TSMC promises MEMS platform in '06

Taipei, Taiwan — After three years of research and development, foundry Taiwan Semiconductor Manufacturing Co. believes 2006 will be the year it begins to see results from its efforts to merge IC process technology with microelectromechanical systems (MEMS).

Bolstered by the success of Digital Light Processing (DLP) technology from Texas Instruments Inc., TSMC sees an increasing opportunity in MEMS, which market researcher In-Stat believes will grow from about an $8 billion market this year to just over $14 billion by 2009.

"Even though shipments of DLP products are still very low, compared to other [MEMS] products, the revenue is very distinguishable. So lots of people are taking interest in this area," said Claire Chen, manager of mainstream technology marketing at TSMC. "Once the IC foundries get into this area, it will enable the MEMS designers to be more profitable and offer more attractive product features."

The MEMS business has been dominated by the likes of Hewlett-Packard Co. and STMicroelectronics, which produce microfluidic devices for inkjet printers, and by Bosch, a sensor maker. Together, these three companies accounted for just over $1 billion in MEMS sales last year, according to Yole Developpement, a French research firm specializing in the MEMS market.

But last year, TI vaulted into the lead with $900 million in revenue from its DLP chips, the heart of a new breed of rear-projection TVs and projectors based on an array of more than a million micromirrors.

Whereas TI's hybrid approach bolts the MEMS device on top of CMOS circuitry, TSMC is hoping to introduce a more highly integrated way of joining CMOS and MEMS. Chen offered scant details on the process and acknowledged that the foundry is still working through some of the hurdles to pulling off such integration.

Both before and during transistor formation, it's difficult to successfully introduce MEMS devices into the process for numerous reasons, ranging from severe changes in wafer topography to the limited choice of materials. Major changes need to be made to the CMOS process to accommodate MEMS devices. That's why companies like ST and TI have opted for the hybrid approach, which integrates the MEMS device post-CMOS — after the back-end interconnection for the ICs is complete, but before assembly.

"For CMOS-MEMS integration, it's really a matter of understanding the way to manage the process, which could be compatible or could be totally incompatible in terms of cost, time of development and so on," said Jean-Christophe Eloy, general manager of Yole Developpement. "It takes time, because it's complex in terms of technology and complex in terms of the ability to reach the right price on the right application."

Just as TSMC is prepping its offering, a host of small U.S. and European MEMS foundries are coming into their own, after some much-needed consolidation. There are still several companies that will likely fail in the near future, Eloy said, but on balance the U.S. and European players should shift to profitability this year.

Eloy said he is seeing a lot of interest in moving some MEMS manufacturing to Asia, especially Taiwan. A few years ago, several companies here invested in building up MEMS capacity. Momentum has slowed since then, largely because of the long device development cycles. But

the island has continued to improve its offerings, through consortiums like the Advanced Microsystem and Packaging Alliance, which includes companies like TSMC, United Microelectronics, Advanced Semiconductor Engineering, Nanya, Quanta and a handful of universities and research labs.

For instance, Taiwan's Industrial Technology Research Institute has set up a prototyping lab that offers a MEMS design platform and access to a manufacturing line, which lowers the barriers for many companies that can't afford the equipment. Another key project at the lab is trying to integrate the IC and MEMS design processes. "Currently, we have to do simulations of MEMS and CMOS piece by piece, and force them together to see what happens," said Peter Chang, deputy engineering director of ITRI's Micosystems Technology Division. "An interoperable platform will make the design process easier."

If TSMC introduces a process next year, it would be a major step forward for MEMS designers, which mostly rely on a fragmented network of small foundries with proprietary process technologies. TSMC's larger resources would not only improve economies of scale, but also bring a standard infrastructure to the table. That could help shorten the typical five to 10 years it takes to bring a product to market.

Yet even before TSMC leaves the gate, some skeptics say that a common process platform applicable to multiple products may never exist because of the specialized nature of MEMS. Moreover, it's unlikely that multiple foundries would rally around a base process that MEMS designers could plug into, this camp maintains.

It would be difficult for a silicon foundry to survive in the market unless it captured a high-volume client in a product like inkjet heads, said Benedetto Vigna, MEMS business unit director at STMicroelectronics. "How many wafers are we now talking about in flash memory? Millions," Vigna said. "But for accelerometers worldwide? Maybe 100,000 to 200,000 — and one 8-inch fab can do 300,000 a year. This is why I don't believe the MEMS silicon foundry model can work like the CMOS silicon foundry."

ST has two standard processes for MEMS production. The first, called Thelma, produces accelerometers, gyroscopes, filters/resonators and some custom motion-based devices. The second, VenSen, is for the pressure and force-sensor markets.

Such standardization is unusual in MEMS. Products usually are so specialized that the process often ends up being equally so. "Each time you are developing a new product, you are developing a new process. It is just a nightmare, and it is killing this industry," said Yole's Eloy. "Reuse of an existing process is key."

ST's Vigna said MEMS designers are starting to see the need for a standard process. Designers, he said, "have to make something real and manufacturable."