MEMS stands for microelectromechanical systems. It is also
sometimes known in Europe as MST, for microsystems technology,
with a slightly broader definition. But in essence, though, the
terms are interchangeable. Despite four decades of development
and some commercial deployments — such as pressure sensors and
accelerometers to trigger automobile airbags — MEMS is only now
showing signs of becoming a field separate from
microelectronics. And it appears to be sufficiently large in
terms of potential markets to justify investment in startup
companies and increased activity by established players.
The signs are there: the breadth and depth of university
research leading to commercial startups, renewed interest among
mainstream chip makers and, perhaps most significantly, the
reaction of the semiconductor equipment makers that have started
to make equipment adapted, if not dedicated, to various types of
MEMS manufacture.
With a fully functioning silicon MEMS wafer fab costing as
little as $60 to $80 million, the barriers to entry into MEMS
are considerably lower than those for microelectronics, where a
large-scale "economical" wafer fab for digital chips can cost $2
billion. That significantly lower cost might seem like a good
thing but the nascent state of MEMS means that the risk of
building up overcapacity in manufacturing is even higher in MEMS
than in microelectronics.
But for the next few years a limiting factor on growth is
likely to be a persistent diversity of component architectures,
process technologies and materials. The strong interdependence
of design, manufacturing and packaging in MEMS means that a
second choke on growth is likely to be the availability of
engineers trained in the complexities of MEMS — renaissance
engineers who have a breadth of skills and knowledge in process,
materials and mechanical and electronic design.
Right now many questions are being asked of the MEMS
protagonists. Is silicon set to triumph as the substrate of
choice because of the pre-existence of a strong equipment and
materials infrastructure? Will it thereby achieve economies of
scale and drive out glass and plastic substrates? Even on
silicon there are a many alternative process technologies for
building MEMS. Which process technologies will succeed? Will
each application demand the optimum manufacturing technology?
Must so many processes co-exist, even at the expense of
requiring many dedicated manufacturing lines? Is MEMS technology
development destined to follow the 30-year progress pioneered by
microelectronics, albeit at an accelerated rate, or will the
differences required by mechanical, optical, fluidic and
biochemical devices and their application environments
necessarily throw the various subclasses of MEMS onto completely
different development paths?
In business terms, does a foundry-and-fabless design-house
model make sense? Or will a few of the multitude of agile
startups with their own manufacturing capability come to
dominate MEMS? Or will a few of the semiconductor giants,
perhaps the companies that have shipped the vast majority of
MEMS components to date, simply acquire those startups when the
time is right, or take the markets from them by dint of deep
customer relationships and deep pockets?
Which technical and business visions for the industry will
catch the broader imagination and take root? And after four
decades of marginal significance why should MEMS make its "big"
commercial entrance now at the beginning of the 21st century?
While the industry tries to answer these questions, it must
be remembered that contained within MEMS technology are some of
the seeds of the "nanotechnology" that will almost inevitably
rise up to complement it — and eventually supplant it.
Technology
By definition, MEMS components contain micrometer-dimensioned
elements, usually with a moving part, sometimes a solid
mechanical one, sometimes a fluid one, and usually integrated
together with at least some electronic circuitry. The
electronics may be only a piezo-resistor network or a capacitive
element to transduce mechanical motion into an electrical
signal, but it can be much more. MEMS with fully integrated
electronics offer the promise of truly complete
"system-on-chip." MEMS encompasses sensors of many types
although optical sensors such multipixel CCD or CMOS camera
chips are usually omitted.
MEMS are usually manufactured in a fashion similar to
integrated circuits, making use of the experience in lithography
and etching. Although silicon is a favored substrate for this
reason, glass, quartz and plastic substrates are sometimes
preferred. MEMS have also evolved natural complementary roles to
microelectronics: as sensor-transducers, which can provide
signals for electronic processing, and as transducer-actuators,
which can move or effect a change in response to that
processing. Typical motions are electrostatically operated
linear and rotary motors, although steam and internal combustion
engines are also being miniaturized.
Whereas microelectronics exploits the semiconducting property
of silicon, MEMS and more elaborate micro-machines also exploit
the structural properties: a low coefficient of thermal
expansion, a high thermal conductivity and a strength that for
its equivalent weight is greater than aluminum. In a
complementary role are piezoelectric materials that produce an
electrical charge when deformed and, conversely, deform under a
voltage — allowing precise electrically controlled motion and
actuation.
A four-wave history
The first wave of MEMS commercialization started in the late
1970s and early 1980s with bulk etched silicon structures, and
back-etched membranes, used to make pressure sensors. As the
thin silicon membrane deforms under pressure, it effects a
piezoresistive track laid on its surface and the change is used
to transform the pressure into an electronic signal. Subsequent
devices include the capacitively sensed moving-mass
accelerometer used to trigger airbag deployment in automobiles,
and gyroscopes for orientation.
A second wave of commercialization arrived in the 1990s,
mainly oriented around the boom in PC and information
technology. Texas Instruments introduced video projection based
on electrostatically actuated tilting micromirror arrays. The
thermally operated inkjet print head remains a high-volume
application.
It can be argued that the third wave of MEMS
commercialization boomed and busted at the turn of the millenium,
with micro-optics as an accompaniment to fiber-optic
communications — by way of all-optical switches and related
devices. Despite the current deflation in that market,
micro-optics will be a strong growth area for MEMS in the long
term.
Other applications that could be the inspiration and
beneficiary of the fourth wave of commercialization include
static and moving devices for radio frequency passives
components, silicon-based audio, biological and neural probes,
so-called lab-on-a-chip biochemical and drug development systems
and microscale drug-delivery systems.
What's up with processes?
The recent upsurge of interest in MEMS has partly come about
with the advent of surface micromachining technology, in which a
sacrificial layer — material that keeps other layers separated
as the structure is being built up — is dissolved in the last
step, allowing the creation of suspended thin moving and
resonant structures.
But as Bernard Courtois of the TIMA Laboratory (Grenoble,
France), a MEMS research group, observed: "There are two ways to
manufacture microsystems: develop processeses specific to
microsystems or use process that have been developed for
microelectronics. Among those later processes, some can be
targeted to microsystems, or it is possible to add special
process steps to accommodate microsystems within integrated
circuits."
Many applications call for departures from traditional
electronics manufacturing, in terms of extra steps, back-surface
processing, unusual metals and more exotic materials, wafer
bonding and so on. Indeed many applications, particularly in the
biological and medical area, do not tolerate silicon as a
substrate.
In many cases, glass and plastics are preferred, and
disposable medical devices are often stamped out in plastic for
reasons of cost.
But for many companies and research groups the existence of
CMOS, silicon-germanium and gallium arsenide process
technologies for microelectronics is a starting point for MEMS
development. The integration of MEMS with electronics on the
same chip could, in theory, improve performance, efficiency and
reliability of the overall circuit and reduce the cost of
manufacturing and packaging.
One dominant approach to high integration is post-processing
of MEMS structures in reserved areas on top of the
microelectronic die by surface micromachining. However, the
temperature at which the previously fabricated microelectronics
is damaged has to be taken into account. Thus, containing MEMS
manufacture within a low temperature regime is a key topic for
those pursuing monolithic integration.
To this end, the Interuniversity Microelectronics Center (IMEC;
Leuven, Belgium), has developed a polycrystalline silicon
germanium (SiGe) deposition technique with a critical
temperature of 450 degrees C compared with 800 degrees C for
polycrystalline silicon. However, with the lower temperature
comes a slower deposition rate. And so a second method with a
much higher deposition rate at a slightly elevated temperature
of 520 degrees C has been developed. The choice of SiGe is an
attempt to intersect with what will become the de facto standard
process for high-frequency electronics, but many others are
looking at mainstream digital CMOS as their starting point.
Earlier this year, IBM announced that it has developed RF
MEMS components using standard production materials on a BiCMOS
process technology at temperatures below 400 degrees C. IBM has
developed MEMS resonators and filters that could replace
discrete passive components in wireless devices.
"There's been a continuing discussion about the integration
of CMOS and MEMS over many years," said Roger Grace, an industry
consultant specializing in MEMS and microsystems. "But about the
only integrated process running in volume today is that for
Analog Devices ADXL-50 accelerometer. The same functionality is
being created by Motorola using two chips, one for the MEMS and
one for the microelectronics with integration at the package
level.
Such debates are commonplace in microelectronics. It is
noticeable that analog and mixed-signal microelectronics are
frequently integrated at the package level as circuits on
separate die. Similarly, intelligent power electronics is
frequently implemented as a multichip solution although others
tout the benefits of intelligent-power process technologies. The
reasons for and against integrating mechanical structures and
large amounts of electronics are even more complex.
The situation is compounded by the fact that, unlike
microelectronics, where standard packages developed quickly and
are the norm (varying, in essence, on pin-count and attach
method), MEMS environmental parameters are very diverse. Some
packages must exclude light while others must allow it onto the
die surface. Some packages must maintain a vacuum over or behind
the die, while others must pipe gases or liquids around a chip.
It is recognized that it is impossible to develop a standard
package for the diversity of MEMS applications. But it is also
highly desirable that for each application category the industry
defines a standard package and its road map.
"MEMS designers tend to make the device first and then think
about test and packaging," observed consultant Roger Grace.
Given that in excess of 95 percent of the cost of a component
can be in the test, packaging and final assembly, getting that
optimized should be more important than building the most
elegant MEMS structure.
Meanwhile, SEMI, the industry group, is starting to put work
into standardization of both packaging and manufacturing
processes.
Thus, for example, Sandia National Laboratories has developed
the Summit V process technology that includes up to five layers
of polysilicon and has licensed it and related design kits out
to commercial companies such as Coventor Inc. and Ardesta LLC
(Ann Arbor, Mich.) a venture capital company. Sandia has
licensed its four-layer Summit IV process technology to
Fairchild Semiconductor Corp. (South Portland, Maine).
This is part of a deliberate policy of infrastructure
commercialization undertaken by Sandia and referred to by Sandia
workers at a recent MEMS symposium as the "Visionary's Dilemma,
or How to evolve from initial demonstration and to industry
standard."
Roger Grace said: "Sandia's policy is to license technology
into industry to get the volume of deployment up so they can
prove reliability sufficiently for their own low volume
applications." The Summit processes are perhaps shown off to
best advantage when building gears, chains and micromachines.
"Summit V is an expensive process. Are there enough
applications to support Summit V?" Grace asked. "Engineers still
want to match the manufacturing process technology to the
application."
According to Grace, only a handful of volume applications for
MEMS have been established so far. Even Texas Instruments'
moving-mirror video projection chip, despite its success in the
desk-top projector market, runs at less than 1 million units per
year, according to Grace.
"We could see a continuation of custom processes and custom
solutions. I don't see them [engineers] compromising."
So is talk of MEMS as a boom market premature?
Marlene Bourne, senior analyst at In-Stat MDR, predicts that
the worldwide market for MEMS will grow from $3.9 billion in
2001 to $9.5 billion in 2006, an average growth rate of 19.5
percent. By comparison, the worldwide market for semiconductor
chips has been hovering at about $150 billion since 1996,
although it is expected to get back to 20 percent growth in
2003.
Nexus, a European industry body, estimated that the market
for microsystems was already $30 billion in 2001 and will rise
to $68 billion by 2005. This disparity with In-Stat's numbers is
due mainly to a broader definition of "microsystems" used by
Nexus. Nexus says microsystems can include whole systems, such
as heart pacemakers, and extend to polymer, glass, metal and
ceramic-based devices.
Bourne said: "I define MEMS by size and as devices with a
mechanical function generally fabricated in silicon although not
exclusively. It's more to do with the process technology, which
could be derived from bulk, surface micromachining or Liga. I am
trying to measure components as supplied to OEMs so I am not
measuring value at the die level, nor at the end-user product."
Business
"The MEMS industry goes in waves, we've seen pressure sensors,
accelerometers and mirror-based optical devices come up," said
Bourne. "RF is the next wave; miniature relays and switches.
They have extreme volume but extreme price pressure," she said.
Bourne also referred to silicon microphones as an upcoming
area. In such devices the traditional microphone diaphragm is
made from a thin silicon membrane. Certainly companies such as
Akustica Inc. (Pittsburgh) and SonionMEMS A/S (Lyngby, Denmark)
are being set up to exploit this possibility.
Akustica is also pursuing the possibility of silicon
sound-generation devices.
Although small size would limit output power, such components
could be well suited to hearing-aid applications. And the
ability to put 255 such silicon loudspeakers on a single die,
for example, would not only increase the loudness of the device
but could allow 8-bit digital addressing of such miniature
"loud" speakers. Similarly, the ability to deploy multiple
microphones and microphones in arrays could open up new
application areas. Such applications might include directional
microphone strips that could be run strung around a space — the
interior of an automobile or home for example — so that a
captured sound is accompanied by positional information.
"What we are seeing is a strong segmentation of the MEMS
field," said Bourne.
And it's true that startups and established chip and systems
companies are playing in such areas as transportation, health
care, telecoms and consumer electronics. And many of the
startups are present in only one application segment. As with
microelectronics, MEMS is becoming a horizontal enabling
technology that may serve a number of broad vertical markets.
"There has been a lot of interest from the traditional
semiconductor manufacturers but that is probably a knee-jerk
reaction to problems in their own sector. The process
technologies are so different," said Bourne.
But are they? As TIMA and others have pointed out, there are
advantages to be gained if you can use a standard process even
if it is modified from a baseline IC process.
The difference between silicon MEMS, MOEMS (micro-optical
electro-mechanical systems) and conventional IC manufacture is
therefore one of degree.
And indeed, this being the case, an entry into MEMS is
appealing to smaller chip manufacturers, which have been under
assault from the top integrated device manufacturers (IDMs) and
the Taiwanese microelectronics foundries. As these smaller chip
companies cannot afford the cost of developing deep-submicron
process technologies and new factories to house them, they must
find niche markets or leave manufacturing altogether.
It is notable that Fairchild Semiconductor and Austria
Microsystems (Unterpremstatten, Austria) have added MEMS to the
range of devices they can manufacture. And that a Xicor 6-inch
wafer fab in Milpitas, Calif., has been acquired by Standard
MEMS Inc. (Burlington, Mass.) to be turned over to MEMS
manufacture; and that the Plymouth, England, wafer fab of
Zarlink Semiconductor was sold to X-Fab Semiconductor Foundries
AG (Erfurt, Germany), which characterizes itself as a
mixed-signal and MEMS device foundry.
But it is also notable that the vast majority of MEMS sold to
date have been made by major semiconductor players such as
Motorola, Analog Devices and Texas Instruments and that
STMicroelectronics is expanding its interests in the area.
The appeal of MEMS to the larger chip manufacturers, the ones
who can still afford to pursue deep-submicron CMOS process
technology, is making an increased margin on older process
technologies and wafer fabs that have been fully amortized by
years of manufacture. In other words, the trailing edge of
microelectronics can become the leading edge of silicon MEMS
manufacture.
In-Stat's Bourne said: "Yes, a number of conventional IC fabs
have said they can do this; but it is not easy. It takes very
specialized expertise, which is in short supply and resides in
people at the universities and start-up companies."
She added that while there is the development of a foundry
business model serving fabless component developers, these
tended to be dedicated MEMS foundries such as Standard MEMS,
Intellisense and Cronos. "Fifty percent of the startups have
in-house fab facilities. You need to have people who understand
the process technology." Which, in turn, highlights the need for
training.
"For ST the interest in MEMS is quite strategic," said
Benedetto Vigna, MEMS business unit manager, STMicroelectronics.
"The main reason is that we want to address new markets with
silicon by exploiting the different properties of silicon. And
because MEMS can be built at about one micron it means we can
reuse existing manufacturing facilities."
ST has been a longtime supplier of thermally operated inkjet
printer heads to Hewlett Packard. In addition it has developed a
range of inertial sensors, including angular and linear
accelerometers. It has also shown a thermally operated optical
switch developed for Agilent.
And, indicating one direction for the silicon MEMS market to
develop, STMicroelectronics struck a deal with Onix Microsystems
Inc. (Richmond, Calif.), a heavily backed Californian startup.
In July 2001 the two companies agreed to co-develop and
manufacture chip sets containing MEMS and ASICs for use in
Onix's optical switching engines. ST's volume manufacturing
capability will provide the chip sets required by Onix.
"In RF MEMS the priority is not so high," said Vigna."We have
nothing commercially developed although one of the great areas
of interest is mechanical switches to replace gallium arsenide
solid-state switches."
According to Vigna there are still issues of reliability for
the industry to demonstrate and, "if it is going in a mobile
phone you need a very low actuation voltage. The wish is for
sub-5 V although the cell phone makers will probably allow us 12
to 15 V. There again if they need 40 V for organic light-mitting
displays we might not need to go so low."
With some startup companies, such as MicroLab Inc. (Chandler,
Ariz.), claiming to have RF switches ready for the market, it
remains to be seen if these devices will remain discrete or be
integrated with SoCs.
Waiting for standards
"I still don't see a CMOS-like universal MEMS process on the
horizon, so there is a kind of fragmentation at present. But
there will be a consolidation; in EDA, in design, in testing. As
soon as standards appear a kind of simplification could happen,"
said ST's Vigna.
"If we could stick to two or three process platforms we could
improve time- to-market for MEMS because they would become
reliable, through the buildup of experience."
"I believe the most successful approach," Vigna continued,
"will be a modular approach with the integration of CMOS and
MEMS at the package level. You could sell the MEMS die with
bumps for flip-chip integration.
"We are pushing two or three main processes," said Vigna.
"The main process is Thelma — for Thick Epitaxial Layer for
Micro Accelerometers. This is a 0.8-micron process with a thick
polysilicon layer for structures and a thin polysilicon for
interconnect."
The technology allows silicon structures to be attached to
the substrate in a few points called anchor points, but which
are free to move in a plane parallel to the substrate itself. To
be compatible with traditional plastic packaging techniques, a
cap is placed over the top of the sensor element to prevent the
molding from fouling the moving elements.
For micro-actuation a similar process is used but without the
cap. Instead, a flexible passivation layer is added. For the
work with Onix on micromirrors a third process is used. This is
because Thelma's polysilicon layers are not as good as
monocrystalline silicon for producing mirror finishes.
Vigna explained "If we could stay on two or three process
platforms we could improve time-to-market because the processes
become more reliable." Vigna is also of the opinion that most
integration will be done at the package level.
So how will design automation tools for MEMS develop?
Vigna's answer is "slowly."
"EDA tools are useful when design becomes a science. But
today MEMS are more like analog circuitry. First you need
standard processes, then you get standard design tools."
Despite the present diversity of the process technologies
some third parties with EDA tools have emerged from such
companies as Ansys Inc. (Canonsburg, Pa.) and Coventor Inc.
(Cary, N.C.).
And just like the semiconductor equipment makers, the
mainstream EDA vendors, such as Cadence Design Systems Inc. (San
Jose, Calif.), are starting to treat MEMS as a distinct
engineering discipline.
Memscap, itself a provider of MEMS design automation tools,
has transformed itself into one of a new generation of dedicated
MEMS fabricators. It is pursuing RF and optical applications and
has developed a number of processes in support of these
applications. More controversially, it has built a dedicated
MEMS wafer fab near its original base in Grenoble, France, and
supported Walsin Lihwa, a Taiwanese company, to build through
licensing agreements.
"The relationship with Walsin is very close. We helped them
build their fab. The idea is that we go optical and they go
wireless. So that we both get both technologies to market very
quickly, and both of us will have second sources," said
Jean-Michel Karam, president and chief executive officer of
Memscap.
Karam's argument is that building a wafer fab is part of the
investment needed to kick-start the MEMS market. "A lot of
companies say they have an RF MEMS switch. But before Nokia
commits to using RF MEMS they need to know you can supply 100
million units a year. You need to prepare for big volume," he
said.
Karam said that the industry will see standardization. "The
best way to make money is to minimize the number of processes —
hen when you make changes you make them in design and not in the
process. Technically you can't generalize this easily," he said.
The outcome, he argued, is that a few processes will become
"standardized" for different high-volume applications in each
industry sector and once they become standardized, economic
pressure will start to drive MEMS designers to use these
processes without alteration.
Agile vs. stable
One might expect STMicro's Vigna to predict that the major
semiconductor companies will reap the MEMS harvest in the years
to come.
After all, ST has experience of manufacturing in volume, has
wafer fabs all over the world in which it could insert MEMS
processes and has close relationships with a number of major
customers.
"Big companies like dealing with big companies. A big company
can sue another big company if things go wrong. Of course
startups are more agile, more flexible and focused to only work
on MEMS."
But Vigna suggested a third way forward for MEMS business, a
way to combine the volume manufacturing capability of a major
company with the flexibility and focus of a startup.
"The first big company to spin off its MEMS group could be
really successful," said Vigna, although he denied that
STMicroelectronics has any plans to set his own MEMS business
unit free.
Memsic Inc. (Andover, Mass.), a spin-off from Analog Devices
Inc. (Norwood, Mass.) formed in 1999 would provide a good
example except that Analog has also retained its own MEMS
activities. In fact, Memsic illustrates another trend: the
rising significance of Taiwan and China.
Taiwan and China
Memsic was founded by Yang Zhao, a graduate of Beijing
University, who served within Analog Devices' micromachined
products division between 1993 and 1999.
Zhao had a particular interest in a thermally based
accelerometer and persuaded Analog Devices to let him take the
technology and develop it in return for an equity position in
his startup. Zhao is a passionate believer in the utility CMOS
process technology for MEMS and now produces highly integrated
accelerometers embedded in complex logic on a single chip.
"We concentrate exclusively on CMOS MEMS. The only way to
succeed is to piggyback on the available technology. You should
copy as much as possible before you start inventing," Zhao said.
Memsic uses TSMC as a foundry for logic at 0.6-micron design
rules before taking the six-inch wafers back and implementing
the MEMS structures at its subsidiary in Wuxi, China. This does
restrict Memsic to low-temperature post-CMOS processing so as
not to damage the electronics on the wafer — but that is the
technique that Memsic has developed.
"It means we can take design to any one of a number of fabs
and even though designs have to be tailored to suit the
processes we would rather do that," said Zhao. "Besides, TSMC
does not have a MEMS process so we do it ourselves."
But TSMC is an investor in Memsic, so will TSMC be without a
MEMS module for very long?
In the Shanghai district of China there are several 8-inch
foundries being built. Tens of billions of dollars are being
spent. So China is going to be one center of silicon
manufacturing. Taiwanese foundries may look to develop more
specialized processes, including ones supporting MEMS to run in
their older fabs on their trailing-edge processes.
"We would be interested in taking on more MEMS business,
provided the technology that's required is in tune with our
basic semiconductor manufacturing technology. . . . I see CMOS
technology as a platform on which to put optical devices, MEMS
or even carbon nanotubes," said Chenming Hu, chief technology
officer of TSMC, in a recent interview with EE Times (http://www.eetimes.com/story/OEG20020517S0044).
Roger Grace said: "Taiwan and China are gearing up to be big
in MEMS. They want to do in MEMS what they've done in
microelectronics. We need to keep very close eyes on that."
Summary
The future of MEMS is therefore full of ambiguities.
Engineers will continue to pursue hundreds of MEMS ideas on
scores of process technologies and packaging solutions, but
standardization will throw up some preferred options based on
silicon.
Agile and innovative start-ups across the Uniteed States.
will drive MEMS applications forward, leaving others in their
wake — but Europe's relatively few hit-the-ground running
spin-offs will also succeed. Meanwhile, Taiwan and China will
repeat in MEMS the success that they have achieved in the world
of microelectronics manufacturing.
But do not expect MEMS to suddenly blossom as
microelectronics did in the 1960s and 1970s. MEMS remains
extremely diverse and difficult. MEMS is microelectronics plus
micromechanics in pursuit of hundreds of applications across all
industry sectors.
Do expect MEMS markets to turn in better long-term growth
than "pure" microelectronics from this point on. As foundry
services develop, engineers will increasingly come to make their
designs fit the process technology rather than the other way
around, making standard EDA tools and design kits economical and
initiating a virtuous cycle.
Perhaps the most significant reason why MEMS can take off now
is that the complexity of micromechanical analysis has — through
the advent of gigahertz processors — become solvable on the
engineer's desktop. One remaining question is whether MEMS
technology has enough time to triumph or must blur into
nanotechnology before it has enjoyed its season in the sun.