The year of the 8-inch MEMS fab

by Jérémie Bouchaud, Wicht Technologie Consulting

Eight-inch MEMS fabs are hardly anything new. In revenue terms, a major part of the MEMS market has been processed on 8-inch lines since 2001, when Texas Instruments moved to 200 mm production. For many years, though, TI was the only significant manufacturer and the opportunity for equipment and wafer suppliers was limited.  Since 2005, a handful of companies have joined the 8-inch group; these are Hewlett Packard, STMicroelectronics, EPCOS and foundries Dalsa, DNP and tMt.  And indeed WTC estimates that in 2006 around 16% of the revenue generated by the MEMS market was produced from activity on 8-inch lines.

So what will the future bring? WTC believes that 2008 will be a transition year and expects the 8-inch revenue share of the MEMS market to more than double from $1.4 billion this year to $3.5 billion in 2011. This represents a growth of 16% to 31% in the same time. The revenue growth will accelerate in 2009 and 2010 as a number of new 8-inch MEMS fabs become operational.

The signs for this are clear. In 2006, only the two companies in the top 30 MEMS manufacturers, TI and HP, were manufacturing MEMS at 200 mm diameters. By 2011, there should be at least 12 players: ST's and Freescale's 8-inch lines are already operational: Bosch, Omron and Canon should follow in the next 18 months. And wtc knows of a further six companies among the top 30 who expect to, but have not yet publicized their plans to move to this diameter.

The foundry offer also currently expands in 200 mm. Dalsa, tMt, DNP and Jazz Semiconductor have started to offer services, while APM, Omron, TSMC Silex come on line by 2009.

Factors driving the adoption of 8-inch wafers

The combination of MEMS for consumer electronics combined with automotive MEMS sensors now enables the large volumes that justify a move to 8-inch, and this route is taken by Bosch, Freescale and others. For consumer markets, operating an 8-inch line is now a marketing argument to convince your customer that you can ramp-up the product in large volume rapidly and that you are not at the limit of your capacity.

Meanwhile we expect a significant part of the inkjet business to move to 8-inch diameter production. HP has run a 200 mm line in Corvallis since 2005. Canon plans to move to 200 mm, MEMjet will start on 8-inch with a Taiwanese MEMS foundry and WTC anticipates that other inkjet players will buy older 200 mm CMOS fabs and convert them to print head production.

However, what is new is that certain mass production level products are beginning with 8-inch. This is the case of MEMS oscillators from Discera through foundry partner Dalsa, SiTime with Jazz Semiconductor and Silicon Clocks with an as yet undisclosed foundry partner.
This is also the case for RF MEMS switches. RFMD is currently building an 8-inch facility dedicated to RF MEMS and NXP will also probably start at this diameter. WiSpry will partner with Jazz Semiconductor.

Another factor is the existence of the infrastructure. The equipment offer is mature, e.g. wafer bonders at EVG and Suss.

Further, a number of manufacturers of mass products are undergoing drastic technology changes in order to cope with the extreme price pressures they face in production sensors for consumer electronics, such as the requirement for new 3D packaging solutions. As each needs to invest in new processes and equipment, this naturally coincides with an upgrade to 8-inch diameter substrates.

Role of foundries in transitionto 8-inch wafers

MEMS and IC foundries with a 200 mm capability already attract companies developing MEMS products for mass markets, as shown by Dalsa for Discera’s oscillators and Jazz Semiconductors, which makes switches for WiSpry and MEMS oscillators for SiTime.

WTC also expects that some large MEMS manufacturers will increasingly cooperate with these 8-inch foundries as their consumer business expands. We do not expect the likes of Analog Devices to externalise all of their MEMS production, but the company can externalise some parts of their processes. This could concern standard processes if the companies want to keep more critical processes in house for strategic reasons. This can also concern special processes, such as through silicon vias (TSV) for 3D packaging.

WTC anticipates that externalisation of part of the processes will affect mainly consumer products, which need very fast production ramp up. Most of the automotive MEMS production will remain in house.

Foundries such as TSMC and Silex are positioning themselves to welcome these partial processes. Silex with its offer for TSVs, while TSMC already manufactures a part of the accelerometer sensor made by MEMSIC and will inherit certain ADI processes for the production of consumer MEMS from the second half of 2008.

Risks associated with scale up

There are however risks associated with moving to the new size, and also ways to mitigate these. One of these risks is over-capacity. One company recently revealed its production had already reached 100,000 devices a week, but WTC knows that 50,000 devices fit on the company's 8-inch wafer. This illustrates the overcapacity risk with 200 mm. A way to overcome this is to share the production of MEMS with other devices, and for example, Bosch will manufacture both MEMS and ASICS in its new fab.

There are anticipated overcapacity issues across the entire industry. With twelve companies expected to be producing on 8-inch diameter wafers among the top 30 companies in 3 years time compared to three today, MEMS production capacity will significantly increase in this time. As a result, wtc anticipates an accelerated price decrease for high volume products like inertial sensors, microphones and pressure sensors.

Another risk is the tuning of processes. Transitioning to a larger wafer size means learning to work with new equipment, and adjusting the process again. More than in any other area, this phase is the most cost intensive and time consuming in MEMS production. While the final tuning still has to be undertaken in a company's own facility, the emergence of 200 mm R&D fabs, such as Minatec in France or SVTC, is a great support in this phase and lowers the risk associated with transferring from 6- to 8-inch.

Investment also represents a major issue. Building a new 200 mm fab from scratch - as Bosch intends - requires a initial capital investment of $150 to $200 million. An alternative that offers lower initial investment is to buy and convert to MEMS one of the ten to fifteen 200 mm CMOS fabs currently for sale. Another option is to cooperate with an 8-inch foundry on either a complete process or a portion of the process.

Each approach has advantages and drawbacks (see table), and each company must examine the options carefully. The optimum solution depends on the starting situation, as follows:

**  For an IC and MEMS producer, it is easier to convert a former 200 mm fab to meet proprietary production requirements. This is the case of Freescale, where a part of the 200 mm IC fab in Texas is now dedicated to MEMS, while the company still produces microcontrollers, power devices and RF ICs.

** A pure play MEMS manufacturer normally opts for a new facility or partners with a foundry. WTC notes that even smaller MEMS manufacturers move into large facilities. In 2004, AMR and GMR sensor company Sensitec took over a major part of the IBM HDD manufacturing facility in Mainz. The compatibility of the processes and drive for more capacity outweighed the disadvantages of owning an oversized infrastructure.

Comments

[Gene Burk]

I'm surprised at the rush to 200 mm facilities by these companies. A simple back of the note pad calculation indicates little likelihood of profitability. Take the case of the resonator for cell phones. ASP of the currently used Quartz device is about $0.20. While the TAM is ~ 1 billion annualy the realistic SAM for any one provider will max out at 400 Million per year. At 50,000 devices per wafer that will mean 1 wafer per hour of production which will be a small fraction of any factories capacity. With the additional optimistic assumption of 50% Gross Margin, which is highly unlikely at that low factory utilization, we have a facility buyback in the time frame of 5 years. That leaves no remaining revenue for profit and some very disgruntled investors.