Professional Documents
Culture Documents
© NRC publication no. 44453: Published in: Can. J. Elect. Comput. Eng., Vol. 27, No. 1, pp. 33-39, Jan. 2002
2
Vi Coff Vi Ron
Z0 Z0 Z0 Z0
1 2 1 2
Vr Vt Vr Vt
FIG. 2: A series switch model in the off state. FIG. 3: A series switch model in the on state.
Vi rline
Vi
Z0
Z0
1 C 2 C′
1 R 2
Vr Vt
Vr Vt
seems to be the opposite in the series and shunt configura- in a circuit, the impedance of the transmission line introduces
tions, the direction of optimum is the same. In both cases, a break frequency that is lower than the cut-off frequency by
minimizing the resistance of the switch optimizes the circuit. the ratio of line impedance to Ron . The FOM provides useful
insight into the fundamental behaviour of switching elements
and lets us compare switching element behaviour in isolation
Summary of switch models from circuits and from device geometry.
The first case for which we want to extract an FOM is not
In the above discussion of series and shunt switch configu- one of our switching elements at all but an ideal element of
rations we have presented an abstract model of a “switching semiconductor material. In Fig. 7 we show a small volume
device” that can be characterized by a small capacitance in of material deep in the semiconductor. We assume that some
one state and a small resistance in the other state. This may external control is exercised to switch the material from an in-
seem like an oversimplification to purists who would like to sulating state to a conducting state. The capacitance of this
use full models for each semiconductor device. However, in small element of material is just C = εr ε0 dxdy/dz. The resis-
practice, the simple 2 parameter model provides quite accu- tance of the same volume of material is R = ρdz/dxdy. The
rate prediction of switching circuit performance. It is also figure of merit is then
straightforward to extract these simple parameters from the
basic properties of each type switch. We have also noted that dx · dy dz
FOM = CR = ε0 εr ρ = ε0 εr ρ. (10)
the direction of optimum performance is the same for either dz dx · dy
series or shunt configuration. That is, we want lower resis-
tance and lower capacitance which leads to the concept of fig- Thus, the geometry cancels and the FOM is just dependent on
ure of merit. the basic material properties of dielectric constant and resis-
tivity. We cannot improve on the FOM by changing geom-
etry. For example, if we want to decrease Ron we can make
FREQUENCY DEPENDENT CHARACTERISTICS dz smaller. At the same time however, that would made Coff
larger in the same proportion as we reduced Ron and the FOM
Figure of Merit would be constant. From this reasoning, we conclude that for
a solid state device, the FOM has a floor value that is deter-
With direct current one uses the off-on resistance ratio to mined by the dielectric constant and the lowest resistivity that
characterize a switch. Since at microwave frequencies the can be achieved in that same volume.
off-state is determined by capacitance, we use the ratio of We write the FOM in terms of semiconductor mobility, µn ,
impedance to compare switches. The impedance ratio for our carrier concentration, N, and electronic charge, q; FOM =
simple two parameter device is just ε0 εr /(µn Nq). Inserting values for GaAs of N = 1017 cm−3 ,
µ = 5000 cm2 /(V·s), and εr = 13.1 [8], we get an FOM of
Zon 14.5 fs. We show in the following sections that this value is
Zratio = = jωCR. (9)
Zoff substantially less than any experimental values found for de-
vices. The higher device values are due to contact resistance
The ratio is frequency-dependent, a trait which we would ex-
and fringing fields. We will also show that MEMS devices
pect since one state is determined by a capacitance. The prod-
already demonstrate a lower FOM.
uct of CR is a characteristic number called the figure of merit
In principle, the FOM of a solid state device like a MES-
(FOM) of a switch. It has been used by several authors with-
FET can be calculated by integrating eqn. (10) remembering
out very much discussion. A smaller FOM is better since it
that in general ρ and ε are functions of position. Also, the
means a small on impedance relative to the off impedance.
electric field distribution may change when the switch is in
The reciprocal of the FOM is also used as a metric and is
different states reducing the direct application of the simple
called the cut-off frequency. The FOM matches what is ex-
model. This has already been done in the form of device mod-
pected intuitively. We can make a better switch element by
els from which we can extract the effective R and C values.
reducing R while C is held constant or by reducing C while
This we will do in the following section.
holding R constant. In an ideal situation we can reduce both
R and C.
While we have developed the FOM from an impedance ra-
tio, we cannot assign it a great deal of physical meaning. In dx
a MEMS switch FOM values are in attoseconds (as), which
corresponds to cut-off frequencies of tens of terahertz (THz). dz ρ , εr E
These frequencies are extrapolations far beyond the range of
validity of the assumptions that we made earlier in developing
our simple switch model. That means that we cannot plot a dy
graph of frequency response for a circuit and find a break in
the THz region. For example, if we consider any series switch FIG. 7: An element of the active area of a semiconductor.
SEMICONDUCTOR SWITCHES
MESFET Area A
IF
n+
Area A
r r n
C C p+
r W Ron
Q
i
C
n+
W S W
5 mWatts, an FOM of 220 fs can be achieved and at 25 mWatt ally takes place at only a small number of “high” spots lead-
that can be reduced to 110 fs. There are several reasons for the ing to a reduced effective surface area. For example, gold-
slow reduction in FOM with increasing power. The increased gold contacts with an area of 20 × 20 µm, a conducting length
recombination rate as noted above is one reason, but there is of 0.4 µm and resistivity of ρ = 2.5x10−6 Ω · cm would give a
also the effect of contact resistance as values of Ron reach the calculated resistance of 2.5x10−5 Ω. In practise, a resistance
1 Ω level. of 0.22 Ω was measured. This is 8800 times greater than ex-
pected which we interpret as an effective area factor of about
10−4 . We calculate the capacitance of a single gap in Fig. 11
Photoconductive as C = ε0 A/G. For the on state resistance, we use an effective
area factor ae , a conducting length l equal to the thickness of
Photoconductive switches have been used for very high- the contact films and a resistivity of ρ to give R = ρl/(A · ae ).
speed optically controlled switching of microwave sig- This leads to an FOM of
nals [14] and there has been an interest in making efficient ε0 ρl
optically controlled switches [1]. A photoconductive switch FOM = CR = . (15)
G · ae
has the geometry of an interdigital capacitor. Under illumi-
nation photo-carriers generated in the semiconductor regions Comparing eqn. (15) with eqn. (10), we see that εr has been
between the fingers provide an Ron which closes the switch. reduced to 1 giving an FOM reduction of 10−1 . The resistivity
The structure of this switch is very similar to that of a MES- of the metal is 10−4 less than the semiconductor giving a re-
FET in that the electrodes are planar. However, the isolation is duction of FOM of 10−5 . The effective area factor is approxi-
somewhat better because the gap is usually made longer than mately 10−4 meaning that much of the benefit of the material
the short channel length of a MESFET. The Ron is dependent change is largely lost due to the effective area factor. How-
on the optical power, the carrier mobility and lifetime in a ever, in contrast to eqn. (10), this FOM includes 2 geometry
manner similar to a PIN diode when optical power is substi- factors, the gap and the conducting length. The conducting
tuted for bias current. Starting with eqn. (10), we can write length is the thickness of the films that make up the contact
the FOM for the photoconductive region as follows: which we cannot reduce without introducing loss in the signal
path to the contact. Thus, the principle degree of freedom that
1 h̄ω is introduced by MEMS is the gap G. It accounts for a factor
FOM = ε0 εr Γ, (14)
µn qτ P of less than 10−1 in the MEMS switch of Fig. 11 if we con-
sider a 2 µm gap to represent a MEMS switch and an 0.4 µm
where µn is the electron mobility, q is the electronic charge, τ
channel length to represent a MESFET. For our comparison
is the carrier lifetime, P is the optical power density, ω is the
in Table I, we take experimental values from the literature to
optical frequency, h̄ is Planck’s constant, and Γ is the quantum
determine FOM. Ron of 0.22 Ω has been reported for a contact
efficiency. While we could could use this theoretical value for
FOM, for our comparison in Table I we take experimentally
determined values of Ron and Coff .
MEMS SWITCHES A
off on
/S12
of freedom available to the switch designer that does not ex- -30 3
ist in a semiconductor switch. That is the ability to set the 4 and 5
S12
gap independently of the materials. The FOM can be reduced -40
7
by as much as the gap, or mechanical travel distance, can be -50 6
increased. Secondly, the change in dielectric from semicon-
ductor to air gives an order of magnitude lower value for ca- -60
pacitance. The third conclusion is that the effective area fac- -70
tor is the dominant “material” property. It is not known if the
-80
factor of 10−4 is a fundamental limit or if improvements can 1 10 100
be made. The study of the physics of metal to metal contact Frequency (GHz)
in small areas may be productive in future improvements in
FOM. FIG. 12: Graph showing the comparison of the FOM for a number
of RF switches. The labelled curves are for 1 - measured S21 NRC
opto, 2 - Opto 40 Ω, 80 fF, 3 - PIN 1 Ω, 110 fF, 4 - Opto 100 Ω,
30 fF, 5 - FET 5 Ω, 100fF, 6 - Rockwell MEMS switch, and 7 - 60 µm
COMPARISON coplanar waveguide gap on quartz measured at NRC. Gray boxes are
the switching ratio for mechanical coaxial switches.
A number of sources were used to compile the data that
is summarized in the following Table I. The capacitance and
resistance data was extracted from commercial information,
published papers or determined experimentally by the authors. first generation devices. This is predicted by our theoretical
The last entry was taken from ARPA’s website. analysis presented above and represents a general advantage
of MEMS devices over semiconductor devices.
A further comparison of the switching elements is pre-
off /Son
sented in Fig. 12. In this graph we plot the ratio of S12 12
for a switching element connected in a series configuration.
For the optical switch and the MEMS switches, this is a con-
figuration that might be be used. The curves for individual
p-i-n diodes and MESFETS were produced by simulation us- APPLICATION OF THE FIGURE OF MERIT
ing Touchstone models. The data for the models was extracted
from commercial data sheets or measured from sample tran-
sistors. The FOM may be of most use when trying to develop a
A point that this graph emphasizes is that semiconductor new switch device. Our use of the FOM was developed af-
devices have similar off-on ratios even though they do not ter attempting to invent a new type of semiconductor switch
function by the same principles. The MEMS devices shown without initially considering the high frequency AC behavior.
are about 30 dB better in switching ratio even though they are The assumption was that if a large change in DC resistance
could be produced, then it would be possible to find a width
to length ratio that would produce a useful switch. In work
which we have reported earlier [1], we developed an optically
TABLE I: Summary of devices used in the comparison. controlled semiconductor switch. What we found was that
Device Class FOM Power Cap. Res. the switching characteristics were similar to that available in
(fs) (mW) (fF) (Ω) other devices even though those devices functioned by appar-
ently quite different physics. This can be seen in Fig. 12 and
IMS-Small Opto 4000 5 80 50
is in fact predicted by eqn. (10) which we subsequently de-
IMS-large Opto 3000 5 30 100
veloped. Which shows that the limit of microwave switch-
NE3290 FET 500 0 100 5 ing behavior is largely predicted by the dielectric constant of
Blackwell AlGaAs FET 270 0 170 1.6 the material in the off-state and by the conductivity in the on-
MA4GP022 GaAs PIN 4.5 220 5 110 2 state. What eqn. (10) makes quite explicit is that there is no
MA4GP022 GaAs PIN 20 110 25 110 1 geometry that will improve this result. What is further demon-
Raytheon MEMS memb. 12 0 35 0.35 strated by eqn. (15) is that if the switch physically moves when
Rockwell MEMS cant. 2.5 0 11 0.22 changing from the off-state to on-state, we can introduce an
COM DEV Coaxial 0.07 0 0.35 0.2 “engineerable” degree of freedom into the device. So far, we
ARPA-proj. MEMS 0.01 0 0.05 0.2 have not found a similar degree of freedom available by using
conventional semiconductor engineering.
cial material under the MEMS part. There will also be restric- ANT
tions on parameters such as process temperature and etching
techniques so that the underlying microwave circuit and sub-
strate are not damaged. A large number of design and fabri- FIG. 13: A simplified block diagram of a satellite payload.
cation methods are being developed for RF MEMS switches.
An overview is given in [7]. There are two basic switching
mechanisms: capacitive coupling where the switching mech- on-board integrated in the form of switch matrices to pro-
anism is due to a capacitance change and metal contacting vide system redundancy. The receiver input/output and low
where opening and closing an ohmic contact is the switching power switch matrices are typically implemented using coax-
mechanism. At NRC, we have been developing a metal con- ial switches while the high power switch matrix is imple-
tacting switch based on SiO2 or SiN fixed-fixed beam MEMS mented using waveguide switches. The RF-MEMS technol-
structure and a polyimide sacrificial layer [2, 3]. ogy could be potentially used to build miniaturized switch
Since the structure has free standing moving parts, the me- matrices to replace the bulky coaxial technology for the re-
chanical properties are of utmost importance. Properties such ceiver input/output as well as the low power switch matrix.
as residual stress, Young’s modulus, and defects must be well One would expect to achieve more than one order of magni-
characterized and their effects on the mechanics of the device tude reduction in mass and volume by replacing the coaxial
understood. This is complicated by the extreme aspect ratios technology with MEMS technology.
of thin film components and by the relative importance of sur- Such mass and volume reduction would have a dramatic
face properties due to the small scale of the devices. There is impact on the economics of a satellite program, since launch
still a great deal of research needed in these difficult areas of costs are related to satellite weight. Alternativley, the mass
mechanics of thin films. saved in the switch matrix could be replaced in other areas to
increase the capability to the satellite. That could be in the
form of increased capacity by adding payload electronics or
extended station keeping life by adding more fuel.
APPLICATIONS OF MEMS SWITCHES
MEMS switches could be also potentially used in beam
forming networks (BFN), particularly, in the design of re-
Of the many possible applications of MEMS switches, we configurable Butler matrices and phase shifters for the multi-
consider only 2 that reflect opposite extremes of the switch beam satellite communication systems.
market. Spacecraft applications demand the highest switching
performance and benefit by mass/volume reductions. At the
other extreme, wireless handheld phones use low cost semi- Wireless Applications
conductor switches.
MEMS switches can replace the SP2T and SP3T switches,
which are currently used in dual-band, and triple-band cell
Satellite Applications phones. These switches are currently implemented using
semiconductor technology. The advantage of using MEMS
Fig. 13 illustrates a simplified block diagram of a satellite technology in this case would be RF-performance improve-
payload. It consists of receive/transmit antenna, a beam form- ment, which would in turn reduce dc power consumption.
ing network (BFN), input filter assembly (IFA), high gain re- This is a very low cost high volume application at present
ceiver (RCV), input and output multiplexers, high power am- filled by GaAs MESFET switches. A commercial MESFET
plifiers (HPA) as well as several switch matrices. A satellite provides about 0.9 dB insertion loss which by itself consumes
system of this type would typically have 100’s of switches about 19% of generated RF power. In principle a MEMS
References
∗ Electronic address: peter.grant@nrc.ca
FIG. 14: A switched filter bank for multi-band receivers.
[1] P.D. Grant, M.W. Denhoff, M. Gao, and R.R. Mansour, “A
Photo-conductive Microwave Switch” in Conference Proceed-
switch could provide 0.2 dB insertion loss which reduces the ings of ANTEM98, Ottawa, ON, 1998, pp. 475-482.
power loss to 4.5%. This improvement is a worthwhile en- [2] M.W. Denhoff, P.D. Grant, M.A. Harry, and M. Yu, “Fabrica-
gineering objective, but as a consumer product it would have tion of a microwave MEMS switch,” in Proceedings of ANTEM
2000, Winnipeg, 30 July - 2 August, 2000, pp. 79-82.
to be available at about the same cost as a GaAs MESFET.
[3] P.D. Grant, L. Winchiu, S. Piek, M.W. Denhoff, M. Yu, and
Presently GaAs MESFET switches cost less than $1 in quan- M.A. Harry, “Microwave Performance of a MEMS Switch De-
tities of 100,000. There are as yet no similar commercially signed for RF and Microwave Routing,” in Proceedings of AN-
available MEMS switches so that we cannot make a cost com- TEM 2000, Winnipeg, 30 July -2 August 2000, pp.83-89.
parison. [4] R.N. Tait, “An IC-compatible process for fabrication of RF
A further requirement for this application is that the switches and tunable capacitors” in Can. J. Elect. Comput. Eng.,
switches must be specified for the environment found in a vol. 25, no. 1, Jan 2000, pp. 26-28.
[5] J.J. Yao and M.F. Chang, “A surface micromachined miniature
hand held phone. The temperature range required is -40 °C
switch for telecommunication applications with signal frequen-
to 85 °C. The state of development of MEMS switches seems cies from DC up to 4 GHz”, in Proc. Transducers 95 – Eurosen-
to be still at a level of ensuring basic functioning. We have sors IX, Stockholm, Sweden, 1995, pp. 384-387 vol. 2.
not found references to RF characterization over the required [6] C.L. Goldsmith, Z. Yao, S. Eshelman, and D. Denniston, “Per-
temperature range. formance of low-loss RF MEMS capacitive switches”, in IEEE
It is expected that future wireless receivers would need to Microwave and Guided Wave Letters, vol. 8, no. 8, Aug. 1998,
operate at several bands covering a wide range of frequencies p. 269.
[7] J. Jason Yao, “RF MEMS from a device perspective,” in J. Mi-
from 900 MHz up to 5 GHz. These receivers will have also
cromech. Microeng., vol. 10, no. 4, Dec. 2000, pp. R9-R38.
to operate in an environment of increasing interference. Such [8] S.M. Sze, Physics of Semiconductor Devices, 2nd. ed., John
requirements could be met with the switched filter bank as Wiley & Sons, New York, 1981.
shown in Fig. 14. The entire band to be covered by the re- [9] Y. Ayasli “Microwave switching with GaAs FETs”, in Mi-
ceiver is divided into several channels where the appropriate crowave Journal, vol. 25, no. 11, Nov. 1982, pp. 61-74.
narrow-band is selected using a switched tunable-filter bank. [10] D.A. Blackwell, D.E. Dawson, and D.C. Buck, “X-band MMIC
In view of current conventional technologies for RF filters, switch with 70 dB isolation and 0.5 dB insertion loss”, in IEEE
Microwave and Millimeter-Wave Monolithic Circuits Sympo-
a receiver with such capability would be bulky, power con-
sium, Orlando, FL, USA, 1995, pp. 97-100.
suming and very expensive. However, with the use of MEMS [11] G. Hiller, “Design with PIN diodes,” Alpha Industries Applica-
technology, MEMS switches could be integrated with MEMS tion note APN 1002, June 1999.
tunable filters to build the whole re-configurable receiver on [12] “Applications of PIN Diodes,” Hewlett-Packard application
one single chip. The availability of high performance MEMS note 922.
switches and filters is a key to miniaturization of such type of [13] “MA4GP Series Data Sheet,” M/A-COM Semiconductor Prod-
wireless receivers. ucts, Burlington, MA.
[14] D.A.M. Khalil and A.M.E. Safwat, “On the improvement of the
performance of the optically controlled microwave switch”, in
IEEE Trans. MTT, vol. 45, no. 8, Part 2, Aug. 1997, pp. 1358-
CONCLUSION
1361.
[15] J.M. Bustillo, R.T. Howe, and R.S. Muller, “Surface Microma-
In comparison with the dominant semiconductor devices, chining for Microelectromechanical Systems,” in Proceedings
MEMS offers a fundamentally superior basis for developing of the IEEE, vol. 86, no. 8, Aug. 1998, pp. 1552-1574.