out with the old in with the new
TRANSCRIPT
But with the continued expansion of MEMS devices
into new areas of application, the limitations of
silicon (Si) usefulness became clear. The need to
combine electronics and MEMS on the same chip
(iMEMS), improve the wear characteristics of moving
parts, and achieve a greater mass of moving parts in
MEMS inertial sensors have led researchers away
from a ‘one-material-fits-all’ approach. Instead the
search is on for materials that more directly serve
specific ends. The need for biocompatibility in the
emerging field of bio-MEMS has added urgency to the
quest for new materials, since Si-based materials
cannot meet every bio-MEMS need.
A more basic problem with Si arises from its surface
properties. At micron dimensions, surface forces are
frequently stronger than the more familiar elastic and inertial
forces that are proportional to volume. Particularly
troublesome is stiction, the tendency for surfaces to stick
together. This is because water – present from humidity or
other sources – in small gaps gives rise to strong capillary
forces. If the materials involved are hydrophilic and the water
evaporates, the materials pull together seeking the few water
molecules remaining. This may stick surfaces together and
render a device inoperative. Though a bare Si surface is
hydrophobic, it quickly oxidizes on exposure to air or water
to become hydrophilic.
Expensive procedures, such as supercritical drying1, are
routinely used to avoid this problem in effecting the final
release of complex Si MEMS structures. A better solution
would be the discovery or creation of a hydrophobic thin film
by Robert Huber and Neal Singer
Out with the oldin with the new
Sandia National Laboratories,Albequerque, NM 87123, USAE-mail: [email protected]
(Image above courtesy of Sandia NationalLaboratories.)
The materials issues facing the
microelectromechanical systems (MEMS) community
can be understood best in terms of the historical
context. The field began almost as an afterthought
among those engaged in integrated circuit
production. These researchers recognized early on
that the same processes used in the production of
circuits could be re-ordered to make very small
mechanical devices. The huge investment made by
the electronics community in silicon technologies –
and the relative ease with which these techniques
could be adapted to device production – made them
an obvious resource for early MEMS designers. It is no
accident that polycrystalline silicon is the most
commonly used structural material.
July/August 200236 ISSN:1369 7021 © Elsevier Science Ltd 2002
REVIEW FEATURE
material, suitable for MEMS fabrication and able to simplify
the release process by allowing water to be removed by a
dehydration bake.
This article describes a few of the promising new materials
being explored at Sandia National Laboratories and other
organizations as solutions to the above problems. Any new
films must have properties that polycrystalline Si does not,
but must be compatible with most of the microfabrication
processes that already exist.
Integrated MEMSAn oft-mentioned advantage of Si surface micromachining
(SMM) is its potential for integrating complete electronic
circuits and complex micromachines on the same chip2. The
advantages are obvious: lower system costs thanks to batch
fabrication of the entire system, better performance because
of lower noise and less capacitive loading, higher reliability
because of fewer interconnects, and simpler packaging.
However in practice, realization of these advantages has
proven to be quite difficult for a number of reasons, including
some very complicated questions involving materials
compatibility.
The material of choice for SMM structures is usually
polycrystalline Si, the same material extensively used in
CMOS (complementary metal-oxide silicon) circuits. In many
ways polycrystalline Si is an ideal material for MEMS. It has
high strength, can be made almost stress-free (an essential
property for free standing structures), and is linearly elastic
at strains up to the fracture point. It can be deposited as
conformal layers that are part of a multilayer structure.
Sandia National Laboratories’ SUMMiT™ process
(see Figs. 1 and 2 for examples) is able to use as many as five
polycrystalline Si layers separated by sacrificial layers of
silicon dioxide (SiO2) and silicon nitride (Si3N4).
Why, then, are there so few integrated MEMS processes
available and why are they so expensive? The answer lies in
the fabrication details.
For micromachines, the polycrystalline Si layers must be
formed stress-free. Stress-free polycrystalline Si requires very
high temperature anneals, in the 1100°C range. But modern
CMOS circuits use submicron dimensions, both horizontally
and vertically. Submicron MOS transistors cannot withstand
the high temperatures over time required to produce stress-
free polycrystalline Si. Different approaches have been taken
by various labs to solve this problem.
Sandia National Laboratories has developed a process in
which the MEMS components are fabricated first in a
depression etched into the wafer surface3. After the
mechanical elements are formed, the depression is
completely back-filled with SiO2 and the wafer is resurfaced
using chemical mechanical polishing (CMP). At this point, the
July/August 2002 37
Fig. 2 Computer-generated cut-away model of a SUMMiT 5™ device. (Courtesy of JamesAllen, Craig Jorgensen, and Victor Yarberry at Sandia National Laboratories.)
Fig. 1 Computer-generated cut-away view of a SUMMiT 5™ device, in this instance atorsional ratchet actuator. (Courtesy of Stephen Barnes, Craig Jorgensen, and VictorYarbury, Sandia National Laboratories.)
wafer goes into the CMOS process where the electronics are
formed by a near-standard process. Since the mechanical
polycrystalline Si has already been exposed to very high
temperature anneals, the relatively low temperature process
for submicron CMOS has little or no effect on the MEMS
device. After completing the CMOS device, a few additional
low temperature steps are needed to complete the metal
interconnections and release the MEMS structures
(see Figs. 3 and 4). This MEMS-first approach is difficult in
practice because of the very long fabrication throughput
times and restrictions on the MEMS design. Therefore, many
attempts have been made to add MEMS to completed circuits
– a second approach to fabricating integrated MEMS.
But a critical need for the MEMS second approach is much
lower temperatures during MEMS fabrication. This need is
one of the drivers in the search for new MEMS-suitable
materials. Two examples of the MEMS second approach are
the Defense Advanced Research Projects Agency (DARPA)
sponsored Application-Specific Integrated-MEMS Process
Service (ASIMPS) alpha run, a prototyping service supported
through a collaboration of Cronos Integrated Microsystems,
MEMSCAP, and the Carnegie Mellon MEMS Laboratory, and
the service from MOSIS.
The first process uses conventional CMOS processing
followed by a sequence of maskless dry-etching steps to
fabricate composite metal/insulator microstructures.
Currently, the process is designed for a 0.6 µm CMOS process
from Austria Micro Systems International4, but temperature
is still an issue.
The integrated MEMS process from MOSIS has much the
same approach. Processing steps are added to a conventional
CMOS-processed multi-project wafer and the MEMS user
performs a post-processing maskless anisotropic etch. At
present the available CMOS is the 2 µm double
polycrystalline Si SCNA and SCPE Orbit process5.
In with the newAmong the more interesting new MEMS materials that might
solve the temperature problem are polycrystalline germanium
(Ge), polycrystalline silicon germanium (SiGe), polycrystalline
diamond films, and amorphous diamond films. All four of
these can be formed in a stress-free condition at much lower
temperatures than polycrystalline Si.
Researchers at the Berkeley Sensor and Actuator Center at
the University of California, Berkeley, under the leadership of
Roger T. Howe, have demonstrated microstructures built on
standard CMOS wafers using both polycrystalline Ge and
SiGe.
Polycrystalline Ge
This material6 can be deposited by conventional low
pressure chemical vapor deposition (LPCVD). Rapid thermal
annealing (RTA) can then be used to give low resistivity,
tensile polycrystalline Ge films for freestanding
microstructures. The time-temperature cycles are compatible
with standard aluminum (2% Si) metallized CMOS wafers.
The mechanical properties of polycrystalline Ge are
comparable to Si and resonators have shown quality factors,
Q, as high as 30 000 in vacuum. Resonators with integrated
REVIEW FEATURE
July/August 200238
Fig. 3 Cross section of a Sandia National Laboratories integrated MEMS structure.
REVIEW FEATURE
CMOS amplifiers have been built to demonstrate the
capabilities of the material.
Polycrystalline SiGe
If used in combination with a sacrificial layer of
polycrystalline Ge, this material shows promise as a
micromechanical structural material for use on top of
conventional CMOS wafers with aluminum (Al) metallization
(see, for example, Fig. 5).
Stress free polycrystalline SiGe can be deposited by CVD at
temperatures low enough to be compatible with completed
CMOS wafers7. These films can be formed into complex
mechanical structures by surface micromachining with post-
CMOS compatible steps. The built-in strain in the
polycrystalline SiGe layers can be controlled from -145 MPa
to +60 MPa by deposition conditions. Surface micromachined
structures have been successfully formed directly on standard
CMOS electronics8. Folded flexure lateral resonators with a Q
value as high as 15 000 in vacuum have been reported.
Amorphous diamond films
Two other new materials being considered for MEMS are
amorphous diamond (aD), grown by pulsed-laser deposition9,
and CVD polycrystalline diamond films10.
Diamond and hard amorphous carbon have recently
emerged as a promising class of materials to improve the
mechanical performance and reliability of MEMS (see, for
example, Fig. 6). Diamond has the highest hardness (~100
GPa) and elastic modulus (~1100 GPa) of all materials.
Amorphous forms of carbon – specifically the hard carbons,
aD, tetrahedral amorphous carbon (ta-C), and diamond-like
carbon (DLC) – can also approach crystalline diamond in
hardness (up to ~90 GPa) and modulus (800+ GPa). It is not
the hardness that is the main appeal of these materials for
the MEMS designers, however, it is their extreme wear
resistance (up to 10 000 times greater than Si),
hydrophobicity, and chemical inertness.
The inherent stiction resistance of these carbon-based
materials (hydrophobic surfaces do not stick together) and
their chemical inertness permit their use in aggressive
chemical environments. Crystalline diamond also possesses
the highest thermal conductivity and the widest range of
July/August 2002 39
Fig. 4 Integrated MEMS chip with complex CMOS circuits and MEMS components on the same chip. The three-axis accelerometer prototype is a joint effort by the University of California,Berkeley and Sandia National Laboratories.
optical transparency (from far IR to UV) of any material,
making it useful for thermal heat sink structures or micro-
optics.
MEMS structures fabricated from mostly CVD
polycrystalline and nanocrystalline diamond, both in the area
of surface micromachining and in mold-based processes, are
making progress. Already-demonstrated diamond
microstructures include: substrates with integrated channels
for active-cooling, micromachined fresnel optics, optical fiber
alignment structures, free-standing capillary tubes,
acceleration sensors, thermally-actuated liquid ejectors,
electrical microswitches, micro-tweezers, diamond cantilevers
with tips for scanning probe microscopy (SPM) applications, a
diamond motor structure, diamond gears, and
nanoindentation tips.
One disadvantage of CVD polycrystalline diamond is
surface roughness – typically quite high, ~1 µm RMS. This
limits the minimum device dimensions to typical layer
thicknesses of several microns, with tens of microns in the
lateral dimension. This inhibits the use of multi-level
processes to create complex MEMS structures.
A new type of hard carbon material recently introduced is
termed stress-free aD. The material is an amorphous mixture
of nanophases of tetrahedrally-coordinated carbon – about
70% of the total – with three-fold coordinated carbon
comprising the remaining 30%. The three-fold coordinated
carbon is not distributed randomly. Instead, it clusters in
conjugated chain-like or sheet-like structures. In the
deposited state, the material exhibits extremely high levels of
compressive film stress (8 GPa), but, surprisingly, with
suitable control of film deposition, 100% stress relief (down
to 0 ±10 MPa) can be achieved. The resulting stress-free films
are hard, optically transparent, moderately conductive, and
best of all, nearly atomically smooth (0.1 nm RMS roughness
on Si, 0.9 nm RMS roughness on SiO2). These properties are
favorable to the use of this material for surface-
micromachined MEMS applications, where small dimensions
and complicated structures are desired.
The low temperatures needed by the aD films open the
possibility of their application to a second approach for
integrated MEMS. It should be possible by post-processing to
add MEMS to previously processed chips. Al metallization
may have to be added after formation of the aD structures,
but there is no reason why this could not be done.
Polycrystalline diamond films
Polycrystalline Si vibrating micromechanical (‘µmechanical’)
resonators with frequencies in the hundreds of MHz that
retain Q values of around 10 000 are now being
demonstrated in research efforts9. The idea is to further
extend frequency ranges into the high-UHF and S-Band
ranges needed for front-end RF applications in today’s
wireless transceivers.
Several approaches to increasing frequency ranges have
been successfully demonstrated, including:
(1) Scaling of devices to smaller and smaller dimensions
through brute force;
REVIEW FEATURE
July/August 200240
Fig. 5 Schematic cross-section of an iMEMS showing the combination of CMOS built with the usual n-type polycrystalline Si and MEMS made with polycrystalline SiGe. (Courtesy of RogerHowe and Andrea Franke (now at Motorola) of the Berkeley Sensors and Actuator Center.)
REVIEW FEATURE
(2) Strategic geometries that take advantage of higher
frequency mode shapes and eliminate Q-degradation
caused by anchor losses; and
(3) alternative materials such as SiC that boost the acoustic
velocity of the resonator structural material, thereby
making it easier to achieve higher frequency without the
need for overly tiny dimensions.
Because of the possibility of ‘scaling-induced’ performance
degradations of the above techniques, (2) and (3) show the
most promise for extending the micromechanical resonator
frequency range while retaining the Q values and degree of
stability needed for application to the low-low bandpass
filters and ultra-stable reference oscillators needed in
communication transceivers.
One variant utilizes CVD polycrystalline diamond with an
acoustic velocity potentially more than two times higher
than that of polycrystalline Si. This material can be used as a
structural material for clamped-clamped beam and folded-
beam micromechanical resonators. Polycrystalline diamond
has great potential to achieve the desired UHF frequencies of
0.3-3 GHz needed in wireless communication transceivers
more easily because the resonance frequency of a mechanical
device is generally directly proportional to the acoustic
velocity of its structural material. Because of its inherent
stability and chemical inertness, diamond can potentially
offer better aging characteristics than polycrystalline Si – an
important consideration for frequency and reference
applications. Finally, polycrystalline diamond can be
machined nearly as easily as polycrystalline Si, making it
amenable to conventional surface micromachining. With
deposition temperatures below 600°C, CVD polycrystalline
diamond can be more easily integrated with highly
conductive metal electrodes for lower losses and higher
power handling at GHz frequencies.
Since CVD polycrystalline diamond CC-beam
micromechanical resonators have demonstrated measured
resonance frequencies from 2.7-9.9 MHz and Q values up to
6225, they easily equal the current output of polycrystalline
Si resonators in this frequency range. In addition,
approximately 39 kHz folded-beam, comb-driven resonators
have been demonstrated with Q values in the range of
20 000. The frequencies seen in these polycrystalline
diamond devices are 15% higher than that of equivalently
sized polycrystalline Si versions. While this is not the value
potentially achievable, even moderate success has
encouraged further work in this area.
A radical approachThe four new materials discussed above, while not standard
for MEMS, still have most attributes in common with
July/August 2002 41
Fig. 6 Surface micromachined structure of amorphous diamond. (Courtesy of John P. Sullivan and Thomas A. Friedmann, Sandia National Laboratories.)
polycrystalline Si. Some form of CVD deposits the films. The
fabrication processes involved are near-standard surface
micromachining. The final structures look pretty much like
the standard product. But still more radical approaches are
being taken. High aspect-ratio structures are being made by
using a deep reactive ion etch (DRIE) to create a mold from a
Si wafer. The mold is backfilled with thin sacrificial layers and
a thicker structural layer of polycrystalline Si, or
polycrystalline SiGe, or in some very recent experiments at
Sandia National Laboratories, by tungsten (W). Researchers at
the Berkeley Sensors and Actuator Center at the University of
California, Berkeley11 have produced microactuators using Si
molds that are 100 µm deep, but only 6 µm wide.
Efforts to use W as a microelectronics structural material
are underway at Sandia National Laboratories12. This effort
fabricates freestanding W devices that are supported above a
Si substrate by columns of Si3N4. The process starts with the
fabrication of a Si mold, as described above. Some trenches
are filled with Si3N4 to provide the support columns, while
other shallower trenches are filled with W, which forms the
MEMS device. The parts are released by etching away the top
layer of the Si wafer, leaving the W free, but the deeper
nitride-filled trenches are only partly exposed to provide the
support columns (see Fig. 7).
This process has a number of very attractive attributes
when compared to polycrystalline Si SMM MEMS. Among
them are high ‘Z’ direction stiffness, much greater mass than
Si SMM parts, higher capacitance between elements, high
lateral forces, and much higher electrical conductivity
compared to thin polycrystalline Si. There also exists the
potential to integrate W parts with CMOS processes because
all processing steps are at low temperatures. The density of
W is eight times that of Si, and the layers can be many
times thicker than polycrystalline Si layers. Thus, the mass
per unit area of the ‘proof mass’ in a SMM inertial
sensor could, possibly, be increased by a factor of 50.
Since the resolution of some MEMS designs is now limited
by thermal noise, such a large mass increase would allow
a major increase in the sensitivity of these inertial
sensors.
REVIEW FEATURE
July/August 200242
Fig. 7 Tungsten MEMS structures made by the Si micromolding process. The support posts are Si3N4. (Courtesy of James G. Fleming, Sandia National Laboratories.)
Micromoldingadvantages
• Deep trench technology allows depths of tens
of microns
• Molding maintains planarity
• Molding greatly reduces or compensates for
stress
• High Z stiffness
• High force attenuation
• High capacitance
• Tungsten for heat spreading?
REVIEW FEATURE
The futureThe new MEMS materials discussed here are just a few of the
many being developed in labs around the world. It is too early
to tell which, if any, will find widespread use. In many cases a
new material is chosen to solve a specific problem with a
specific MEMS device or application. Unless the specific
application generates a large enough volume to support the
generally high cost of both development and specialized
production equipment needed for the unique material, the
material will not become part of the established technology.
Both microelectronic and MEMS manufacturing are very
expensive undertakings. It takes a very high volume to support a
production line economically. Evidence of this is found in the
number of integrated circuit (IC) companies that have gone ‘fab-
less’, using contract-manufacturing services known as foundries
instead. This strategy works with IC chips because a wide variety
of applications can all use the same fabrication line. This is not
necessarily so for MEMS, especially if the MEMS product uses a
new material to solve some unique problem. If the device has
too limited an application, the unique material will remain a
laboratory curiosity, and Si, with all its handicaps, will remain
the material of choice. MT
July/August 2002 43
REFERENCES
1. Mulhern, G.T., et al., Proc. 7th Int. Conf. Solid-State Sensors, Actuators, andTransducers, Yokohama, Japan, (1993) pp. 296-299
2. Sniegowski, J.J., and de Boer, M. P., Annu. Rev. Mater. Sci., (2000) 30, p. 299-333
3. Smith, J.H., et al., Proc. IEDM (1995) p. 609
4. See http://www.ece.cmu.edu/~mems/asimps_short_course/
5. See http://www.mosis.org/Technical/Designsupport/nist-mems-1.html
6. Franke, A.E., et al., 12th International Workshop on Micro Electro MechanicalSystems (1999) Orlando, USA
7. Sedky, S., et al., J. Microelectromechanical Systems (1998) 7(4), p. 365-372
8. Franke, A.E., et al., Solid State Sensors and Actuators Workshop (2000) SouthCarolina, USA
9. Sullivan, J.P., et al., Mat. Res. Symp. Proc. (2001) 657
10. Wang, L., et al., 15th IEEE International Conference on Micro Electro MechanicalSystems (2002) pp. 657-660
11. Heck, J.M., et al., Transducers '99, Sendai, Japan (1999) pp. 328-331
12. Fleming, J., Mat. Res. Symp. Proc., (2002) 729, to be published