright
MODELLING OF SPRING CONSTANT OF A NON-UNIFORM RF
MEMS BRIDGE

S. C. Sahaa, U. Hankeb, G. U. Jensenc, T. Sæthera

a
Department of Electronics and Telecommunications (IET), Norwegian University of Science and
Technology (NTNU), 7491 Trondheim, Norway. E-mail: [email protected],
[email protected]
b
Institute of Microsystems Technology, Vestfold University College, 3103 Tønsberg, Norway. E-
mail: [email protected]
c
SINTEF ICT, Gaustadalleen 23, 0371 Oslo, Norway. E-mail: [email protected]

Has been submitted to International Journal of Modelling and Simulation (ACTA


press/IASTED)
Is not included due to copyright
Tuning of resist slope for RF MEMS with hardbaking parameters

Shimul Chandra Saha 1, Håkon Sagberg 2, Geir Uri Jensen 2, and Trond Sæther 1

1
Department of Electronics and Telecommunications, Norwegian University of Science and
Technology (NTNU), NO-7491 Trondheim, Norway.
2
Department of Microsystems and Nanotechnology, SINTEF ICT, PO Box 124 Blindern, NO-0314
Oslo, Norway

In proceedings of 19th international Microprocess and Nanotechnology Conference


(MNC 2006), pp 314-315, October 2006, Kanagawa, Japan.
Is not included due to copyright
Tuning of resist slope with hard-baking parameters and release
methods of extra hard photoresist for RF MEMS switches

Shimul Chandra Saha*1, Håkon Sagberg2, Erik Poppe2, Geir Uri Jensen2, Tor A. Fjeldly1and
Trond Sæther1
1
Department of Electronics and Telecommunications (IET), Norwegian University of Science and
Technology (NTNU), NO-7491 Trondheim, Norway
2
Department of Microsystems and Nanotechnology, SINTEF ICT, PO Box 124 Blindern, NO-0314
Oslo, Norway

Journal of Sensor and Actuators A: Phys. Vol. 142 (2), pp 452-461, 2008.

This article is reprinted with kind permission from Elsevier, sciencedirect.com


This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy

Available online at www.sciencedirect.com

Sensors and Actuators A 143 (2008) 452–461

Tuning of resist slope with hard-baking parameters and release


methods of extra hard photoresist for RF MEMS switches
Shimul Chandra Saha a,∗ , Håkon Sagberg b , Erik Poppe b , Geir Uri Jensen b ,
Tor A. Fjeldly a , Trond Sæther a
a Department of Electronics and Telecommunications (IET), Norwegian University of Science and Technology (NTNU),
NO-7491 Trondheim, Norway
b Department of Microsystems and Nanotechnology, SINTEF ICT, P.O. Box 124 Blindern, NO-0314 Oslo, Norway

Received 6 July 2007; received in revised form 12 October 2007; accepted 21 October 2007
Available online 1 November 2007

Abstract
We have studied the properties of photoresist (resist) as a sacrificial material for fabricating suspended metal bridges and developed a process to
release RF MEMS switches by dissolving thermal cross-linked extra hard photoresist. At the edges of the patterned photoresist, the contact angle
with the substrate can be tuned using hard-baking parameters. From the experiment we have found that the resist angle reduces both with baking
temperature and time. The baking temperature has a stronger effect on resist slope than the baking time has. It is also observed that at a certain
baking temperature, and after a certain time, the resist angle does not reduce further. The RF MEMS switch with photoresist as the sacrificial
layer can be released by plasma ashing to strip the initial resist skin, followed by dissolution of the bulk resist in Microposit 1165 solution with
heating. The heating of the Microposit 1165 solution accelerates and improves the dissolving process. The Piranha solution is also very effective
in dissolving hard resist for releasing RF MEMS switches.
© 2007 Elsevier B.V. All rights reserved.

Keywords: RF MEMS; Resist angle; Hard-baking parameters; Thermal cross-linking; Resist skin; Plasma stripping

1. Introduction First of all, it will reduce the number of required process steps.
If we use another material as a sacrificial layer, it will require
Radio frequency microelectromechanical systems (RF extra process steps like patterning and etching. Second, we can
MEMS) devices are becoming increasingly popular owing to tune the resist angle for a climbing anchor pattern bridge at the
their very good performances at RF and microwave frequen- edge, by optimizing the photolithography process parameters
cies [1,2]. RF MEMS switches have very low insertion loss like hard-baking parameters (temperature and time), exposure
in on-state and very high isolation in off-state. Electrostati- gap (gap between the mask and the resist surface), etc. For sac-
cally actuated MEMS switches consume very little power. They rificial materials like oxide or metal, it would be much more
are very linear and have very high third order intercept point difficult to obtain the desired sloping profile or tune its angle at
compared with any other solid-state counterparts. Surface micro- the edge.
machining is the most common fabrication process for RF In order to have a simple fabrication process for the suspended
MEMS. We are developing a process to fabricate capacitive RF bridge, a climbing anchor is preferred to a vertical anchor. If we
MEMS switches at SINTEF MiNaLab, Oslo, Norway. want to fabricate a vertical anchor as shown in Fig. 2, we have to
In order to fabricate suspended bridges or cantilevers, pho- sputter the metal first for the anchor and then do the lithography
toresist can be used as a sacrificial layer [3], as shown in Fig. 1. to define the vertical anchor. Subsequently, we need to sput-
There are several advantages in using resist as a sacrificial layer. ter again and define the bridge, which increases the number of
required process steps. RF MEMS switches perform switching
(on/off) by the mechanical movement (up/down) of the sus-
∗ Corresponding author. Tel.: +47 7359 2784; fax: +47 7359 1441. pended bridge. Fig. 3 shows an actuated bridge in downstate.
E-mail address: [email protected] (S.C. Saha). For reliable switch operation, the anchor should have sufficient

0924-4247/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.sna.2007.10.067
Author's personal copy

S.C. Saha et al. / Sensors and Actuators A 143 (2008) 452–461 453

baking time. This is more efficient compared with work pub-


lished on resist angle tuning by either baking temperature or
baking time [4,6]. The exposure gap between the mask and the
resist surface could also be optimized to tune the resist angle as
mentioned in [7,8]. We have also developed a systematic pro-
cess to release the RF MEMS switches by dissolving the hard
resist. From the experiment we have found that we can use the
plasma stripping to ash the initial strong resist skin. Afterwards
Fig. 1. A cross-section view of the sputtered bridge on sacrificial layer with we can use heated Microposit 1165 to dissolve the rest of the
climbing anchor. resist completely. This process looks very efficient and we can
obtain very clean switches without any resist residue on the sub-
strate. It was also found that the Piranha solution could be very
effective in dissolving hard and cross-linked resist. The Piranha
solution is a mixture of H2 O2 and H2 SO4 (1:2) and it was heated
at 140 ◦ C. To dissolve hard resist with the ordinary method of
using a typical solvent would require hours or even a day of
soaking in the solvent with ultrasound stirring. Even so in some
cases it would not completely dissolve the resist. Our developed
method is very successful and time-efficient. This work will be
very helpful for fabrication of any kind of surface micromachin-
Fig. 2. A cross-section view of the suspended bridge with vertical anchor. ing MEMS device for tuning the initial sacrificial layer angle
and to release it at the final stage of the process.
thickness compared with the rest of the bridge. To guarantee A review of the thermal reflow and thermal cross-linking of
this, the resist angle should be fewer than 45◦ at the (steepest) photoresist is presented in Section 2. The experimental methods,
position of the anchor. If we have a steeper angle, more than with reference to previous work, are presented in Section 3. The
45◦ , the metal thickness of the anchor may be significantly thin- results and discussion are presented in Section 4, followed by
ner than the rest of the bridge [4]. The improvement of the step the conclusion in Section 5.
coverage with decreasing sidewall angle is also supported by
(Figs. 5, 7 and 8 presented by [5]). The thinner metal at anchor 2. Thermal reflow of photoresist, thermal cross-linking
will reduce the reliability of the suspended bridge significantly. of photoresist and formation of hard skin
If the angle is very small, the planarity of the switch will be
worse. To obtain a smaller angle, the resist needs to be hard- 2.1. Thermal reflow of photoresist
baked at a higher temperature for an extended period of time.
The resist will be harder and it will be difficult to dissolve during When photoresist is hard-baked at a temperature higher
the release process of the MEMS switch afterwards. than the glass transition temperature (Tg ), the resist viscosity
We deposited gold by dc sputtering to form the suspended decreases and at a certain temperature the resist becomes fluid
bridge on the tuned resist. After lithography and patterning of with a fixed surface tension. The system tries to achieve a state
the switch, the sacrificial resist need to be dissolved to release of minimal potential energy and therefore minimizes its surface.
the suspended bridge. Acetone and Microposit 1165 provided by Since the smallest surface for a given volume is a sphere, a cubic
Shipley can be used for dissolving resist. We used both acetone pattern tends to form a hemispherical surface [9]. The thermal
and Microposit 1165 with heating but found some residue left reflow of resist depends on various factors such as baking tem-
and the samples were not fully cleaned. The residue looks like perature, baking time, resist characteristics and resist volume,
threads, which gives us the indication of thermal cross-linking. etc. According to [10], the shrink bias or shrinking of the crit-
The residue can cause possible stiction during actuation of the ical dimension (CD) as shown in Fig. 4, can be expressed as
switch. a functional relationship of the baking temperature, normalized
Here we have shown that it is possible to obtain a desired volume V(Vn ), and the native property (Kr ) of the resist, as shown
angle by optimizing the combination of baking temperature and in Eq. (1):

CD = (Tb , Vn , Kr ) (1)

Here Tb is the baking temperature for a fixed time and Vn is the


normalized value as square area–circular area [10].
The critical dimension shrinkage depending on temperature
is given by

CD = B × exp(A × Tb ) + C (2)

Fig. 3. An actuated MEMS switch in down state. where A, B, C are constants [10].
Author's personal copy

454 S.C. Saha et al. / Sensors and Actuators A 143 (2008) 452–461

pre-baking at 80 ◦ C for 10 min, the dissolution rate of the resist


is greatly decreased, and when the resist is baked at 120 ◦ C
the dissolution rate is very low, as presented by Moon et al.
(Fig. 2; [13]). Post-exposure baking also increases the thermal
cross-linking in the unexposed area. That means that the unex-
posed resist is hardened twice, which makes the resist difficult
to dissolve. Miyagawa et al. [14] also presented work on thermal
cross-linking of positive tone photoresist with o-naphthoquinone
diazide (NQD). They mentioned that in the polymer matrix the
thermal reaction of multifunctional NQD compound with co-
polymers with hydroxyl groups yielded sufficiently cross-linked
structures at high temperatures around 140 ◦ C for 10 min. It was
also observed that heating made a great difference in solubil-
ity between exposed and unexposed areas. They also found that
the gel fraction (the degree of cross-linking) increased with the
post-exposure baking temperature (Fig. 8; [14]).

2.3. Formation of hard resist and skin on the surface

From Section 2.2 we can see that for high temperature and
Fig. 4. The change in critical dimension (CD) before thermal reflow (a) and longer baking time a thermal cross-linking occurs for positive
after thermal reflow (b).
tone resist (in fact, this is also the case for negative resist [15,16]).
The resist at the unexposed area becomes twice hardened by
The resist reflow also depends on the baking time. For a
both pre-bake and post-bake. To release RF MEMS switches
typical resist structure the thermal reflow length is given by [11]
we have to dissolve mainly unexposed and baked resist. The

μt increase in the resist hardness with thermal cross-linking is also
L= (3) supported by [17]. Kawai [17] has presented work on the ther-
ρ
mal properties of photoresist. The resist used was a Novolak
where ρ is the resist density, t the baking time and μ is the resin mixture and he found that the Knoop hardness of the resist
fluid viscosity. The fluid viscosity μ(T) at specific temperature increases with baking temperature (Fig. 1; [17]). The work also
is given by shows the indentation rate of AFM (atomic force microscope)
  tips with baking temperature, which shows a depth profile at
E
μ(T ) = μ0 exp (4) mid-temperature range. The author suggests that the data for the
RT
depth profile region of hardness show that a hardened rigid resist
where E is the activation energy, μ0 the pre-exponential factor layer is formed on the resist surface and that the inner layer still
and R is the universal gas constant. A fitted viscosity function remains soft. This hard layer is called skin and is also mentioned
used for simulation is given by [11] in Ref. [18]. According to [18], the hard skin can form as a result
  of reactive ion etching (RIE) or ion implantation. This can also
4848.2
μ = (4.84414 × 1016 )exp − (5) be caused by hard baking of the resist or if the resist is exposed to
T
a high temperature process. In the processing of the switches, the
From Eqs. (2) and (3) it can be seen that the baking temperature resist experiences several high temperature cycles. One of them
has a more pronounced effect on the resist edge (angle) than the is high temperature hard baking, around 115–120 ◦ C, to tune the
baking time. The resist extension or angle at the edge can be resist angle, as mentioned in Section 1. During the metal sput-
varied more effectively with temperature than with time. This tering for the suspended bridge, the sacrificial resist can also
is also supported by [12]. The HIPR resist containing Novolac experience high temperature back sputtering.
resin, which we used for our experiment, has a glass transition For the positive tone resist, a short delay time between coat-
temperature of Tg = 90 ◦ C. ing and DUV exposure can produce a ‘lip’ at the top surface of
the relief image. A long delay time between coating and expo-
2.2. Thermal cross-linking in positive tone resists sure causes a thin skin to form between resist and air interface
[19].
Moon et al. [13] presented results on three-component pho-
toresist based on thermal cross-linking. From their experiment it 3. Experiments
was found that the vinyl ether groups react with hydroxyl groups
of the binder polymer at high temperatures (from pre-baking) A two-mask process was used for our experiments. The aim
through electrophilic addition reaction to form cross-linked net- of the experiments was to develop several process steps. One was
works. These authors also suggest that thermal cross-linking to tune the resist angle at the anchor position for the climbing
does not occur at temperatures around or below 50 ◦ C. But after anchor/bridge. Another was to develop a successful release pro-
Author's personal copy

S.C. Saha et al. / Sensors and Actuators A 143 (2008) 452–461 455

the wavelength of the exposure light. This is rather complicated,


because it will require a special filter for wavelength selection.
This will increase the cost of the fabrication. We have performed
an experiment to inspect the combined effects of baking time and
temperature on resist angle.
Some results of the resist angle tuning have already been pre-
sented as a short abstract [20]. We used two types of resist for the
resist tuning experiment, HIPR 6517 and diluted AZ 4562. We
targeted a thickness of 2.5 ␮m for the sacrificial layer. For HIPR
6517, we got a thickness of 2.55 ± 0.01 ␮m at a spin speed of
2000 rpm. AZ 4562 is a thick resist and we diluted it with sol-
vent in a 20:7 ratio. We obtained a thickness of 2.55 ± 0.03 ␮m at
3250 rpm spin speed. Here 0.01–0.03 ␮m was the deviation over
the wafer. The solvent used was MICROPOSIT EC SOLVENT
(2-methoxy-1-methylethyl acetate) from Rohm and Haas. After
spinning, the resist was soft-baked at 100 ◦ C for 90 s on a hot-
plate. Then we exposed the resist in automatic exposure tools
through the anchor mask (opening for metal deposition to sub-
strate). The exposure power was 35 mW/cm2 . We exposed the
HIPR 6517 resist for 5 s with an exposure gap of 25 ␮m. The
diluted AZ 4562 resist was exposed for 5 s with an exposure
gap of 30 ␮m. We also varied the exposure gap for a few wafers
to investigate its effect on the resist slope. After exposure, we
developed all the wafers in MF 312 solution (1:1) for 1 min.
After development, the resist was hard-baked on a hotplate
at temperatures ranging from 100 ◦ C to 130 ◦ C for duration of
90–180 s. A few wafers were also baked for a longer time (5 min)
and at a higher temperature (150 ◦ C). For some wafers the expo-
Fig. 5. A brief overview of the process flow for two masks process. sure gap was changed to 50 ␮m, instead of the standard gap.
The detailed conditions and obtained resist angle are presented
in Section 4.
cess by dissolving the sacrificial resist and subsequently drying After baking we cut the wafers through the anchor position
the switches. or opening in the sacrificial resist layer. Then we mounted the
A brief process overview is shown in Fig. 5. A p-type, one wafers vertically on the sample holder. We inspected the edge
side polished, 350 ␮m thick (1 0 0) 4 silicon wafer was used profile of the patterned resist in a scanning electron microscope
for the fabrication (a). A 2.5 ␮m thick resist was coated on the (SEM). We took SEM images at several positions on each wafer.
silicon substrate by spin coating (b). The resist was pre-baked Owing to charging it can be difficult to obtain a clear SEM
at 100 ◦ C for 90 s on a hotplate and then exposed to form the image of polymer materials, so we used a very low vacuum
opening by means of the anchor mask. The resist was developed (∼0.5 mbar) and low excitation voltage (3–5 kV) for the SEM
in the MF 312 (1:1) solution for 1 min (c). The wafers were then inspection. Other surface profile measurement systems, such as
hard-baked at the desired baking parameters, and then gold was a profilometer (needle), optical interferometry and atomic force
sputtered onto the sacrificial layer and substrate. The thickness microscopy (AFM), can also be used for this purpose. The pro-
was close to 1 ␮m (d). The gold was patterned by the switch mask filometer uses a sharp needle and a small amount of force/weight
and etched in a KI4 solution to form the switch (e). Subsequently, to scan the surface. The resist is usually soft, which may cause
the resist was dissolved by the process developed in this work, the needle to penetrate beyond the surface into the bulk resist.
followed by drying and release in a CO2 critical point dryer (f). As we were interested only in the steepest part with a very sharp
image of the rising part, this system was not suitable for find-
3.1. Resist angle tuning ing the exact angle. Optical interferometers usually give a very
accurate result, but not for samples where the surface has steep
Some work has been published on the dependence of resist slopes as in our example. In our sample the ratio of the vertical
angle on different parameters. The resist angle depends on the distance to the horizontal distance at the sloping part was close
baking time [4], the baking temperature [6] and the exposure to unity, which made it difficult to get an accurate surface profile
gap [7]. The work in [4] aims for a 45◦ angle. Refs. [6,7] seek of this region. The AFM would have been a good option, but we
to obtain a vertical profile. The goal in Ref. [4] is to obtain the did not have access to it in our lab.
desired angle by varying the baking time. The authors of [6] try The reliability of the sputtered bridge depends on the steep-
to optimize the resist angle by tuning the baking temperature. est angle of the resist. We measured the resist angle at the
Ref. [7] tries to optimize the desired resist angle by controlling steepest part of the edge for different baking conditions. The
Author's personal copy

456 S.C. Saha et al. / Sensors and Actuators A 143 (2008) 452–461

Fig. 7. A top view of the switch with possible way of access the resist by solvent.

ters and in our design the bridge width ranged from 20 ␮m to


100 ␮m. The solvent also has to dissolve the resist underneath
Fig. 6. A schematic view of angle measurement from SEM image.
the bridge from the side, as shown in Fig. 7, and this will take
longer to clean the resist fully. Microposit 1165 is regarded as
a better option to dissolve hard resist compare with acetone. So
resist angles were obtained by measuring the horizontal and we used Microposit 1165 for our experiment. From discussion
vertical distance at the steepest part of the anchor, using the equa- with the relevant people, it was found that heating could acceler-
tion θ = tan−1 (Vertical distance (Y)/Horizontal distance (X)), as ate the dissolution of resist in the solution. It was also suggested
shown in Fig. 6. [18] that a short plasma strip at the beginning could remove
the skin. The rest of the bulk resist can be effectively dissolved
3.2. Release methods of hard resist by Microposit 1165 with heating. It was also learned that the
Piranha solution could effectively dissolve hard resist. We con-
Owing to availability and ready-to-use facilities, we used the ducted an experiment to dissolve hard resist with possible skin
HIPR 6517 resist as a sacrificial layer for our test device fab- with all the different options available in our lab.
rication. This positive tone resist, HIPR 6517, is provided by Another suggestion [18] was that ashing of the top 10–20%
Fujifilm. The chemical compounds of the resist are ethyl lac- of the total resist would be enough for subsequent release. So
tate (45–60%), ethyl-3-ethoxypropionate (20 to <25%), Novolac we targeted to ash the (around) 300 nm thick resist skin with
resin (15–25%), and naphthoquinone diazide ester derivative plasma ashing. We did a brief plasma stripping trial to measure
(3–10%). the ashing rate of the plasma tool in our lab. From this experiment
Acetone can be used to dissolve ordinary resist. We kept we found that the plasma tool can ash 300 nm skin at 150 W in
the sample in acetone for several hours (overnight) and also 45 min. We chose low power intentionally to avoid high temper-
heated the solvent for 1–2 h with magnetic stirring. After drying, ature during stripping, which could buckle the patterned bridge
we inspected the samples in an optical microscope and found by annealing.
that the resist was not completely dissolved and a lot of residue We selected eight samples with the same process parameters
material remained on the substrate. The Shipley Microposit 1165 (i.e. baking temperature, gold sputtering, etc.) for our experi-
resist remover is specially used to dissolve hard resist. We also ment. The resist was pre-baked at 100 ◦ C for 90 s and post-baked
used this solvent to release our switches. We kept the sample at 120 ◦ C for 5 min on a hotplate. The samples were labelled RS
in Microposit 1165 for some 4–5 h followed by an additional (resist stripping) 1–8 for convenience of identification. A brief
1–2 h of heating with stirring, and found that still some resist plan of the experiment is shown in Table 1. We ashed four sam-
residues remained on the sample. The residue looks like threads, ples, RS 1–4, in a plasma stripper. Four samples, RS 5–8, were
which gave us the indication of thermal cross-linking. If the kept without ashing. After that, we put all the samples RS 1–8
resist residue is left after release it could change the electrical in Microposit 1165 for about 90 min at room temperature with
characteristic of the switch and possibly also cause stiction. magnetic stirring. Then we put the specific samples RS 1, 2, 5
In order to release RF MEMS switches, the solvents have to and 6 in heated Microposit 1165 for 60 min with magnetic stir-
dissolve both the resist on the substrate around the bridge and ring. We heated the solution up to a temperature of around 60 ◦ C.
also the resist underneath the suspended bridge. The sacrificial The rest of the samples we kept in Microposit 1165 for the same
layers are usually few micrometers thick and in our case it was time at room temperature. Finally we put all the samples again
2.55 ␮m. The bridge width may be several tens of microme- in Microposit 1165 at room temperature with magnetic stirring

Table 1
The experimental setup to dissolve hard resist with plasma ashing and Piranha dip
Plasma strip (150 W, 45 min) No plasma strip

Heating Microposit 1165 (∼60 ◦ C) No heating Microposit 1165 Heating Microposit 1165 (∼60 ◦ C) No heating Microposit 1165

Piranha dip, 60 s No Piranha dip Piranha dip, 60 s No Piranha dip Piranha dip, 30 s No Piranha dip Piranha dip, 30 s No Piranha dip

RS 1, very clean RS 2, very clean RS 4, very clean RS 3, almost clean RS 5, Very clean RS 6, thin skin left RS 7, very clean RS 8, not clean
Author's personal copy

S.C. Saha et al. / Sensors and Actuators A 143 (2008) 452–461 457

Table 2
The resist angle of HIPR 6517 for different baking conditions
Baking temp. (◦ C) Baking time (s) Exposure gap (␮m) Resist angle (◦ )

No bake 25 65
100 90 25 59
100 120 25 57
100 180 25 55
115 90 25 49
115 120 25 45
115 180 25 43
115 300 25 41
130 90 25 34
130 180 25 35
150 90 25 28
100 90 50 45

for another 60 min. After this, we dipped the specific samples


RS 1,4,5 and 7 in a Piranha solution. Two samples, RS 1 and
4 were kept in Piranha for 30 s and the remaining two samples,
RS 5 and 8, were kept for 1 min. This was done intentionally
to check if the Piranha attacked the gold switches. We did not
find any evidence of such an attack in either of these samples.
After the experiment, we dried all the samples and inspected
them in an optical microscope. In order to avoid stiction dur-
ing wet release, we used CO2 critical point dryer to release our
fabricated RF MEMS switches.

4. Results and discussion

4.1. Resist angle tuning

The variation of the resist angle with time and temperature can
be seen from the SEM images shown in Fig. 8. For comparison
the SEM images for HIPR 6517 have been scaled to the same
magnification and compared in Fig. 8. From Fig. 8 it can be
easily seen that the resist angles decreased with both baking
temperature and baking time. The measured resist angles, for
both HIPR 6517 and diluted AZ 4562, with different baking
conditions, are shown in Tables 2 and 3. A variation of resist
angle with baking temperature for a baking time of 90 s is shown
in Fig. 9. Also the variation of resist angle with baking time at
a baking temperature of 115 ◦ C is shown in Fig. 10. Fig. 8. The SEM images for HIPR resist at same scale. Baking temp 100 ◦ C
(top), 115 ◦ C (second from top), 130 ◦ C (third from top) all for 90 s baking time.
From Tables 2 and 3, it can be seen that the resist angle Baking temp 115 ◦ C for 120 s baking time (bottom).
decreases with both baking temperature and time. The baking
temperature has greater effect on resist angle than the baking
time does. It is also seen from Fig. 9 that the angle decreases at
Table 3 a lower rate at higher temperatures. The baking time has very
The resist angle of diluted AZ 4562 for different baking conditions little effect at higher baking temperatures. Also, the resist angle
Baking temp. (◦ C) Baking time (s) Exposure gap (␮m) Resist angle (◦ )
decreases little after a certain period of baking. The effect of
baking temperature and time has similar effect on both kinds of
No bake 30 64 resist. The resist angles of diluted AZ 4562 decrease at a slower
100 90 30 59
100 120 30 59
rate with baking time than HIPR 6517, as can be seen from
100 180 30 56 Fig. 10.
115 90 30 47 Comparing our work with [4,6], it can be seen from Fig. 9 and
115 120 30 46 the data from Puchner et al. (Fig. 5; [6]) that the variation of the
115 180 30 43 angle with temperature is sharper in Ref. [6] (large resist pattern)
130 90 30 37
115 90 50 40
than in our work. From Fig. 10 and the work presented by Grenier
(page 42; [4]) it can be seen that the resist angle decreases at a
Author's personal copy

458 S.C. Saha et al. / Sensors and Actuators A 143 (2008) 452–461

Fig. 9. Resist angle vs. baking temperature for 90 s baking time. Fig. 10. Resist angle vs. baking time at 115 ◦ C baking temperature.

Fig. 11. The optical image of all the released samples together. The samples identification nos. RS 1–8 are marked on each image.
Author's personal copy

S.C. Saha et al. / Sensors and Actuators A 143 (2008) 452–461 459

higher rate with increasing baking time [4] compared with our
work. For both works the rate of reduction in angle decreases
with time. In Ref. [4] the reduction in angle stops after a certain
time but in our work it continues to decrease with time. An
exact comparison is difficult because different hotplates were
used. The initial angle was a little smaller in our experiment
than the work presented in Refs. [4,6].
From the experiment it is seen that we can obtain an angle of
around 45◦ by baking at 115 ◦ C for 120 s. From Tables 2 and 3
it is also seen that the exposure gap has a strong effect on the
resist angle, and the resist angle decreases with a higher exposure
gap. From this experiment we have seen that we can combine
the baking temperature with baking time in an optimum way to
obtain a desired angle.
Fig. 12. The released cantilever by plsma strip at 400 watts for 45 min (It can
4.2. Release method of hard resist be seen easily that at anchor the cantilever buckled up).

The images of the dried samples are shown in Fig. 11. From at low power. For some samples we used a high power around
Fig. 11B it can be seen that the hard resist can be dissolved com- 400 watts for 1/2–2 h but the bridge and cantilever were buckled
pletely by plasma stripping and hot Microposit 1165 solution. It owing to the heating produced by plasma power. The anchor part
is also seen from Fig. 11B and F that the ashing has great effect got a high peak and the suspended part buckled down, which can
on the release. The sample in Fig. 11F, without ashing and with be seen from the cantilever group shown in Fig. 12. The plasma
heating, still has some resist skin residue left, which can be seen stripping method can be made effective by introducing holes in
on the bridge edge. It is also seen from Figs. 11F and H that the suspended part, allowing the plasma to access the resist quite
heating has an effect on the dissolution of the resist. The amount easily and quickly.
of resist residues left on the sample without heating is much In order to test that there was no residue left underneath the
higher than with heating. From Fig. 11H it can be seen that the bridge, some cantilevers were folded into an upside-down posi-
sample without ashing, heating and Piranha dip has still left a tion and inspected in an optical microscope. For comparison
lot of resist skin and residue. From Fig. 11 A, D, E and G, it can a sample from ordinary release and one released by plasma
be seen that Piranha is very effective in dissolving hard resist. It stripping and heating (RS 2) were inspected. The comparison
can dissolve the resist even without plasma ashing. One has to is shown in Fig. 13. From this figure it can be seen that RS
be careful, however, that the high temperature (around 140 ◦ C) 2 is completely clean while the sample with ordinary release
does not affect the device performance and that Piranha does not has some residue left on the backside of the bridge and can-
attack the metal membrane. tilever. Using these developed methods, we have fabricated the
It is also possible to use only plasma stripping to release the complete switch with transmission line and dielectric layer. The
RF MEMS switches. This is a dry process and usually simple SEM images of the released switch have been shown in Fig. 14
to use. The vertical strip of 2.55 ␮m thick resist is very fast but (top view) and Fig. 15 (15◦ tilted).
it takes a long time to etch the resist underneath the bridge. The In order to release the MEMS structure completely, with-
plasma cannot attack (access) the resist directly as the suspended out any residue and threads on the bridge and substrate, one
gold bridge works as a mask. So the resist underneath the bridge can use plasma stripping first to remove the top layer of thin
(∼100 ␮m wide) has to be ashed from the side, through the initial skin. Afterwards the sample can be put in Microposit 1165 with
gap of the bridge (2.55 ␮m), which will be quite time-consuming possible heating. It is also advantageous to use some kind of

Fig. 13. The back side of the released cantilever folded backside upward, with our develop process, RS 2 (left) and with ordiany slovent release (right).
Author's personal copy

460 S.C. Saha et al. / Sensors and Actuators A 143 (2008) 452–461

seen that plasma ashing followed by Microposit 1165 remover


with possible heating can successfully dissolve the thermally
cross-linked hard resist and release the RF MEMS switches.
This work will be very helpful in tuning the resist angle and
release of RF MEMS switches, which use resist as a sacrificial
layer. The results of this experiment are applicable to any kind
of surface micromachined MEMS device.

Acknowledgements

The authors are grateful to the Norwegian Research Coun-


cil for sponsoring the work through the SMiDA project (no.
159559/130) and the IRRFT project (no. 159259/I40). The
authors also thank MiNa Lab personnel for their support in the
clean room.

References
Fig. 14. The SEM image of a released RF MEMS switch (top view).
[1] G.M. Rebeiz, J.B. Muldavin, RF MEMS switches and switch circuits, IEEE
Microwave Magazine 2 (4) (2001) 59–71.
[2] H.J. De Los Santos, S. Rassoulian, J. Maciel, MEMS for future microwave
system, in: Proceedings of 2005 IEEE MTT-S, Long beach, CA, USA,
2005, pp. 905–908.
[3] K. Walsh, J. Norville, T. Yu-Chong, Photoresist as a sacrificial layer by dis-
solution in acetone, in: Proceedings of 2001 IEEE Int’l MEMS conference,
Interlaken, Switzerland, 2001, pp. 114–117.
[4] K. Grenier, (LAAS-CNRS), IMEC Training: MEMS in RF and MW Elec-
tronics, Belgium, April, 2005.
[5] T. Smy, R.N. Tait, K.L. Westra, M.J. Brett, Simulation of density and step
coverage for via metallization, in: Proceedings of VMIC conference, Santa
Clara, CA, USA, 1989, pp. 292–298.
[6] H. Puchner, R. Castagnetti, W. Pyke, Minimizing thick resist sidewall slope
dependence on design geometry by optimizing bake condition, Microelec-
tron. Eng. 53 (2000) 429–432.
[7] C.J. Hamel, A low cost robust method to vertical sidewalls for thick resist
processing, in: Proceedings of 2004 IEEE/SEMI Int’l Electronics Manu-
facturing Technology Symposium, San Jose, CA, USA, 2004, pp. 247–250.
[8] W. Kang, E. Rabe, S. Kopetz, A. Neyer, Novel exposure methods based
on reflection and refraction effects in the field of SU-8 lithography, J.
Micromech. Microeng. 16 (2006) 821–831.
Fig. 15. The enlarged SEM image of the middle section of the switch bridge, [9] S. Audran, B. Faure, B. Mortini, J. Regolini, G. Schlatter, G. Hadziioan-
above the electrode (tilted 15◦ ). nou, Study of mechanism involved in photoresist microlens formation,
Microelectron. Eng. 83 (2006) 1087–1090.
stirring, which will improve the release quality and accelerate [10] J.-S. Kim, C.-W. Koh, G. Lee, J.-C. Jung, K.S. Shin, Novel routes towards
the process. This method is very effective and also less time- sub-70nm contact windows by using new KrF photoresist, in: Proceedings
consuming. The total dissolution process may take 3–4 h. The of SPIE, vol. 4345, 2001, pp. 232–240.
ordinary solvent methods require hours and even days and also [11] S.-K. Kim, Resist reflow process simulation study for contact hole pattern,
require ultrasound stirring for violent shaking, and sometimes J. Vac. Sci. Tech. B 24 (1) (2006) 200–204.
[12] T. Leveder, S. Landis, L. Davoust, N. Chaix, Optimization of demolding
this is still not enough to clean the sample completely. We have temperature for throughput improvement of nanoprint lithography, Micro-
used this process on several occasions to release our samples and electron. Eng. 84 (2007) 953–957.
found it to be very effective without leaving any kind of residue [13] S.-Y. Moon, C.-M. Chung, T. Tamaoka, Three component photoresist
on the substrate. based on thermal crosslinking and acidolytic cleavage, Polymer 41 (2000)
4013–4019.
[14] K. Miyagawa, K. Naruse, S. Ohinishi, K. Yamaguchi, K. Seko, N. Numa,
5. Conclusions N. Iwasawa, Study of thermal crosslinking reaction of O-naphthoquinone
diazides and application to electrodeposition positive photoresist, Prog.
We have developed a method to tune the resist angle by differ- Org. Coat. 42 (2001) 20–28.
ent hard-baking parameters and optimized a process to release [15] K.H. Chae, K.H. Sung, A polymeric photobase generator containing
oxime-urethane groups: crosslinking reaction and application to negative
the RF MEMS switch without any resist residue left. The baking
photoresist, J. Polym. Sci. A: Polym. Chem. 42 (2004) 975–984.
temperature and time can be combined in an optimum way to get [16] H.-H. Liu, W.-T. Chen, F.-T. Wu, Characterization of Negative tone pho-
a desired angle. The baking temperature has a prominent effect toresist based on acid catalyzed dehydration crosslinking of novolak having
on the resist angle. From the release method experiment it is pendant carboxyl group, J. Polym. Res. 9 (2002) 251–256.
Author's personal copy

S.C. Saha et al. / Sensors and Actuators A 143 (2008) 452–461 461

[17] A. Kawai, Measuring the thermal properties of photoresist thin film using Science and Technology, before he started as a Research Scientist at SINTEF
atomic force microscopy, Thin Solid Films 273 (1996) 308–311. ICT, micromechanics and nanotechnology in 2001. The main field of work at
[18] K. Shaw, online correspondence. SINTEF is process development for Si MEMS.
[19] S.A. MacDonld, Airborne chemical contamination of a chemically ampli-
fied resist, Proc. SPIE 1466 (1991) 2–12. Geir U. Jensen received the MSc degree in physics and the PhD degree in
[20] S.C. Saha, H. Sagberg, E. Poppe, G.U. Jensen, T. Sæther, Tuning of resist electrical engineering from the Norwegian Institute of Technology, Trondheim,
slope for RF MEMS with hard baking parameters, in: Proceedings of MNC in 1984 and 1989, respectively. His PhD work was on Monte Carlo simulation of
2006, Kanagawa, Japan, 2006, pp. 314–315, Abstract. transport in III–V semiconductor devices. From 1989 to 1990, he was a Research
Fellow at the University of Minnesota, Minneapolis, and from 1990 to 1991,
he was a Research Scientist at the University of Virginia. Charlottesville. He
Biographies was with SINTEF DELAB, Trondheim, from 1991 to 1993, and with Telenor
R&D, Kjeller, Norway, from 1993 to 1995. His research during this period
Shimul Chandra Saha received the BSc degree in Electrical and Electron- was in the area of design and fabrication of III–V semiconductor electronic
ics Engineering from Bangladesh University of Engineering and Technology and optoelectronic devices. Since 1995, he has been with SINTEF Electronics
(BUET), Dhaka, Bangladesh in 2001 and MSc degree in Hardware for Wireless and Cybernetics, Oslo, Norway, where he works on silicon radiation detectors,
Communications from Chalmers University of Technology (CTH), Gothenburg, micromachined sensors and RF MEMS.
Sweden in 2004. He is currently working toward his PhD degree in electrical
engineering (focusing on RF MEMS switches and switch circuits) at the depart- Tor A. Fjeldly received the MSc degree in physics from the Norwegian Insti-
tute of Technology and the PhD degree from Brown University, Providence, RI.
ment of Electronics and Telecommunications (IET), Norwegian University of
Science and Technology (NTNU), Trondheim, Norway. His current research He was employed with Max-Planck-Institute for Solid State Physics, Stuttgart,
work involves modeling, design and simulation of RF MEMS switches and Germany, and with the SINTEF research organization, Norway. Since 1983, he
has been with the Norwegian University of Science and Technology (NTNU),
MEMS filter at RF and microwave frequency. He also works on designing
the mask and fabrication of the MEMS switches and circuits in clean room where he is a Professor of Electrical Engineering. He is presently located at the
University Graduate Center at Kjeller, Norway. Dr. Fjeldly has been a Vis-
in collaboration with SINTEF Micro- and NanoLab (MiNa lab), Oslo, Norway.
iting Professor at University of Virginia, Charlottesville, VA, at Rensselaer
From September 2001 to August 2002, he worked as a lecturer in department of
Electrical and Electronics Engineering (EEE), Chittagong University of Engi- Polytechnic Institute, Troy, NY, and at University of Rovira i Virgili, Tarrag-
ona, Spain. His present research interests include nanoscale semiconductor
neering and Technology (CUET), Chittagong, Bangladesh. From January to
devices and RF MEMS. He has written about 200 scientific papers, several
June 2003, he worked in a project on DDS for base station application at Eric-
sson AB, Molndal, Sweden. His research interest includes RF MEMS switch book chapters, and has co-authored and co-edited several books. Dr. Fjeldly
and switch circuits, RF and microwave circuits for wireless communications and is a Fellow of IEEE and a member of the Norwegian Academy of Technical
micro-fabrication. Sciences.

Håkon Sagberg received the siv.ing. (MSc) degree in physics from Norwegian Trond Sæther was born in Ålesund, Norway, in 1958. He received the MSc and
PhD degree in electrical engineering from the Norwegian University of Science
Institute of Technology, Trondheim, in 1996, and his PhD from the University
and Technology, Trondheim, Norway, in 1981 and 1991, respectively. From 1981
of Oslo in 2006. He has been affiliated with SINTEF since 1998, working as a
research scientist at SINTEF Department of Microsystems and Nanotechnology to 1983, he was a Research Scientist at the microelectronics group at SINTEF in
Trondheim. In 1983, he cofounded the fabless semiconductor company Nordic
since 2005. His main fields of research are optical MEMS and RF MEMS.
VLSI, where he worked as the Technical Director until July 1999. In July 1999,
Erik Poppe received a MSc degree from the Department of Physics at the he joined the Circuits and Systems Group at the Norwegian University of Science
Norwegian Institute of Technology in 1994. He then worked as a Research and Technology as a full Professor. His research interests include RF CMOS,
Assistant in semiconductor laser fabrication at the Norwegian University of analog-to-digital converters and microelectromechanical systems.
Metallization scheme and release methods for fabrication of RF
MEMS switches

Shimul Chandra Saha a, Håkon Sagberg b, Erik Poppe b, Geir Uri Jensen b, and Trond Sæther
a

a
Department of Electronics and Telecommunications, NTNU, NO-7491 Trondheim, Norway.
b
SINTEF ICT, PO Box 124 Blindern, NO-0314 Oslo, Norway

In proceedings of 33rd Micro- and Nano- Engineering conference (MNE 2007),


September 2007, pp 773-774, Copenhagen, Denmark.
Metallization scheme and release methods for fabrication of RF MEMS switches
Shimul Chandra Saha a, Håkon Sagberg b, Erik Poppe b, Geir Uri Jensen b, and Trond Sæther a
a
Department of Electronics and Telecommunications, NTNU, NO-7491 Trondheim, Norway.
Ph: +47 735 92784, E-mail: [email protected]
b
SINTEF ICT, PO Box 124 Blindern, NO-0314 Oslo, Norway
Keywords: RF MEMS, sputtering, residual stress, stress gradient, resist skin, release method.
1. Introduction: Radio frequency micro electro mechanical systems (RF MEMS) devices are
becoming increasingly popular due to their very good RF performance at high frequencies. RF MEMS
switches have very low insertion loss in the on state and very high isolation in the off state.
Electrostatically actuated MEMS switches consume very little power. They are very linear and have
very high third order intercept point compared to any other solid state counterparts. A co-planar
waveguide (CPW) MEMS shunt capacitive switch is shown in figure 1. The capacitance is formed
between the suspended metal bridge and the bottom electrode, with a dielectric in between. When the
bridge is in up state the capacitance is small (fF) and the switch is on. When the bridge is actuated as
shown in Fig. 1 (right), the capacitance increases by two orders of magnitude the signal is shorted to
ground, and the switch turns off. We are developing a process to fabricate surface micromachined
capacitive RF MEMS switches in the SINTEF Micro and Nano Laboratories (MiNa), Oslo, Norway.
2. Design and Fabrication: Successful fabrication of suspended metal bridges requires three process
steps: 1) Deposition and patterning of a sacrificial layer, 2) Deposition and patterning of the metal
layer, and 3) Removal of the sacrificial layer. We have used a common positive photoresist (HiPR
6517) as a sacrificial layer. The photoresist is baked and the reflow creates a gradually increasing
thickness at the edges. This is done in order to achieve a reliable bridge profile, rising with an angle of
approximately 45q where it contacts the substrate [1]. We have used DC magnetron sputtering to
deposit gold on the sacrificial layer to form the bridge. The reliability and performance of the switch
depend on the mechanical properties of the bridge, and especially pull-down voltage and switching
time are influenced by residual stress and stress gradients. The stress of the sputtered gold film
depends on sputtering parameters such as pressure and power [2]. We have aimed for a low tensile
stress, because compressive stress may cause buckling, and too much tensile stress will increase the
pull-down voltage. The sputtering power must be low enough to avoid over-heating and burning the
sacrificial resist. After patterning the gold bridge, the sacrificial resist needs to be dissolved to release
the bridge. A thin layer of skin can be formed on the top surface of resist after the lithographic process
and sputtering which makes it difficult to dissolve using a simple acetone soak. Some of the possible
reasons for formation of the skin are, baking, and a delay between coating and exposure of resist [3].
3. Experiments and Results: Coating, patterning and baking of the sacrificial photoresist has been
described in [1]. For all sputtering experiments described below we used standard lithography.
The sputtering pressure was varied from 10 mTorr to 20 mTorr to tune the film stress. We used a
low sputtering power of 500 Watts for all experiments. The sputtering conditions and obtained results
are shown in table 1. Example images of structures sputtered at 10 mTorr and 500 watts are shown in
Figure 2. In order to measure the residual stress and stress gradient of the sputtered film, we used
wafer curvature measurements and micro-machined test structures [4]. For wafer curvature method,
we deposited gold directly on wafers and measured the wafers curvature before and after film
deposition. In micro-machined test structures methods, cantilevers were used to measure the stress
gradient, bridges were used to measure compressive stress, and guckel rings with CoventorWare
simulations were used to measure the tensile stress. A lower tensile stress is obtained at higher
pressure. A decrease in negative stress gradient was also found with increasing sputtering pressure.
To remove the sacrificial resist, acetone was quite ineffective, and we obtained better results with
Shipley Microposit 1165 resist remover. An experiment plan as shown in table 2 was carried out to
explore the different options for removal of resist skin. The samples were put in Microposit 1165 for
about 90 min at normal temp, followed by 60 min with heating and another 60 min at normal temp, in
all cases with magnetic stirring. Afterwards the samples were dried in a critical point dryer. Two
released samples RS 2 and RS 8 are shown in figure 3. From the release test, it is found that the
plasma ashing and heating has a great effect on the release process as shown in figure 3.
4. Conclusion: We can use 20 mTorr at 500 watts to deposit low stress planar gold. Plasma ashing,
followed by Microposit 1165 with heating can be used for successful release of RF MEMS switch.
Figure 1: A side view of RE MEMS capacitive shunt bridge with sacrificial resist (left) and an actuated bridge (right).
Table 1: The sputtering parameters and measured residual stress and stress gradients
Wafer curvature methods Micro-machined test structures method
Sputtering Residual stress Sputtering Stress gradient Residual stress,
pressure* (mTorr) tensile (MPa) pressure* (mTorr) -(MPa/Pm) tensile (MPa)
10 42 10 35 70
16 31 15 5.2 55
24 31 20 4.8 45
*All the wafers are sputtered at 500 watts. The uncertainty of the stress value is 5-15 MPa.
Table 2: The experimental setup to dissolve hard resist with plasma ashing and piranha and obtained results
Plasma strip (150 watt, 45 min) No plasma strip
Heating Microposit 1165 No heating Microposit Heating Microposit 1165 No heating Microposit
(~60q C) 1165 (~60q C) 1165
Piranha No Piranha No Piranha dip, No piranha Piranha No
dip, 60 s piranha dip,60 s piranha 30 s dip dip, 30 s piranha
dip dip dip
RS 1 RS 2 RS 4 RS 3 RS 5 RS 6 RS 7 RS 8
Very clean Very Very Almost Very clean Thin skin Very Not
clean clean clean left clean clean

Figure 2: Optical images of buckled guckel ring and cantilever sputtered at 10 mTorr and 500 watt

Figure 3: Released sample RS 2 (left) and Released sample RS 8(right)

Acknowledgments: Authors are grateful to Norwegian Research Council for sponsoring the work through the
SMiDA project (No 159559/130) and the IRRFT project (No 159259/I40).
References:
[1] S. C. Saha, Tuning of resist slope for RF MEMS with hard baking parameters, Proceedings of MNC 2006.
[2] H Windischmann, Critical Reviews in Solid State and Materials Science, 17 (6): p 547-596 (1992).
[3] S.A. MacDonld, Proc. SPIE, vol 1466, p 2-12, (1991).
[4] C.-W. Baek, Sensors and Actuators A 117, p 17-27, (2005).
Tuneable Low-Pass Filter from C to X band with RF MEMS
Capacitance and Transmission line

Shimul Chandra Saha*1, Ulrik Hanke2, Håkon Sagberg3, Tor A. Fjeldly1and Trond Sæther1
1
Department of Electronics and Telecommunications (IET), Norwegian University of Science and
Technology (NTNU), NO-7491 Trondheim, Norway
2
Institute for Microsystems Technology, Vestfold University College, 3103 Tønsberg, Norway
3
Department of Microsystems and Nanotechnology, SINTEF ICT, PO Box 124, Blindern, NO-0314
Oslo, Norway

Has been submitted to Microelectronics Journal


Tuneable Low-Pass Filter from C to X band with RF MEMS
Capacitance and Transmission line
Shimul Chandra Saha*1, Ulrik Hanke2, Håkon Sagberg3, Tor A. Fjeldly1and Trond

Sæther1
1
Department of Electronics and Telecommunications (IET), Norwegian University of

Science and Technology (NTNU), NO-7491 Trondheim, Norway


2
Institute for Microsystems Technology, Vestfold University College, 3103 Tønsberg,

Norway
3
Department of Microsystems and Nanotechnology, SINTEF ICT, PO Box 124,

Blindern, NO-0314 Oslo, Norway

*Contact Author, E-mail: [email protected], Phone: +47 7359 2784, Mobile:

+47 905 36957, Fax: +47 7359 1441

Abstract: In this paper we have shown the design of an RF MEMS tuneable low-pass

filter. The theory of traditional parallel plate capacitors and high capacitance ratio RF

MEMS shunt capacitive switches is described. The variation of capacitance with

actuation voltage for the high capacitance ratio switch is verified with simulation. The

low-pass filter design uses both distributed transmission lines and RF MEMS

capacitances to replace the lumped elements. The use of RF MEMS variable

capacitances gives the flexibility of tuning the cut-off frequency of the low-pass filter.

We have designed a low-pass filter at 7 GHz cut-off frequency using the theory of

stepped impedance transmission lines and MEMS capacitances. A 3D electromagnetic

simulation of the filter is performed and agrees quite well with ADS simulation. A

prototype of the tuneable low-pass filter is realised by use of parallel plate capacitances

for verification. The variable shunt capacitances are formed by combination of a

1
number of ordinary parallel plate RF MEMS capacitances. The cut-off frequency is

tuned from C to X band by actuating different combinations of parallel capacitive

bridges. The measurement results agree well with the simulation result.

Keywords: Low-pass filter, MEMS, Tuneable capacitance, Tuneable filter.

2
1. Introduction

Radio frequency micro electromechanical systems (RF MEMS) technology offers an

attractive capability for RF systems, particularly in support of switching and tuning

functions. One such component is a micro electromechanical voltage tuneable capacitor,

which can enable a wide tuning range and high quality (Q) factor. In this paper we

present a design of a tuneable low-pass filter at 7 GHz combining RF MEMS capacitors

and high impedance transmission lines. The low-pass filter is designed by means of the

theory of stepped impedance transmission line filters. The filter can be developed with

surface micromachining without the need for any external inductors. This will give

flexibility in integration and easy fabrication processing.

Compared with solid state varactors, MEMS tuneable capacitors have the advantage of

lower loss and a potentially larger tuning range [1]. The interconnection loss and noise

can also be reduced compared with the use of off-chip solid state RF components.

Among all the MEMS capacitors developed to date, the parallel plate configuration is

the most commonly used. The actuation mechanism is mostly electrostatic as it

consumes very little power during actuation. A parallel plate tuneable capacitor can be

fabricated with surface micromachining techniques. The theoretical tuning range of

such a capacitor is, however, limited to 50%, as the beam will collapse to the downstate

when the gap is less than 2/3 of the initial gap. High tuning range capacitors have been

demonstrated recently. Among those, Zou et al. [1] developed a varactor with a tuning

range of 69.8%. Two steps in the suspended beam are implemented. Another varactor

reported by Dussopt et al. [2] has a tuning range of 90%. The switch fabrication is

3
somewhat complicated with several staircase profiles. Still, these capacitance ratios are

not very high. Improved capacitance ratio (CR) RF MEMS capacitors have been

reported by Nieminen et al. [3] CR 2.25 and Rijks et al. [4] CR 17.

In standard filter design at low frequency, the MEMS varactor is used as a lumped

element with off-chip inductors. At high frequency, some low-pass filter designs were

reported with integrated transmission lines and MEMS switches (see Cai et al. [5] and

Fanget al. [6]). In [5] the tuning of the filter is obtained by changing the transmission

line length using MEMS switches. In [6] planar spiral inductor and shunt capacitor are

used but without any tuning capability. Also, low-pass and band-pass filters integrating

capacitive MEMS switches and short inductive transmission lines have been realised

above 10 GHz [7, 8, 9]. In [7], the tuning was in steps instead of being continuous and

the tuning of the cut-off frequency was obtained by switching between different

combination of inductance and capacitance. In [8, 9] the cut-off frequency was tuned by

changing the inductance using MEMS switching. In this paper we have shown a design

of a low pass filter with a high impedance transmission line and RF MEMS capacitors,

where the tuning is obtained by capacitance tuning. Also it is shown that a continuous

tuning of the frequency can be obtained instead of a stepwise tuning.

We show a design of high capacitance ratio RF MEMS shunt switches in a co-planar

wave guide (CPW) configuration. The methodology for the shunt switch is similar to

the one presented in [1, 3]. Instead of g2=3g1 as mentioned in [3], we proposed a design

with g2>3g1 (sec 3, Fig 2). This design will have better control in tuning the capacitance

than the design mentioned [3], as it has less chance of collapsing to the downstate. The

4
MEMS capacitive shunt switch is used in a stepped impedance low-pass filter instead of

a low impedance transmission line. The series inductor in the low-pass filter is

implemented using a short section of a high impedance transmission line [7]. A

preliminary theory of low-pass filters using transmission lines and MEMS capacitances

is presented in [10, 11]. This filter is much more elegant and compact than filters based

on traditional O/8 long stub and transmission line filters [12]. The designed filter has a

large tuning range, over 60% of its nominal cut-off frequency. Although step impedance

filters don’t have a very sharp roll-off, they can be used for certain applications where

compact size and a wide tuning range are required and less roll-off can be accepted. The

low pass filter, for example can be used in harmonic suppression above the fundamental

frequency in frequency synthesizer. The high tuning range will be very useful in the

systems with multiple frequency band operation.

In this paper, the theory of parallel plate and the high capacitance ratio RF MEMS shunt

switches are described in Section 2. The design of a high capacitance ratio MEMS

switch and simulation of the pull-down voltage and the variation of the capacitance with

actuation voltage are performed with Coventorware“ in Section 3. A third-order 3 dB

ripple chebyshev low-pass filter is described in Section 4. The filter is designed and

optimised with Advanced Design System (ADS“) from Agilent. A comparison between

ADS and Electromagnetic Design System (EMDS) is also presented in the Section 4. A

prototype of the filter is fabricated by our own facility at SINTEF MiNaLab in Oslo,

Norway. In the first prototype, instead of a high capacitance ratio switch, 3 ordinary

parallel plate bridges were used as tuneable capacitances. The tuning of the capacitance

was obtained by actuating different combinations of bridges. The fabrication of the filter

5
is presented in Section 5. The measurement of the filter is presented in Section 6. A

comparison of the measurement result with re-simulation of the filter, taking into

account the effects of process constraints, is discussed in Section 7.

2. Theory of capacitive switch

In this section we discuss the simple parallel plate switch and the high capacitance ratio

switch. Both switches are used in shunt configuration by means of a CPW transmission

line.

2.1 Theory of traditional parallel plate capacitive switch

A MEMS capacitive switch consists of a suspended top plate and a fixed bottom

electrode plate with an overlap area Ae=LeuWe, shown in Figs. 1 and 2. Le and We are

the length and the width of the electrode, respectively. A DC bias voltage, VDC, is

applied between the bridge and the two separate actuation electrodes on the two sides of

the centre conductor as shown in Fig. 1. All electrodes are covered with a dielectric with

thickness td and relative permittivity İr. The initial gap is g0 at the centre of middle

electrode when no DC voltage is applied. When a DC voltage is applied between the

bridge and the actuation electrodes, the initial gap reduces according to the actuation

voltage [13]. If the fringing field effect is neglected, the value of the capacitance

between the suspended bridge and centre conductor for a deflection x of the bridge can

be determined by

6
Ae
C x H 0. (1)
td
( g 0  x) 
Hr

When VDC is applied between actuation electrodes and bridge, an attractive electrostatic

force Fe is generated with the value

1 Ca x VDC
2

Fe (2)
2 ( g  x  td )
a0
Hr

Here Ca(x) is the total capacitance between the bridge and the two actuation electrodes,

x is the distance the top plate has moved from its initial position, VDC is the actuation

voltage and gao is the initial gap between actuation electrodes and suspended bridge.

This electrostatic force tries to pull the top plate down. The beam has a mechanical

spring constant (stiffness) which tries to pull the beam back to the upstate position. For

small deflections the electrostatic and elastic force are equal in magnitude. When the

beam is deflected one-third of the initial gap, the electrostatic force becomes stronger

than the mechanical restoring force. At this point the beam collapses fully to the down

state and this voltage is called the pull-down voltage. Before pull-down or when x<go/3

the capacitance increases to maximum 150% of the original value. So the maximum

tuning range of the electrostatic actuated parallel plate capacitor is 50%.

2.2 Theory of high tuning range capacitive switch

The side view of the proposed high tuning range switch is shown in Fig. 2 (a) according

to the topology mentioned in [3]. Instead of g2=3g1 as in [3], we proposed g2>3g1. It

consists of separate actuation electrodes beside the centre conductor of the CPW

7
transmission line. In this case, the centre conductor is higher than the actuation

electrodes. When g1 is less than or equal to g2/3, the capacitance ratio can be very high

as the downstate capacitance will depend only on the thickness and the dielectric

constant of the dielectric. The theoretical capacitance ratio for a given deflection can be

derived as

td
g1 
C ( x) Hr
(3)
C0 td
( g1  x ) 
Hr

C0 is the initial capacitance when x=0. When x=g1, the beam touches the centre

electrode and the maximum capacitance ratio becomes

Cd Hr
g1 1 (4)
C0 td

Here Cd is the downstate capacitance when x=g1.

From equation 4 we can see that the capacitance ratio can be very high by choosing

appropriate initial gap and maintaining a very small gap between dielectric and top

plate. The proposed design has several advantages over the switch mentioned in the

references [1] and [2]. It can have a very high capacitance tuning range. Owing to the

dielectric the switch will not short-circuit when the beam is fully pulled down. As the

switch has a g1< g2/3, it is less prone to collapse to the downstate. The switch mentioned

in [4] has extra bumps to avoid a full collapse of the beam. Finally, it has potential for a

simpler fabrication process than the switches mentioned in [1, 2, 4], as the top beam is

planar and no step is required. The proposed switch has the potential of higher

capacitance ratio than the work mentioned [3]. The additional process steps required for

8
the proposed switch are the electroplating or sputtering of the centre conductor to make

it higher and planarization of the sacrificial layer. Relevant works on planarization >14,

15@ can be very useful for fabrication of such kind of switches.

3. Design and simulation of the high capacitance ratio RF MEMS shunt switch in

CPW configuration

The proposed high capacitance ratio switch in side and top view is shown in Figs. 2a ,

2b respectively. The switch is designed on a CPW 50 : transmission line. The proposed

fabrication of the switch is done using a standard high resistivity silicon substrate with

280 Pm thickness. The relative dielectric constant of the substrate is 11.9 and the

resistivity is 8000 :-cm. For a 50 : CPW transmission line, the dimensions are 110

Pm/204 Pm /110 Pm (G/Wc/G) (see Fig. 2). The dimensions could also be decreased by

reducing the CPW gap (G) and the actuation electrodes. This will increase the required

actuation voltage, which may be a disadvantage. A reduced g1 will result in a higher

upstate capacitance, which is required for a filter operating at a few GHz.

The dimensions of the proposed beam are as follows, as shown in Table 1. The beam

width at the actuation electrode area We, is 200 Pm. The beam width at the two ends

Wb, is 100 Pm. This will reduce the beam stiffness, which will reduce the actuation

voltage for tuning the capacitance. The gap height at the centre g1 is 0.5 Pm, and the gap

height at actuation electrode area g2 is 2 Pm, which gives a very wide tuning range of

the capacitance ratio even before full contact with the centre electrode. The length of the

actuation electrodes is 100 Pm for each, with a gap of 5 Pm on each side of the

9
electrode for isolation. The total beam length is 750 Pm. This long bridge is chosen to

reduce the actuation voltage. The thickness of the beam is 2 Pm and the proposed

material for the beam is gold. The thickness and length of the beam can be subject to a

trade-off to optimise actuation voltage. Gold is the most frequently-used metal for RF

MEMS switches, owing to its ductility and low Young’s modulus. The thickness of the

transmission line ground and actuation electrodes is 1 Pm. The thickness of the centre

conductor is 2.3 Pm with 0.2 Pm Si3N4 dielectric on top. The thickness of the dielectric

can be reduced, in case it is necessary to have higher capacitance at downstate.

The simulation of the switch with respect to displacement and capacitance change with

actuation voltage is done in Coventorware“. The displacement from the top position of

the beam at centre and capacitance variation with actuation voltage are shown in Fig. 3.

From this it can be seen that the top plate deflects fully to the centre conductor at 12.8

volts. In the Coventorware architect simulation the residual stress is not included. The

tensile stress will increase the actuation voltage for the same deflection. The simulation

of capacitance vs actuation voltage including residual stress can be performed using

CoventorWare analyser. The capacitance in the upstate with zero actuation voltage is

around 690 fF. The upstate capacitance value is high, because of the large overlap area.

Before the beam is in contact with the centre electrode, the capacitance reaches 2.8 pF

at 11.8 volts, which gives a capacitance ratio of 4. When the beam is in contact with the

centre electrode at 20.5 volt actuation voltage, the capacitance is 15 pF which gives a

capacitance ratio of more than 20. Fringing fields are not included in the Coventorware

simulation. The real capacitance in upstate can be higher than the parallel plate

10
capacitance to some extent owing to fringing fields [15, p.91]. As the upstate

capacitance is very high for our design the percentage of fringing capacitance will be

small. To include the fringing filed effect, a CoventorWare Analyzer MemElectro

simulation is performed with a very fine mesh. The capacitance at upstate is found to be

702 pF. The fringing capacitance is only a 2% higher than the Architect result. The

fringing field can also increase the downstate capacitance to some extent. In downstate

however, the capacitance will be reduced owing to roughness.

4. Theory, design and simulation of the proposed filter

In this section the theory and design of the stepped impedance low-pass filter are

described.

4.1 Theory of the stepped-impedance filter

It is well known [12] that if the length of the transmission line is short (ȕl< ʌ/4) and the

characteristic impedance is high the transmission line can be represented by the

equivalent circuit shown in Fig. 4 where the reactance is given by

XL Z 0 E" (5)

Here the characteristic impedance of the line is Z0, the length is l and the propagation

constant is ȕ.

So a series inductor can be represented by a short length of high characteristic

impedance transmission line. In the lumped elements filter model, the shunt capacitor

will be replaced by an RF MEMS capacitor which will give the cut-off frequency tuning

capability.

11
4.2 A third-order low pass filter topology design and simulation in ADS and

comparison with EMDS (Electromagnetic Design System)

A third-order low pass filter topology is shown in Fig. 5. The basic filter elements are

extracted from the Chebyshev equal ripple low pass filter prototype [10, 11, 12]. The

normalized filter elements for a 3 dB ripple Chebyshev are given by g1=3.3487,

g2=0.7117, g3=3.3487, and g4=1.00 is the load impedance [12]. A ripple of 3 dB is used

to give a sharper roll-off above the cut-off frequency. We have chosen a nominal cut-

off frequency at 7.0 GHz. With a maximum transmission line impedance of 80: the

length of the transmission line becomes 25. 5° and the nominal capacitance value

becomes 1.52 pF when the cut-off frequency is 7 GHz. After optimization of the filter

cut-off frequency in ADS the nominal shunt capacitor value becomes 1.1 pF and the

length of the 80 : line becomes 1467 Pm (30.7˚) for the desired responses. The

substrate parameters used for simulation are mentioned in Section 3. For an 80: CPW

transmission line, the dimensions are 160Pm/53Pm/160Pm (G/W/G). A higher value,

such as 100 : or more, can be chosen for the impedance of the series inductance. This

will make the transmission line very narrow. One has to keep in mind the process

parameters, because if the transmission line is thin, it will introduce extra insertion loss

in the pass band also.

We have simulated the designed filter in ADS using the transmission line and the

capacitance values obtained from CoventorWare. For verification we have simulated the

filter with high capacitance ratio bridges in 3D electromagnetic simulation from

Electromagnetic Design System (EMDS). A top view of the proposed filter is shown in

12
Fig. 6. The dimensions of the bridge are presented in Sections 3. The width of the centre

conductor for the 50 : transmission lines is 204 µm with a gap of 110 µm. The

thickness of the centre electrode dielectric is (Si3N4) 200 nm. A relative permittivity of

7 is used for the nitride to calculate the capacitance. The shunt capacitance is varied

from 0.69 pF (when the bridges are at the initial position) to 2.81 pF (when the bridges

are at 0.4 µm down from the initial position). Here it is assumed that the bridges remain

parallel to the bottom electrode when they deflect from the initial position. The

simulation result and comparison between ADS and EMDS are shown in Fig. 7.

In the figure. BD means deflection of the bridge from its original position. The red line

is the simulation result in EMDS and the black line is the simulation in ADS. From Fig.

7 it can be seen that the cut-off frequency is tuned from 9 GHz to 4.7 GHz when the

bridges deflects from the initial position to 0.4 µm. The tuning of the filter is 4.3 GHz

which is 60% of the nominal cut off frequency 7 GHz. Also the simulation in ADS

closely matches the 3D electromagnetic simulation in EMDS. The insertion loss is

slightly high in the pass band as a 3 dB ripple Chebyshev filter element is used in the

design. It is also observed that for higher capacitance value the insertion loss increases.

As the capacitances are in shunt, for a higher capacitance values the shunt impedance

will be lower and will increase the loss in pass band. For clarity the return losses are not

included in the Fig. 7. In the pass band a higher return loss is present. There are few

reasons for this. A 3 dB ripple chevyshev filter approach is used for the design. Also

only capacitance tuning is used for the variation of cut-off frequency. In order to

improve the return loss in pass band, a 0.1 dB or 0.5 dB ripple chevyshev or flat

butterworth filter approach can be used. Capacitive tuning with an inductive inverter

13
can be used to reduce the return loss for the entire tuning range [16, 17]. Simultaneous

tuning of both capacitance and inductance with stub can also be used to reduce return

loss and insertion loss for the entire operating frequency range of the filter [18].

5. Fabrication of the filter

A top view of the fabricated filter is shown in Fig. 8. The fabrication was done at

SINTEF MiNa lab in Oslo. Since the fabrication of RF MEMS is new to this facility, a

simplified version of the filter was preferred, using traditional parallel plate capacitors.

This will increase the probability of successful fabrication. Instead of high capacitance

ratio switches, a combination of 3 parallel bridges is used for each capacitance as shown

in Fig. 8. Each shunt capacitance is a combination of two 12 µm wide (on the two sides)

and one 15 µm wide (middle) shunt bridge. The variation of the capacitance can be

obtained by actuating individual bridges, or combining one or more bridges. The width

of the centre conductor for the 50 : transmission lines is 140 µm with a gap of 80 µm.

This is different from the dimensions mentioned in Section 3. Because of the non-

uniform beam approach, a wider centre conductor is required to obtain the required

capacitance. The thickness of the centre electrode and the actuation electrode dielectric

is (Si3N4) 220 nm. The estimated total roughness of the dielectric and gold is to be 15

nm. Aluminium is planned for use as a transmission line.

A four-mask process was used to fabricate the proposed filter. The cross-section view of

the process flow of the capacitive bridges is shown in Fig. 9. A 280 µm thick silicon

wafer with 4-8 k:-cm (specified by supplier) resistivity was used for the fabrication. A

500 nm thick oxide was grown on the silicon wafer to obtain lower loss (A). Then a 500

14
nm thick Al was deposited by sputtering and patterned by wet etching (B). A 220 nm

thick Si3N4 was deposited as dielectric by plasma-enhanced chemical vapour deposition

(PECVD) and patterned by dry etch (C). HiPR 6517 photoresist was used as a sacrificial

layer. The resist thickness was 2.55 µm and patterned by standard lithography and

special baking methods (D) as described in detail in [19]. Then 1.1 µm thick gold was

sputtered and patterned by wet etch (E). The gold sputtering was done with very low

power and a sputtering pressure of 20 mTorr to obtain a low tensile stress [20]. When

the gold was sputtered, the transmission line was also sputtered with 1.1 µm additional

gold on aluminum. Low tensile stress ensures low-pull down voltage and gives better

reliability. Finally the switches and filters were released with our developed methods

and dried in a critical point dryer [19].

6. Fabricated low-pass filter and measured results

A SEM image of the fabricated low pass filter is shown in Fig. 10. An enlarged image at

the capacitance region is also shown in Fig. 11. As shown in Fig. 8 the actuation

electrodes are separated from the RF signal electrode, and each bridge has its own set of

actuation electrodes. In Fig. 11 the centre narrow line is the 80 Ohm transmission line,

which is equivalent to a series inductor. The capacitors are at both ends of the 80 Ohm

transmission line, shunting a 50 Ohm transmission line. In the figure, the three parallel

bridges of shunt capacitors can be seen. The extended 50 Ohm transmission lines are

used for measurement purpose with a CPW probe.

15
The fabricated filters were measured in a vector network analyser. The filter

characteristic was measured up to 30 GHz. Standard Line-Reflect-Reflect-Match

(LRRM) methods were used for calibration of the network analyser. The measured filter

S parameters are shown in Fig. 12. We have measured the filter characteristics with one

bridge down (15 µm wide centre bridge), green line, and 2 bridges down (15 µm and 12

µm wide bridges), purple line. Due to lack of actuation facility, we could not actuate all

three parallel bridges together.

7. The re-simulation in the light of fabrication constraint and comparison

After fabrication the filters were inspected in an optical interferometer to measure the

actual width and height of the bridge. The actual initial heights of the bridges were 2.25

µm and the width was reduced on average 1 µm from each side owing to over-etching.

Gold is usually smooth and it was found that the average roughness of the gold was 7

nm. The Si3N4 average roughness was 9 nm. But the main contribution to the roughness

was from aluminum. This is because of the high temp (300 q C) Si3N4 deposition. From

an atomic force microscope (AFM) measurement, it was found that the height of the

aluminum bumps was about 80 nm. As the nitride was deposited on aluminum, the total

roughness for the downstate capacitance was 96 nm. Considering the over etch and

roughness, the downstate capacitance becomes 0.38 pF for the 15 µm bridge (13 um

real width). The filter was re-simulated with the above parameters. The simulation was

16
compared with measurement results of the fabricated filter as shown in Fig. 13. The

measured results agree better with the simulated results, given the effect of over-etching

and roughness. There is a slightly higher cut-off frequency in the measured filter. This

can happen because of non-planarity of the bridge at downstate, which will further

reduce the downstate capacitance.

By using same roughness as mentioned earlier, the estimated downstate capacitance

becomes 0.30 pF for the 12 µm bridge (10 µm real width). The total capacitance at

downstate becomes 0.68 pF when two bridges are down. The simulated and measured

results for this state were compared and are shown in Fig. 13. From this figure it can be

seen that the measured filter again has a higher cut-off frequency for the reason

discussed above. Also, the narrower bridge will have relatively more over-etching and

less planarity.

8. Conclusion

We have presented a tuneable low-pass filter design with transmission line and MEMS

capacitance. A high capacitance ratio shunt switch is proposed, which can be used for

continuous tuning of the cut-off frequency. A simulation of the filter with high

capacitance ratio switch is performed in EMDS and compared with ADS. The results

agree with each other very closely. A simpler parallel plate capacitance was investigated

for the first prototype fabrication and comparison of the filter performances. The tuning

17
of the cut-off frequency is obtained by actuation of several parallel bridges. The

measured results agree well with the simulated result considering the fabrication effects.

The filter roll-off is not very sharp, but can be improved with the use of high impedance

transmission line and a higher order of the filter. But this may have some drawbacks.

As only the capacitance is used for tuning and the inductance is fixed, higher order

filters will have reduced tuning range. Also it will be complicated to tune all the

capacitor simultaneously. High impedance transmission line will increase the insertion

loss significantly as the center line will be very narrow. The filter is very compact and

has very good control of the tuning frequency.

Acknowledgement

The authors are grateful to the Norwegian Research Council for sponsoring the work

through the SMiDA project (No 159559/130) and the IRRFT project (No 159259/I40).

The authors also thank MiNa Lab personnel for their support in the clean room.

18
Reference:

[1] J Zou., C. Liu, J. Schutt-Aine, , J. Chen and , S. Kang, ‘Development of a wide

tuning range MEMS Tunable capacitor for Wireless Communication Systems’,.

International Electron Devices Meet. Tech. Digest 2000, 403-406.

[2] L. Dussopt and G. Rebeiz, High-Q Millimeter-Wave MEMS varactors: ‘Extended

Tuning range and Discrete position Designs’, 2002 IEEE MTT-S Digest, 1205-1208.

[3] H. Nieminen, V. Ermolov and T. Ryhånen: ‘Microelectromechanical Capacitor

with wide tuning range’, IEE Electronics letters, Vol. 37, (24), (2001)1451-1452.

[4] T. S.M. Rijks, J. T. M. V. Beek, P. G. Steeneken , M. J. Ulenaers, J. D. Coster

and R. Puers: ‘RF MEMS tunable capacitor with large tuning ratio’, 17th IEEE

International Conference on. (MEMS) Micro Electro Mechanical Systems, 2004 777-

780.

[5] M. Cai, X. Guo, Y. Li, L. Liu, Z. Lai: ‘Design and modeling of reconfigurable

MEMS low pass filter, Solid-State and Integrated Circuit Technology’, 2006. ICSICT

'06. 8th International Conference on, 563-565 .

[6] J. Fang, , Z.W. Liu, , Z.M. Chen, L.T. Liu,; Z.J Li,.: ‘Realization of an Integrated

Planar LC Low-Pass Filter with Modified Surface Micromachining Technology’, IEEE

conference on Electron devices and Solid-State circuits, 19-21 Dec. 2005, 729-732.

[7] D. Peroulis, S. Pacheco, K. Sarabandi and L. P.B. Katehi: ‘Tunable Lumped

Components with Applications to Reconfigurable MEMS Filters’, Microwave

Symposium Digest, 2001 IEEE MTT-S, Volume 1, May 2001, vol.1 341 - 344

[8] S. Lee, J. Kim, Y. Kim and Y. Kwon, ‘Millimeter-Wave MEMS tunable low pass

filter with reconfigurable series inductor and capacitive shunt switches’, IEEE

Microwave and Wireless Components letters, vol 15, (10), (2005), 691-693.

19
[9] J. Park, S. Lee, J. Kim, H. Kim, Y. Kim and Y. Kwon: ‘Reconfigurable

millimeter-wave filters using CPW-based periodic structures with novel multiple-

contact MEMS switches’, IEEE Journal of Microelectromechanical Systems, Vol. 14,

(3), (2005), 456-463.

[10] S. C. Saha and T.Sæther: ‘Modeling and Simulation of Low Pass Filter using RF

MEMS Capacitance and Transmission line’, Proceeding IMAPS Nordic 2005 pp 155-

159.

[11] S. C. Saha, U. Hanke, T. Sæther: ‘Modeling, Design and Simulation of Tunable

Band Pass Filter using RF MEMS Capacitance and Transmission line’, SPIE

International Symposium, Microelectronics, MEMS and Nanotechnology, Proc. of

SPIE Vol. 6035, Microelectronics: Design, Technology and Packaging II, 11-14

December, Brisbane, Australia, 2005.

[12] D. M. Pozar, ‘Microwave Engineering’, Ch 8, Second Edition, John Willey &

Sons, 1998.

>13@ G. M. Rebeiz: ‘RF MEMS Theory, Design and Technology’, Ch 2 and 3, 2nd

Edition, John Willey & Sons, 2003.

>14@ M. Ohnmacht, V. Seidemann, S. Buttgenbach, Microcoils and microrelays- an

optimal multilayer fabrication process, Sensor and Actuators, vol 83, (2000), 124-129,

[15] L. Wang, A. Cui, J.-S. Hong, E. P. McErlean, R. B. Greed, D. C. Voyce,

Fabrication of high power RF MEMS switches, Journal of Microelectronic Engineering,

Article in Press, doi 10.1016/j.mee.2006.01.067.

[16] K. Entesari, G. M. Rebeiz, A differential 4-bit 6.5-10 GHz RF MEMS tunable

filter, IEEE Transaction of Microwave Theory and Techniques, vol 53, (3) part 2,

(2005) 1103-1110.

20
[17] G. L. Matthaei, E. Young, and E. M. T. Jones, Microwave Filters, Impedance-

Matching Networks, and Coupling Structures. Norwood, MA: Artech House, 1980.

[18] B. Pillans, A. Malczewski, R. Allison, J. Brank, 6-15 GHz RF MEMS tunable

filters, 2005 IEEE MTT-S, pp 919-922.

[19] S. C. Saha, H. Sagberg, E. Poppe, G. U. Jensen, T. A Fjeldly, and T. Sæther,

Tuning of resist slope with hardbaking parameters and release methods of extra hard

photoresist for RF MEMS switches, Journal of Sensor and Actuators A: Phys. vol. 142

(2) (2008) 452-461,.

[20] S. C. Saha, H. Sagberg, E. Poppe, G. U. Jensen, and T. Sæther ‘Metallization

scheme and release methods for fabrication of RF MEMS switches’, Proceedings of

33rd Micro- and Nano- Engineering conference (MNE 2007), September 2007,

Copenhagen, Denmark.

21
List of Table and figures:

Table:

Table 1: The dimensions of the high capacitance ratio switch, shown in Fig. 2.

Figures:

Fig. 1: Side view of a traditional parallel plate capacitive switch in CPW configuration

with separate actuation electrodes.

Fig. 2: Side view (a) and top view (b) of the proposed high capacitance ratio switch

Fig. 3: The displacement vs. actuation voltage at the middle of the beam (a) and the

capacitance vs. actuation voltage (b).

Fig. 4: Equivalent circuit for short length and high characteristic impedance

transmission line

Fig. 5: Third order low pass filter topology

Fig. 6: The top view of the proposed filter with high capacitance ratio switch

Fig. 7: The variation of cut-off frequency with changing shunt capacitance, with a

comparison in EMDS.

Fig. 8: The top view of the proposed low pass filter with actuation mechanism.

Fig. 9: The 4 masks process flow of the low-pass filter.

Fig. 10: A SEM image of the fabricated low-pass filter (top view). Three narrow bridges

at both sides represent the tunable capacitance. Chip size (2.3 mm×1 mm)

Fig. 11: An enlarged image of the low-pass filter at the capacitance area

Fig. 12: The measured filter characteristic with 1 bridge down (15 um), pink line and 2

bridges (15 um and 12 um) down, green line.

22
Fig. 13: Comparison of the filter performances between re-simulated (blue line) and

measured results (black line).

Table 1:

Dimensional Value ( µm)


parameters
We 200
Wb 100
Le 204
Lb 750
We 204
La 100
G 110
g1 0.5
g2 2
tb 2
td 0.2

23
Fig. 1

Fig. 2a

Fig. 2b

24
Fig. 3a

Fig. 3b

Fig. 4

25
Fig. 5

Fig. 6

26
Fig. 7

Fig. 8

27
Fig. 9

28
Fig. 10

29
Fig. 11

30
Fig. 12

Fig. 13

31
Tunable Band-Pass Filter at C band with RF MEMS Capacitance and
Transmission line

Shimul Chandra Saha1, Ulrik Hanke2, Håkon Sagberg3, Tor A. Fjeldly1 and Trond Sæther1
1
Department of Electronics and Telecommunications (IET), Norwegian University of Science and
Technology (NTNU), NO-7491 Trondheim, Norway (e-mail: [email protected]; [email protected],
[email protected])
2
Institute for Microsystems Technology, Vestfold University College, 3103 Tønsberg, Norway (e-
mail: [email protected] )
3
Department of Microsystems and Nanotechnology, SINTEF ICT, PO Box 124, Blindern, NO-0314
Oslo, Norway (e-mail: [email protected]) .

Has been submitted to Journal of European Microwave Association (EuMA)


Is not included due to copyright
The measurement of S parameters of the RF MEMS switches
fabricated in SINTEF MiNa Lab

A MEMO on the measurement results of the fabricated CPW RF MEMS switches.


The measurement of S parameters of the RF MEMS switches fabricated in SINTEF
MiNa Lab

1. Introduction:
The S parameter of the RF MEMS capacitive shunt bridge, at upstate and downstate is measured.
Switches were fabricated with transmission line and dielectric in SINTEF MiNa lab in Oslo. The
released switches looked completely clean and fully suspended when inspected in a optical
interferometer from WYKO. A vector network analyzer from Agilent, with a coplanar waveguide
(CPW) probe was used for the S parameter measurements. The SINTEF ICT department at Trondheim
has developed a bipolar actuation source, which was used for reliability tests. A DC actuation is also
used to measure the S parameters at upstate and downstate. The S parameters in upstate are close to
the expected results from the simulation. In downstate the isolation is lower than the expected value,
due to a reduced capacitance. It is found that if the actuation pulse width is reduced, the number of
operating cycles increase and vice versa. The bridge sticks to the downstate after a certain number of
cycles. The first version of the switch (bridge thickness ~ 1.1 µm) successfully operated up to
~300.000 cycles. The reason of stiction can be either charging at the electrodes (most probably) or
mechanical failure of the suspended beam. A second version with thicker gold (bridge thickness ~ 1.2
µm) operated successfully more than 1.000.000 cycles.

2. The dimensions of the switches:


Several versions of the switch were designed and processed. The switch with title MAIN is fabricated
as a reference design. An optical image of the fabricated and released RF MEMS switch (SHORT) is
shown in Figure 1. The silicon is slightly over-etched (2.6 µm) during etching of tungsten transmission
line (SF6+C4F4). A SEM image of the switch (MAIN) is shown in Figure 2. Some other designs, with
increased length and width are also included to test different parameters, like pulldown voltage, S
parameters, etc. The switches were fabricated on 50 ohm CPW transmission line in shunt
configuration. The center conductor width is 100 µm and the gap between the center conductor and
ground is 62 µm as shown in Figures 1 and 2. The thickness of the dielectric is 210 nm. The initial gap
of the switch is ~2.25 µm. A slight reduction has occurred (coated thickness was 2.50 µm) due to
baking and other high temp processes. The tungsten transmission line thickness is a500 nm with an
additional layer of gold of 1.1 µm. The gold covers the entire tungsten transmission line except the
dielectric area and the region below the suspended beam. The gold bridge thickness is 1.1 µm. The
dimensions and measured pull-down voltage of the shunt bridges are shown in Table 1 below.
Table 1: The dimensions and pull-down voltage of different version of the shunt switch fabricated at
SINTEF MiNa lab (first round).
Title Length of the Width of the Thickness of the Pull-down
suspended bridge suspended bridge gold bridge voltage (V)
(µm) (µm) (µm)
MAIN (1, 2) 300 80 1.1 17
MAIN (with CPW 300 80 1.1 18.8
ground shorted)
SHORT 250 80 1.1 17.6
LONG 500 80 1.1 15.4
V LONG 500 80 1.1 15.5
WIDE BRIDGE 300 120 1.1 18.8
WIDE V LONG 500 120 1.1 14.9

Figure 1: An optical image of the capacitive shunt switch fabricated in SINTEF MiNa lab Oslo.
Figure 2: A SEM image of the RF MEMS switch (MAIN) fabricated at SINTEF MiNa Lab.

3. Measurement results:
The S parameters of the switches fabricated in 1st round are measured with a vector network analyzer
HP 8510 from Agilent. The standard calibration method LRRM (line-reflect-reflect-match) is used for
calibration. The measured S parameters of some of the switches are shown in Figures 3-6.
Figure 3: The insertion loss and return loss of the switch MAIN (1) in upstate

Figure 4: The isolation of the switch MAIN (1) in downstate


Figure 5: The insertion loss and return loss of the switch WIDE BRIDGE in upstate

Figure 6: The isolation of the switch WIDE BRIDGE in downstate

For comparison the dimensions and the simulated results of the proposed switch at paper 3 is
mentioned. The nominal switch length is Lb=600 Pm. The initial gap height of the beam is g0=3.75
Pm. The dielectric material is silicon nitride with a relative permittivity of 8 and a thickness of 1000
Å. Gold is chosen as the membrane material with a thickness of tb=1 Pm. The width of the beam is Wb
= 50 Pm. The simulated S parameters are shown in Figures 7 and 8.

Figure 7: The simulated insertion loss (solid) and return loss (dotted) in upstate (paper 3).

Figure 8: The simulated isolation in downstate with 50 Å roughness (thick line) and 100 Å roughness
(thin line) (paper 3).
4. Result analysis:
The measured insertion loss at upstate for the MAIN switch (version 1) is less than 0.5 dB up to 30
GHz. The return loss is -16 dB at 30 GHz. The insertion loss and return loss are slightly higher than
the simulated result in the paper 3. This is due to a higher upstate capacitance than the simulated value.
There are two main reasons for higher upstate capacitance: a) the width of the bridge is 80 µm instead
of 50 µm in the simulation and b) the initial gap is 2.25 µm instead of 3.75 µm used in simulation.
Using return loss and back calculation the upstate capacitance value was found to be 50 fF. The
insertion loss of the switch will be reduced to some extent if we deduct the loss of transmission line
from the measured result. The isolation is much lower than the simulated value. At 10 GHz the
isolation is -7 dB (fig 4) compared to -15 dB (fig 8) simulated value. The downstate capacitance is
reduced significantly although the width of the beam is made larger. There are some reasons for the
reduced capacitance; a) The dielectric thickness is 220 nm instead of 100 nm used in simulation, b)
The roughness of the TL, dielectric and gold may reduce the downstate capacitance (~ 8-10 nm), c)
The bridge may not be completely flat in downstate, which can be the dominant reason in our case and
d) due to over etching, the beam width is reduced slightly. From back calculations using the formula
mentioned [4], the downstate capacitance becomes 1.43 pF. In the simulations the isolation increased
up to the resonant frequency at 25 GHz (fig 8) and then decreased again at higher frequencies. In the
measured result we do not see this effect since the resonant frequency is higher than 30 GHz. This is
due to reduced capacitance and reduced inductance. The inductance value will de reduced due to wider
and shorter beam. From the measurement a capacitive ratio of 28 was obtained for our switch. From
the stress test a stress value of ~40 MPa was found in the sputtered gold. The measured pull-down
voltage is lower than the expected value. This may happen, due to the thinner bridge at the rising and
falling edges at the middle of the bridge, as can be seen in figures 1 and 2. A wider bridge (WIDE
BRIDGE) with a width of 120 µm is also measured. The measured result is shown in figure 5 and 6.
As expected the insertion loss and return loss both increased due to higher upstate capacitance. The
isolation is increased to some extent due to a higher downstate capacitance. The isolation is -10 dB at
10 GHz. Still this is lower than the expected isolation.

5. Reliability test:
The bipolar actuation source used for the RF MEMS switch reliability test is shown in Figure 9.
Initially the pulse amplitude is slightly higher than pull down voltage. Then the voltage amplitude is
reduced to the hold down voltage. As mentioned in the introduction the switches stick after 300.000
cycles with bipolar actuation. The actuation frequency is 250 Hz. The active actuation part, when the
peak amplitude of the voltage is equal to or greater than pull down voltage is 100 µs. Then a reduced
voltage was applied to keep the bridge down (called hold down voltage). After that the actuation
voltage is brought back to zero to release the beam in upstate. In order to improve the reliability and
find the cause of stiction, a second round of fabrication is performed. The fabrication parameters are
same for the first round except a thicker (1.2 µm instead of 1.1 µm) sputtered gold bridge. The
reliability test shows that the switch can operate more than 1.000.000 cycles before stiction. The pull
down voltage for the MAIN switch was 30 volt. Charge stiction is very common in the capacitive
switching and this can be responsible for our switch also. A better test setup with nitrogen
environment is planned for later reliability test. Also an additional layer of gold at anchor is planned
for fabrication of later version of the switch. This may increase the reliability keeping pull down
voltage at reasonable limit.

Figure 9: The bipolar actuation signals used for the RF MEMS reliability test
6. Conclusions
The work provides some key achievement. These are the first ever RF MEMS switches to be produced
in Norway. The switches have a reasonable low pull-down voltage compared with the present state-of-
the-art switches reported elsewhere. The switches employ rather simple beam structures. The
capacitance ratio of the switches is also very promising compare to recent work published in
elsewhere, considering Si3N4 used as a dielectric. The number of successful operating cycles before
stiction exceeds more than 1.000.000 cycles. The operating cycles may increase when we do the test in
the nitrogen environment. The reliability of the switch is very promising compare to the present state
of art switches. The work will provide strong motivation and encouragement for future work on RF
MEMS in Norway and international community. Based on the MEMO and the second version of the
switch, an article is prepared and accepted in MEMSWAVE 2008, Greece and added as Paper no 15.
A capacitive shunt RF MEMS switch for coplanar waveguide
transmission lines

Bengt Holter1, Karsten Husby1, Håkon Sagberg2, Geir U. Jensen2,


Ulrik Hanke3, and Shimul Chandra Saha4 , Trond Sæther4

1
SINTEF ICT, Communication Systems, O. S. Bragstads plass 2C, 7465 Trondheim, Norway, +47
73593000
2
SINTEF ICT, Microsystems and Nanotechnology, Gaustadalleen 23, 0373 Oslo, Norway, +47
22067300
3
Vestfold University College, Microsystem Technology, 3103 Tønsberg, Norway, +47 33031000
4
Norwegian University of Science and Technology, 7491 Trondheim, Norway, +47 73595000

In press, MEMSWAVE 2008, 30th June -3rd July 2008, Crete, Greece.
A capacitive shunt RF MEMS switch for coplanar
waveguide transmission lines
Bengt Holter1, Karsten Husby1, Håkon Sagberg2, Geir U. Jensen2,
Ulrik Hanke3, Shimul Chandra Saha4, and Trond Sæther4

1
SINTEF ICT, Communication Systems, O. S. Bragstads plass 2C, 7465 Trondheim, Norway, +47 73593000
2
SINTEF ICT, Microsystems and Nanotechnology, Gaustadalleen 23, 0373 Oslo, Norway, +47 22067300
3
Vestfold University College, Microsystem Technology, 3103 Tønsberg, Norway, +47 33031000
4
Norwegian University of Science and Technology, 7491 Trondheim, Norway, +47 73595000

Abstract — In this paper, performance results are


presented for the first RF MEMS switch ever to be produced
in Norway. It is a shunt capacitive switch integrated in a
coplanar waveguide configuration. The switch uses a bipolar
(square wave) bias voltage for both actuation and hold-
down, and is designed to operate in the frequency range 10-
32GHz. The fabrication process has been developed and
performed at the Micro and Nano Laboratory at SINTEF,
which has the only independent complete silicon processing
line in Norway. In the targeted frequency range (10-32GHz),
the measured insertion loss is between -0.17 and -0.4dB, the
return loss between -23 and -12dB, and the isolation is
between 7.5 and 20dB. The achieved capacitance ratio of the
switch is 25. Some stiction problems have been observed for
either a too high pull-in voltage or a too high hold-down
voltage.

Fig. 1. The SINTEF RF MEMS switch


I. INTRODUCTION (SEM) image of the fabricated switch is depicted in Fig. 1
RF MEMS (Radio Frequency Micro-ElectroMechanical (top view). When the bias voltage is applied, an
Systems) technology is currently regarded as a key electrostatic force causes the bridge to bend down onto a
enabling technology for future development of wireless thin dielectric layer that covers the t-line beneath the
communication systems due to superior performance and bridge. In this down-state position, the bridge connects
flexibility compared to existing technologies. In the t-line to ground (through a capacitance), realizing an
particular, RF MEMS components may be applied in RF short for microwave frequencies. When the bias
future cognitive radio systems to enable reconfigurable voltage is removed, the electrostatic force is relaxed,
RF front-ends such as flexible filtering, adaptive causing the bridge to return to its original up-state
matching networks, and more efficient antenna designs. position. In the literature, detailed analysis on modeling
For the last four years, SINTEF has conducted research and design of CPW shunt switches are given in [1]-[3].
within a project entitled Integrated reconfigurable radio
front-end technologies (IRRFT), to gain experience in
II. FABRICATION PROCESS
using RF MEMS components and devices in
reconfigurable RF front-ends. One of the main objectives The switches are fabricated on high-resistivity (4-8
was to develop knowledge and experience to design and kΩcm) Si with 500nm of thermal silicon dioxide. The
process an in-house RF MEMS switch, where the focus switch bridge measures (80×300)ȝm, and is made of
was on optimizing the probability of processing success sputtered gold (Au) on top of a 2.5ȝm thick sacrificial
rather than achieving ambitious electrical specifications. photoresist layer. The 1.2ȝm thick Au bridge has a
In this paper, the fabrication process and some selected relatively low tensile stress (30-60 MPa), and ‘climbs’ the
performance results of the developed switch are sloping sidewalls of the patterned photoresist. Below the
presented. It is the first RF MEMS switch ever to be bridge, the bottom electrode is made of sputtered tungsten
produced in Norway, and it has been fabricated at the (W) covered by a 200nm thick plasma-enhanced chemical
Micro and Nano Laboratory at SINTEF. vapour deposited (PECVD) silicon nitride layer (Si3N4).
The switch is a coplanar waveguide (CPW) capacitive Only four photomask levels were used, defining (1)
shunt switch, where the suspended MEMS bridge is The W for electrode and CPW bottom layer (2) The Si3N4
actuated by applying a bipolar square wave bias voltage layer covering the bottom electrode (3) The patch of
between the bridge and the center conductor of the CPW sacrificial photoresist that creates the bridge-to-dielectric
transmission line (t-line). A scanning electron microscope gap, and (4) The Au bridge and CPW upper layer. An
additional layer of metal (Au) is often included for for good mechanical as well as electrical properties. The
strengthening the bridge anchors and improving the t-line correct sloping sidewalls were achieved with a
RF properties. Such a layer would probably have had a combination of proximity exposure and careful fine-
noticeable influence on the pull-in voltage, at least, but tuning of hard baking parameters (temperature and time).
the SINTEF switch operates remarkably well without that We used a standard DNQ-novolac positive photoresist
layer. For implementation of circuits based on these (HIPR 6517), which is spin coated at 2000rpm and
switches, an additional step/mask for polysilicon bias exposed at a gap of 25ȝm. The optimum hotplate
lines is used. hardbake conditions were found to be 115oC for 300s [4].
The temperature during subsequent processing steps must
A. Bottom electrode
be kept well below the hard baking temperature, in order
The bottom electrode consists of a 100ȝm wide and to prevent excessive baking of the photoresist.
500nm thick stripe of sputter-deposited W with a 12-
D. Bridge
15nm Ti adhesion layer. Patterning is done by reactive
ion etching (RIE). A refractory metal, W withstands the A 1.2ȝm Au layer, with an adhesion layer of 10nm
subsequent high-temperature processing without forming NiCr, is deposited using low-power sputtering at
hillocks as, for instance, aluminium would do. A smooth relatively high pressure (20mTorr). The NiCr/Au is
electrode surface is essential for a high down-state deposited partly on W (CPW lines), and partly on
capacitance. Unlike e.g., Al, W has no hillock formation. photoresist (bridges). A low sputtering power ensures that
Since the W electrode also serves as a section of the t- the sacrificial photoresist is not much affected, and the
line, a sufficiently thick electrode is needed for sputtering pressure was tuned to achieve a low tensile
transmission quality. However, two other effects call for a stress in the Au film. The resulting biaxial stress of the
thin electrode layer: First, the subsequent deposition steps suspended Au bridge was characterized to lie in the range
(nitride deposition, sacrificial resist and Au sputter) are 30-60MPa using test structures such as Guckel rings.
(at least partly) conformal, so the edges of the electrode The Au film was patterned using standard
will create steps in the Au bridge, as illustrated in Fig. 2, photolithography and a potassium iodide (KI) solution.
which also shows the shape of the resulting bridge before The sacrificial photoresist was removed using O2 plasma
and after release. Second, sputtered W has a large internal and Shipley 1165 microposit remover, and then the NiCr
compressive stress (~1GPa); thus a thick W layer would layer underneath the Au in the bridge was etched. Finally,
produce wafer curvature and also adhesion problems the switches were dried with supercritical CO2 drying to
(peeling). As such, the selected thickness of 500nm for avoid stiction. Images of the device are shown in Fig. 3.
the bottom electrode is a compromise.

Fig. 3 SEM images of the fabricated switch in its idle (top)


Fig. 2 Surface profile of the bridge in three states: Pre- and actuated (bottom) states.
release (dotted curve), post-release (solid curve), and actuated
(dashed curve). The heights are relative to the top of the W-
E. Mechanical properties and switch operation
NiCr-Au stack. Pre-release data were obtained with optical
profilometry on surfaces partially covered with photoresist, and
the profile has been scaled and offset to obtain a good fit to the
A bias voltage applied between the bottom electrode
post-release data. and the bridge will cause the latter to move towards the
former. The spring constant is often calculated under the
B. Dielectric layer assumptions of a fixed-fixed bridge and that there are no
A 200nm thick layer of Si3N4 serves as the dielectric steps in the bridge. These assumptions do not apply in our
layer in the capacitive switch. The Si3N4 is deposited case. When released, the bridge contracts and the parts
using PECVD at 300oC, and a (180×140)ȝm patch is close to the anchors deform slightly. This creates a sag or
patterned by RIE. The relative permittivity of the Si3N4 is curvature of the released bridge, as seen in Fig. 2. Also,
around 8, but the surface roughness of the Si3N4 as well the bridge steps on each side of the bottom electrode
as that of the underside of the Au bridge results in a lower result in local curvature changes and a softening of the
’effective’ permittivity. bridge. A several ȝm thick additional top metal layer
would reduce the first sag, but some electrode induced
C. Sacrifical photoresist sag will always be present. However, the SINTEF switch
While achieving vertical photoresist sidewalls is the exhibits an overall good performance. The measured pull-
goal of many high-resolution photoprocesses, the switch in voltage is 30V, and the down-state and up-state
requires a process resulting in roughly 45o sloping capacitance are 1.2pF and 49fF, respectively.
sidewalls so that the Au slope/wall remains thick enough
III. RF PERFORMANCE
The SINTEF RF MEMS switch has been characterized by
observing static S-parameters during steady state
operation, and by observing settling dynamics during
transient operation. The measurements have been
performed directly on wafer by the use of a probe station
according to the measurement set up displayed in Fig. 4.
The use of a bipolar bias voltage was selected in order
to reduce the possibility of dielectric charging commonly
observed in capacitive switches [3]. According to [3], the
dielectric will not charge if the bipolar signal has a period Fig. 5 The probe tip (GSG 200) ready for touch down on the
that is shorter than the charging time of the dielectric. SINTEF RF MEMS switch denoted MAIN
Since the electrostatic force of the bias signal is
A. S-parameters
proportional to the square of the voltage, it will represent
a constant attractive force to the bridge even if the For the S-parameter measurements, the bias voltage
polarity changes. This force will however go to zero in was selected to be a rectangular bipolar signal at 10kHz
the transition between the two bipolar levels, but as long slightly above the pull-in voltage, hence using a hold-in
as this transition phase is fast enough (20-100ns [3]), the voltage that is identical to the pull-in voltage. The switch
bridge will not have time to react to the change. Based on was released after 10s by setting the bias voltage to zero.
these facts and the results reported in [Sec. 7.2, 3], a Under these conditions, none of the tested samples
square wave bipolar bias voltage has been selected for returned to their initial up state position. However, as
both the actuation and hold-down cycle of the switch shown in Section IIIB, the switches did release when the
presented in this paper. bias voltage was quickly and gracefully reduced after
In our case, the bias/control voltage is generated by a pull-in to a lower hold-in voltage.
high voltage arbitrary bipolar pulse generator, which is The measured S-parameters of the SINTEF switch are
isolated from the sensitive measurement equipment by presented in Fig. 6 and Fig. 7. In Fig. 6, insertion and
tees and attenuators. Due to the flexibility of this reflection loss are presented for the open switch (up-state)
generator, it is possible to address mechanical switch whereas in Fig. 7, similar results are presented for the
dynamics and to identify a bias voltage for fast and closed switch (down-state). All the measured results are
careful excitation. The selected bias tee in Fig. 4 compared to a modelled result based on the lumped RLC
represents a capacitive loading of the pulse generator. model shown in Fig. 8.
However, the generator still managed to deliver up to
60V pulses at a slew rate of 600V/ȝs. In Fig. 5, a
magnified view of the probe tip ready for touch down on
the RF MEMS switch is shown.
20GHz R&S
CW spectrum
source analyzer
Atten
3dB HP8510
network analyzer

HP33150A HP33150A
BIAS TEE SUMMIT 9000 BIAS TEE
Probe Station
SINTEF
RF MEMS
Switch

Control voltage

Arbitrary high voltage


Fig. 6 Insertion and reflection loss for the open switch
bipolar pulse generator (black) together with the modeled result (red) for the T-model
with C=49fF, R=0.82 and L=15pH.
Fig. 4 Measurement set up for either S-parameters with
network analyzer or for dynamic transient response with a high
speed synchronized spectrum analyzer.
movement is possibly caused by a stiction mechanism not
yet identified. Charge buildup, mechanical failure and
viscous tension from contamination are some candidate
explanations for this behavior.

Fig. 7 Insertion and reflection loss for the closed switch


(black) together with the modeled result (red) for the T-model
with C=1.24pF, R=1.0 and L=6.5pH.

Fig. 9 Dynamic on/off response of the RF MEMS switch at


20GHz after one million cycles with a spectrum analyzer in zero
span mode.

CONCLUSION
This paper has presented Norway’s first RF MEMS
switch, a capacitive shunt switch produced at SINTEF.
Fig. 8 Lumped T-model of a capacitive shunt switch Between 10-32GHz, the measured insertion loss is
between -0.17 and -0.4dB, the return loss is between -23
The impedance of the switch as a single port device and -12dB, and the isolation is between 7.5 and 20dB.
with the output port left open is The achieved capacitance ratio of the switch is 25.
, where is the
characteristic impedance of the t-line, and is the
ACKNOWLEDGEMENT
magnitude of the reflection coefficient. The highest -
factor is found at the phase angle , where the total This work has been funded by the Norwegian Research
reflection coefficient is pure imaginary. From the Council through the IRRFT (159259/I40) and SMIDA
measured S-parameters, the down-state -factor of the (159559/130) projects. The authors would like to thank
switch has been estimated to about 20-30 according to the Prof. Gabriel Rebeiz of the University of California at
San Diego for his valuable comments to the process
equation .
development of the switch.
B. Settling dynamics
The bias voltage level affects the settling time of the REFERENCES
switch, so fast operation is possible if the bias voltage is [1] J. B. Muldavin and G. M. Rebeiz, “High isolation CPW
increased. However, it has been observed that fast MEMS shunt switches - Part 1: Modeling,” IEEE
operation also introduces a mechanical bounce in the Transactions on Microwave Theory and Techniques, vol.
switch. Due to this, the settling speed was aimed to about 48, no. 6, pp. 1045-1052, June 2000.
100ȝs. In Fig. 9, the RF attenuation through the switch is [2] J. B. Muldavin and G. M. Rebeiz, “High isolation CPW
MEMS shunt switches - Part 2: Design,” IEEE
depicted as measured by a Rohde & Schwarz spectrum
Transactions on Microwave Theory and Techniques, vol.
analyzer. The switch is operated with the bipolar bias 48, no. 6, pp. 1053-1056, June 2000.
voltage and a 30V pull-in voltage, which is gracefully [3] G. M. Rebeiz, RF MEMS: Theory, Design, and
reduced to a hold-in voltage of approximately 12V. After Technology, John Wiley & Sons, 2003.
360ȝs, the bias voltage goes from -12V to zero and the [4] S. C. Saha, H. Sagberg, E. Poppe, G. U. Jensen, T. A.
bridge starts to return to its original up-state position. Fjeldly, and T. Sæther, “Tuning of resist slope with hard-
The switches have been turned on and off at a rate of baking parameters and release methods of extra hard
250Hz to provoke for any failure. After some time, a photoresist for RF MEMS switches,” Journal of Sensor
stiction tendency was observed. The delayed rise up and Actuators A: Phys., vol. 142 (2), pp. 452-461, 2008.

You might also like