MEMS: the second wave
The micro-electromechanical systems industry is going through a rebirth as consumer products start to adopt micromachined devices.
Steve Wainwright, general manager for Freescale Semiconductor in Europe, the Middle East and Asia, reckons you can feel the benefit of micro-electromechanical systems (MEMS) whenever you sit in a car: "If you feel you are a better driver than you were ten years ago, I would say that it's down to the MEMS in the electronic stability controller."
But, for many people in the electronics industry, MEMS was a speciality business that catered for a small number of big-volume customers. It's hard to find a car that does not have any MEMS sensors or actuators in it. The airbag has an accelerometer. The stability controller has its own accelerometer to detect pitch, roll and yaw so that it can apply corrections when the measurements start to look wrong. And each tyre could well have its own MEMS-based pressure sensor inside it.
Before the motor industry took MEMS seriously, it was 'the next big thing' in technology, with big predictions of how much it would drive the industry. Then things settled down and a relatively small selection of vendors, led by companies such as Analog Devices and Freescale, found markets in the automotive business.
Then a games console from a company everyone thought had been swept aside by Sony and Microsoft arrived. The Nintendo Wii showed that you didn't need expensive ultra-realistic 3D graphics to produce a successful console. What you needed was a cheap motion sensor, or three.
The Apple iPhone reinforced in consumer electronics engineering managers' minds that they had to take MEMS seriously, because their competition had already got there.
Benedetto Vigna, group vice-president and general manager of MEMS and sensors at STMicroelectronics said at the recent GSA and IET International Semiconductor Forum: "MEMS is no longer an emerging technology. Products such as the Apple iPhone have provided the second epiphany."
Massimo Sivilotti, chief scientist at Tanner EDA, one of the few design-tools companies to tweak its layout software for the MEMS industry, recalls: "Just a few years ago, those accelerometers, such as the ones used in the Wii, were not on the radar scan of most companies."
The sudden re-emergence of MEMS, this time in consumer devices, shows how quickly the manufacturers have to react to changes in demand. Sivilotti claims the original plan for the kind of accelerometer used in the Wii was to satisfy the market for satellite-navigation units.
"They were for assisted GPS," says Sivilotti. But, he adds, companies developed more sensitive antennas to cope with trees obscuring the weak timing-signals sent from space. Better processing helped deal with the way the high-frequency signal would bounce off the hard surfaces in buildings in the "urban canyons" of densely populated cities. "And some bright spark figured out very few people get lost in tunnels, so that all they really needed was an odometer," says Sivilotti.
Although the switch looks simple enough - simply replace one bunch of customers with another - the changes required are more subtle, Sivilotti explains. The sensor used in a Wii Nunchuk or an iPhone is a comparatively simple design, and does not call for the accuracy needed by applications such as assisted-GPS. This is potentially an opportunity for companies able to work on a cheap design and a headache for those who have invested in designing for a much more accurate system.
"We talk about the MEMS market as though MEMS is a single thing," says Sivilotti, describing how a simple design based on a moving air bubble, much like a miniature spirit level, made it into a souvenir torch from the Beijing Olympics because it was so much cheaper than a regular accelerometer.
"Everyone has had their eyes opened by consumer-grade MEMS.If you don't have to build something very precise, it is much easier to do," reckons Sivilotti.
High projected growth for consumer MEMS - possibly as high as 30 per cent per year and 50 per cent in the mobile-phone segment - will have a dramatic effect on the overall market, says Wainwright. "It will be interesting to see how that changes the established order of suppliers in MEMS."
The MEMS industry resembles the pre-CMOS days of the standard silicon-chip business. Many suppliers have their own secret-sauce process and there are very strong interactions between design and manufacturing. As CMOS picked up steam in the 1980s, its rise went hand-in-hand with the rise of abstraction.
Only the processor manufacturers spent a lot of time working on custom logic. Most teams were happy with the standard-cell design and the synthesis-driven workflows it made possible. As long as your design passed all the checks, the chances were that silicon delivery was as eventual as a trip to the photocopier. Foundries such as Xfab hope a similar trend will see the rise of fabless MEMS companies.
There are, however, problems with the idea of outsourcing production to foundries.
"There are no real standards because processes are being developed simultaneously with product development," claims Sivilotti. "It's not like designing an IC where, once you have the substrate, everything else falls into place."
But with MEMS, Sivilotti points out, packaging is an integral part of the sensor or actuator. It's not just an epoxy-based sealant with metal pins. Sivilotti points to the example of a heart stimulator designed to go into a catheter used during surgery that can use electric pulses to stop and start the organ's beating. "With that, you have the problem of designing a package that can survive a hostile, electrolytic environment," he explains.
Not only do the structures and surrounding circuits need to be custom-designed, so do the packages. However, a shift to the foundry model may lead to increased standardisation and openness from manufacturers. Fab operators like to keep their process data under wraps because it is useful information not just for customers but for competitors.
Trial and error
Very often, the only way to explore how a MEMS design will work is to make it. And when it fails - as it inevitably will - tweak the design, process or both and make it again.
Luckily, MEMS masks are bargains compared with those needed for advanced CMOS work because the geometries are often measured in micrometres not tens of nanometres. This means that many companies have simply adopted prototyping-intensive approaches to design. But the costs start to rachet up if the project calls for an integration of MEMS and the more delicately structured circuits of regular CMOS. This is where software to analyse the results of experiments and make predictions could prove useful. The plan of Process Relations, for example, (see box, p40) is to combine its XperiDesk software with process kits that pass the necessary process data to designers in encrypted form.
Other issues may slow the combination of CMOS with MEMS in a system-on-chip (SoC) approach and favour the use of system-in-package (SiP) instead. As MEMS design teams already have to worry about the package more intensely than their CMOS-only counterparts, the leap to putting an extra device into a package does not look as great. However, they do need to consider issues such as outgassing from one component affecting the MEMS devices, which often need to be kept clear of contaminants.
Volker Herbig, manager of strategic marketing at Xfab, says the decision to go down the SoC or SiP path is "very heavily application dependent. For pressure sensors there is a need to go there".
Herbig contrasts the cost of MEMS versus CMOS processing in terms of the number of layers, each involving multiple steps in the fab, needed to make a complete chip: "MEMS is a seven, eight, nine, ten layer process. If you then go to 0.35μm CMOS you need at least 16 or 17 layers and you can easily end up with 22 layers. The problem is that the CMOS portion is generally quite small in comparison to the MEMS part."
Although you only have to make one chip, much of the cost of CMOS processing is wasted because the bulk of the area involved uses none of it. In an environment where the product cost is largely down to the total area multiplied by the layer-processing costs, that is not good news. "You have to look closely at this dynamic," says Herbig.
The economics are not that good the other way round. David Ruffieux, an RF and analogue design expert at the Swiss Centre for Electronics and Microtechnology (CSEM), uses the example of MEMS-based clock chips, which are often smaller than 1mm2. "You could have produced a billion units by the time you have run the second wafer lot. So, why go with the SoC approach?" he asks, as the CMOS section would generally dwarf the MEMS portion but still have to pay for the additional deep-etching steps needed to produce the mechanical resonators.
Sivilotti says if the MEMS portion is relatively simple - some accelerometer designs do not need the deep etch that some of their more expensive counterparts do - then SoC can make sense. "The very, very low end of the commercial accelerometer space is very cost sensitive. So the least amount of post-processing required is an advantage. This is where SoC may be an advantage," reckons Sivilotti.
Test for MEMS
In consumer designs, it might be possible to trade off the accuracy of the MEMS portion for increased digital processing, which can be implemented in CMOS on the same chip. "The imperfection of MEMS can be compensated for at the system level," says Ruffieux.
And Sivilotti says: "The diversity of the MEMS space makes electronics look really simple."
Wainwright agrees: "Packaging, integration and test are completely different from [CMOS] SoC in terms of where the cost and complexity lie."
In contrast to CMOS, where there is an agreement among companies on implementation through techniques such as boundary scan and probe test, Sivilotti explains: "Test infrastructure doesn't really exist in MEMS. So you have to have a close relationship with the foundry."
Issues such as this mean MEMS is unlikely to see the rise of abstraction that took place in CMOS, Sivilotti concludes: "There will be no encapsulation of design methodologies. There will be a lot of iteration. First silicon success is not going to come to MEMS for some time."