As fabless MEMS manufacturers start to penetrate higher-volume markets Valerie Thompson takes a look at how the fabless model is evolving. Most important, such a business model has yet to prove itself.
No fabless MEMS maker is generating massive sales, although two manufacturers have cracked Yole Developpment's Top 30 rankings (based on sales in 2005): Knowles Electronics LLC (Itasca, Ill.), making MEMS acoustic sensors and microphones, and Intersema SA, a Swiss pressure sensor manufacturer.
The market research firm's preliminary data for 2006 sales suggests that Knowles will jump up the list, and by next year, "I expect two more fabless players, Invensense (Santa Clara, Calif.), Akustica Inc. (Pittsburgh, PA) or even SiTime Corp. (Sunnyvale, Calif.), to enter the Top 30 in 2007," said Yole's Jean Christophe Eloy. That would provide evidence of the viability of the model.
Another clue would be a fabless firm breaking the $100 million-a-year turnover barrier. Equally telling would be a merger or acquisition deal valuing a pure-play fabless MEMS startup at $100 million or more. But, said Eloy, there has been no such deal yet.
Supply and demand
Still, the nascent nature of the fabless model does not mean that participants in the value chain are not trying to make it work.
A market is in the making. On the one hand, foundries are looking for volume customers. Five or six years ago, hype about MEMS prompted investment in infrastructure, tooling and mask setups to create MEMS foundry capacity at semiconductor makers, electronics manufacturers and smaller specialized fabs.
"A lot of money was put into the ground that is now enabling this second generation of MEMS startups to take off," said Tim Wilson of venture capital firm Partech International. He is a member of the board at two MEMS startups, Invensense and Discera Inc. (San Jose, Calif.).
"The investment in MEMS capacity by semiconductor firms goes hand in hand with the shift to 300-mm foundries. What are they going to do with the old fabs?" asked Mike Leavy, managing director of Point 35 Microstructures Ltd. (Livingston, Scotland). The answer to that rhetorical question is make higher-value MEMS devices on those already depreciated lines.
And although there is a fairly steady flow of announcements regarding foundries and fabless partnerships--involving names like Dalsa (Canada), X-Fab (Germany) Semefab (Scotland), Innovative Micro Technology (Santa Barbara, Calif.), and even Chartered Semiconductor Manufacturing Ltd. (Singapore) and Taiwan Semiconductor Manufacturing Co. (TSMC), there is still excess semiconductor capacity.
"Our research has determined that the capacity of the MEMS/MST [microsystem technology] industry is approximately 40 to 50 percent in excess of its current needs," wrote the analysts at Roger Grace Associates in its latest MEMS industry report card.
On the other hand, more MEMS manufacturers are adopting the fabless model. There are three types. One is the specialist in a type of electronics--audio, for example. Two such companies are Knowles, which has been acquired by Dover Electronics, a subsidiary of Dover Corp., and Sonion MEMS A/S (Roskilde, Denmark), a subsidiary of Sonion.
Both make a range of products, as well as MEMS microphones, an application where MEMS has high potential as a replacement technology " see chart on microphone markets below.
Another type is the venture-backed startup. Some examples are Akustica, which has raised close to $30 million in venture capital, Invensense, which just raised $11 million in a second round, and SiTime. In fact, the fabless model for MEMS startups is the only one that VCs are backing, say market observers " see table below.
Even some big-name independent device manufacturers (IDM) active in the MEMS market are looking to outsource to foundries, according to Thomas Hartung, vice president of sales and marketing with X-Fab Semiconductor. "There are parallels to the evolution of the CMOS in that some IDMs move towards a fab-lite model," said Hartung, who confirmed that several large U.S. MEMS makers are in discussions with his firm. He declined to name them.
Process and MEMS law
Key for all three types of fabless MEMS manufacturer is the use of CMOS-like processes, as well as targeting higher-volume applications. This would enable them to convince foundry partners that they can help fill the partners' fabs in the first place.
The standardization seen in CMOS is still a long way off for MEMS. Yole has even dubbed it the MEMS law, "One MEMS, one process," a kind of anti-Moore's Law.
Nonetheless, most MEMS products on the market are produced in captive fabs, which means there are many processes and few standard MEMS processes that fabless companies can design for, according to a Coventor white paper. This presents a chicken-and-egg conundrum that is likely to be resolved gradually in favor of a more economically efficient fabless model.
As more MEMS are produced in independent foundries, processes will gradually standardize, increasing volume and reducing cost and making such processes more attractive to fabless MEMS makers. "The goal is to bring CMOS and MEMS closer together. And although it is already possible to run some MEMS products in the CMOS lines, the majority of processes are nonstandard," said Hartung of X-Fab, which is a foundry partner of Akustica.
The lack of standard processes means that EDA tools cannot be used to the extent they are in fabless semiconductor companies. "In CMOS, a fabless firm can design against process specs provided by the foundries, send the Cadence files to the fab and get back your die. But when it comes to MEMS, it doesn't work like that," said Mark Hesketh, cofounder and chief technical officer of Oligon Ltd. (Edinburgh, Scotland), a CMOS MEMS startup making silicon microphones, now part of Wolfson Microelectronics plc (Edinburgh).
Several vendors that provide EDA tools " Coventor, Intellisense, SoftMEMS and Ansys " have invested in new modeling, design and packaging libraries for MEMS sensors and related devices, but industry practitioners say that they still have difficulties in meeting the needs of the MEMS design community.
X-Fab's Hartung agreed, but did report some progress. "It is possible to design against tech flow and materials specs. And there are design kits for pressure sensors. We are working on similar tools for motion sensors," he said.
"Ideally, you want building blocks leading to simple devices in a fully standardized process, but to do that it means making compromises," said Oligon CEO Richard Laming.
Diversity of processes
The upshot in practical terms is diversity of processes, which means that fabless MEMS manufacturers have to find a foundry partner willing to take on their processes, and then transfer the technology to the foundry partner.
Limiting the number of fabs to choose from is a hurdle, but not an insurmountable one. "We had no problem finding a foundry partner. It is not an issue," said Oligon's Hesketh, who explained that Oligon developed its process at a CMOS fab. It's a single chip solution integrating transducers and electronics.
Discera had a similar experience: "Our CMOS MEMS device is not as unique a process as our competitors'. We've invested a lot time and effort into minimizing the number of unique processes and transferred that [intellectual property] to our foundry partner. It was a process-oriented design from the start," said Discera CEO Tom Willey. Willey's firm works with Dalsa, a Canadian foundry and Austriamicrosystems for ASICs with logistics controlled in Bangkok, Thailand.
Three partners might seem a lot to get a product out the door, but according to Willey, "it is not untypical."
Indeed, rival startup SiTime, which is also venture-backed, has a similar array of partners, including TSMC and Jazz Semiconductor as foundries, and Carsem for packaging, according to its Web site. And Memsic Inc. (Norwood, Mass., and Wuxi, China), which has introduced CMOS motion sensors, works with TSMC in Taiwan and Nantong-Fujitsu Microelectronics (packaging) in China.
"You can choose the fab based on your process and volume and level of product development," said Steve Nasiri, CEO at InvenSense. He declined to name his firm's foundry and packaging partners.
Despite some of these hurdles, there is reason to believe the fabless model will endure. For one thing, foundries do not typically undertake MEMS design, which leaves a market opportunity for finely focused design teams, and some processes are becoming more generic.
Maybe the next MEMS law will be "One MEMS manufacturer. One process."
Insiders say the fabless/foundry business model is important for MEMS development.
They see similarities and parallels, but no one is saying that MEMS will follow the same fabless path as CMOS, not least because of the much greater diversity required in the packaging of MEMS silicon.
