The big foundries now see serious money in silicon's Cinderella.
The received wisdom about the market for microelectromechanical systems (MEMS) is that it has been on the cusp of huge profits for more than a decade. However, MEMS is also an extremely fragmented sector that has struggled to achieve economies of scale and to export manufacturing technologies from one project to another.
Jean-Christophe Eloy, founder and CEO of analyst firm Yole Developpement, famously summed up MEMS's main problem up as 'one product, one process, one package'. Speaking at last Summer's Semicon West conference, he said, 'I first identified this issue in 2001 and it still has not been solved for most companies.'
As the name suggests, MEMS typically draw on concepts from several different branches of engineering. Mechanical and electronics are the best known, but, for example, the last five years have also seen rapid growth in the integration of bio-engineering for sensor applications. Such combinations tend to have highly specific end-uses - indeed, only a handful of MEMS devices have annual markets worth more than $200m.
At the same time, the intellectual property (IP) that underpins these combinations has tended to be closely guarded and large-scale innovators have been wary of outsourcing production to third-party foundries until the core technology has matured. Meanwhile, many such contract manufacturers have themselves come to see most MEMS opportunities as representing a lot of hard work for comparatively little return. Moreover, because they are often micro-machines, the yields can look far worse than those for traditional silicon chips - and to a foundry, yield is the most important number of all.
MEMS's promise has thus been held back by its tendency to behave like the semiconductor equivalent of bespoke tailoring. Certain 'cuts' may catch on - the demand for injket print-heads and accelerometers for airbags remains 'classic' - but by and large things come made-to-measure and therefore expensive.
This year, though, could be the one in which things start to change. Earlier this month, GlobalFoundries, the Abu Dhabi-backed chip foundry, declared that it aims to be the world's largest MEMS manufacturer by 2015. For sure, the company may see MEMS mastery as easier to achieve than in just about any other semiconductor field, given Taiwanese foundry TSMC's huge lead over its rivals in traditional CMOS-based processes. But, here's a curious thing: TSMC itself has also started talking seriously about MEMS, after many years of scepticism. And neither company features in the sector's current Top 20 foundry ranking - although they may dispute how that list is assembled.
This change of heart can best be attributed to the convergence of several factors. Emerging applications have the potential to deliver higher volumes than in the past. Innovators have woken up to outsourcing. Foundries have begun to ferret out economies of scale.
As was the case throughout industry, 2009 was a difficult year for MEMS, but a recovery is on the way. Jean-Christophe Eloy estimates last year's total market value at just under $7bn. He expects this to be followed by 13 per cent growth in 2010, followed by an annual growth rate over the 2011-2015 period of 14-15 per cent. This suggests the MEMS will be worth $16bn by 2015.
'However, if you look at [the market] in terms of devices, the growth there is much more important. It will go from close to 3.2 billion units in 2009 to more than 10 billion units in 2015,' adds Eloy.
Reducing the cost of existing MEMS hits that go into the consumer electronics and automotive industries will play an important role. But Eloy is also now forecasting rapid surges in demand for some newer products. This could be led by digital compasses that are being integrated into cellular handsets to provide location-based services. These compasses alone are expected to represent 2.2 billion of the devices sold in 2015, just a little fewer than 14 per cent of the overall space.
The talk throughout engineering is of inter-disciplinary products, and MEMS can claim to be the poster child for that trend - all that it offers, and all the obstacles that remain.
At the same time, manufacturing innovations are being brought to bear as MEMS also moves from being a market based on discrete functionality to one that takes a system-based approach, placing greater emphasis on elements such as software. These dynamics are familiar from the mainstream semiconductor business.
'We used four- and six-inch wafers in 2000 and now we are moving to six- and mainly eight-inch wafers by volume,' Eloy says. 'Only a few players might be on eight-inch, but that is where the majority of growth is happening based on manufacturing capabilities.
'We have also moved from wire-bonding side-by-side to stacked die. There is more innovative packaging and companies are using it for added value. We have then also moved to SOI [silicon-on-insulator] wafers and epi-poly approaches to silicon. And we have removed capping and implemented 3D TSV (through-silicon-via).
'All of these things are already in development or in early production.'
An important factor here is that foundries equipped with such technologies are needed in a system-led world, so that the developers can concentrate more on the software and other innovations that will differentiate their offerings.
So it is that the contribution to broadening MEMS commercialisation offered by the growth of the foundry sector is not simply significant but essential, and at all levels of the business.
Smaller players have long had to work with manufacturers that specialise in low-to-medium-volume production, because building their own fabs is far beyond a start-up's resources. However, major companies have typically operated their own fabs (notwithstanding the long-standing relationships in print-heads maintained by STMicroelectronics as a foundry for Hewlett-Packard and Texas Instruments as one for Lexmark).
However, in the last 18 months, two major automotive companies, Delphi and Continental Automotive Systems, have announced that they are ceasing in-house manufacturing and transferring the work to foundry partners. During a panel at Semicon West, Asif Chowdhury, a director in the Worldwide Manufacturing Organization at Analog Devices (ADI), said that his company is also looking for third-party fabricators, and a major influence on its thinking has also been the volatility in demand that typifies consumer electronics. Part of ADI's MEMS portfolio includes the chips that enable the 3D gameplay on Nintendo's Wiimote.
'[The move towards contract manufacturing] been a key trend over three years and has been amplified by the downturn in the last year,' Eloy says. 'According to the data we have, at least five companies are involved in such changes at the moment with a total business value of $350m. That's a huge transfer to the foundries.'
There is, however, the inevitable gap 'twixt cup and lip. Eloy believes that there are still economies of scale that the foundries will need to apply more stringently and promote more heavily, much as they have done in the CMOS world. MEMS is a historically messy domain, as ADI's Chowdhury noted: 'It can be the case that even within one company, there is no standard. There are different test platforms, different packaging platforms.'
This kind of dispersed effort may well be costing the MEMS industry dearly. Yole's research suggests that it takes an average of four years to bring a product to market at a cost of $45m, with a company burning through three or four CEOs in that time. Such figures might seem modest in comparison with the costs dogging chip design at advanced nodes, but it again must be remembered that MEMS devices typically target smaller markets.
As 800lb gorillas like TSMC and GlobalFoundries investigate MEMS's potential, it is worth bearing in mind that only six of the sector's 20 largest suppliers have annual sales in excess of $20m. Boosting those numbers will largely depend on breaking the inefficient 'one product-one process-one package' flow.
'The customer has to understand that making a million is very different from making one. There's a learning curve; there's going from prototype to production and that takes time and money,' Yan Loke, vice president of Engineering at Canadian foundry Micralyne, told the same Semicon West panel as Chowdhury.
The MEMS Industry Group is soon to release a foundry checklist that companies can use to guide not only choices of manufacturing partner, but also development strategies. In the meantime, there are some fundamentals to note, and foremost among these is exploitation of the opportunities for reuse that are already available.
'The way to be successful and really help the customers is to come up with more platforms that are reproducible and reusable,' John Foster, CEO of US foundry Innovative Micro Technology, told Semicon West. 'That's really what people need. It speeds up development, and lowers the cost enormously.'
This may always have been the case - but as foundries get more business, they get more experience. So it is that Foster's company can now use its involvement in numerous technological areas to effectively cross-pollinate its internal IP portfolio.
'We're in all these different businesses, so we can survive, so we have the economic diversity,' he explained. 'There's something that happens when you get the economic diversity and it is that you get a technical diversity from working with customers in all these different areas. So, we might make a breakthrough in a biomedical application and turn that around and use it in a high-end navigation application. That happens literally every week - and it's very exciting.'
This kind of reuse also suggests that the relationships between MEMS partners and their foundries may need to be much closer than those in more discrete technological fields - and they could cover a wider range of issues.
'It's not just about finding capacity,' John McKillop, managing director of the MEMS research company Tekton Consulting, told Semicon West. 'It's about finding a fit at the management level, at the expectation level, at the IP level, at the whole range of levels in the company and then executing.
'The thing I keep telling my clients is that the choice of foundry is the most significant decision a start-up is ever going to make. It's their most significant financial decision and it will have more to do with their success or failure than any other decision they take. They have to be careful. They have to be methodical. And they have to execute.'
Foster said that the degree of alignment here could go as far as seeing the foundry take on many of the functions of an incubator. 'Companies come in with venture capitalists and we help them,' he said. 'That's one other service we provide - we have five companies living inside our facility.'
One of the great debates in all branches of electronics remains 'how close is too close?' Big players always want multiple sources, although Chowdhury said that this is easier said than done in today's MEMS foundry sector. Meanwhile, smaller ones live in fear of their great new innovation getting 'locked in', so that one manufacturer controls access to the market.
'This is one of my favourite soap boxes,' said McKillop. 'I think that the critical issue our industry faces will be the availability of an open-source process at a large foundry. It doesn't have to be a well-defined PDK [product development kit] or have all the fancy design tools that you have for CMOS. But it's got to be open-source, so it's available to allcomers with clear design rules in a high-volume, low-cost fab. I think that the evolution of that kind of process will make us a well-defined industry in the future.'
Is this where muscular players like TSMC could come in and eat the medium-sized foundries for lunch? Is this the gap that the MEMS Industry Group's checklist will start to fill? As desirable as open-source technologies may be, there are those MEMS players who believe the sector is simply too diverse in its sources of inspiration.
'One moment you're working with optical MEMS. Or then it's accelerometers. Or gyros. And we do a lot of things related to biology,' said Foster. 'It's not that standardisation isn't a good idea. There are just so many things, it's impossible.'
However, there are those who say that within the technological panoply there are some issues that could be addressed specifically. Again, Chowdhury noted that companies bringing their own internal operations into line could be an important starting point.
'Maybe as a first step, companies could come up with one advanced packaging platform to serve all their needs,' he said. 'That in itself would be a huge move in serving the need to reduce costs.'
In the meantime, exploiting such reuse as is available and following the general design-for-manufacturing maxim of pushing process considerations (as well as those for packaging and other back-end elements) to as early as possible in a design flow would appear both logical and necessary in the short term. More than that, it looks increasingly possible.