MEMS leaders’ differing takes on the implications of their sector’s increasingly tight relations with its pushy big brother CMOS generated the most interesting discussion at the recent MEMS Executive Congress. With IC foundries rushing into MEMS production, IC makers looking to integrate new MEMS sensing functions into their products, and MEMS suppliers integrating their structures with IC chips and packages, MEMS is clearly moving more into the CMOS stream. But despite the irresistible advantages of the established low cost semiconductor volume infrastructure, specialty MEMS makers worried about their business getting subsumed in CMOS .

MEMS production only makes sense for IC foundries if there are high volumes, argued Richard Payne, VP of microfabrication at Pixtronix. “They can’t deal with small volumes,” he noted. “It will shut out those with small applications.”

Others were uneasy about the CMOS foundries’ long term commitment to the MEMS business. “MEMS will be the CMOS foundries’ pet rabbit,” suggested Akustica chairman and CTO Ken Gabriel. “After a while they’ll get tired of taking care of it and will turn it out in the back yard to fend for itself and forget to feed it and it will die.” He suggested the foundries would lose interest in their MEMS customers as soon as CMOS picked up, to which one wag in the audience countered, “Why do you think CMOS will pick up?”

Strong growth for MEMS foundries.

Source: Yole Developpement

With MEMS growing at 10%-50%, much faster than CMOS, it could become as much as 10% of total wafer turnover for a foundry, someone else countered.

The big CMOS foundries, with fully depreciated fabs they can convert to MEMS production, can produce at least some straightforward MEMS devices in high volume at low costs, potentially opening more markets for the MEMS devices. “If you can port your MEMS designs to TSV and CMOS, you can have a 40 to 1 volume ratio on your processes over specialty MEMS technologies. That means tremendous changes in the cost structure,” argued LV Sensors CTO and chairman Janusz Bryzek.

Jack McCauley, director of R&D at Red Octane, maker of the Guitar Hero game, noted that these lower costs meant that the three largest consumer game products now all include MEMS devices, which they couldn’t three years ago. And makers of these kind of high volume products have to use suppliers they can count on to deliver the volumes they need, and second sources.

“Maybe [the CMOS foundries] are more like two pet rabbits,” proposed another executive in the audience. “And we all know what happens when you have two rabbits.”

In fact these major production resources and cost pressures suggest that there will be no more MEMS startups that do their own manufacturing, argued Jean-Christophe Yole, managing director of Yole Developpement, whose presentation said that all MEMS startups from now on will use existing MEMS or IC foundries for production.

Except for a few cases where integration makes a crucial difference in size or performance, it’s also predominately cost that is driving MEMS makers to integrate MEMS directly on the same chip with CMOS circuitry--or not. Payne noted that though his new startup has a LCD replacement technology that offers 4x lower power, it’s still the 5% lower cost that’s mainly driving customer interest. Wire bonding the MEMS device with its ASIC in a stacked package remains cheaper and quicker than developing a single device for most applications, but math can vary widely for the vast variety of MEMS devices and producers. Payne said that back in his days at Analog Devices, the company integrated its MEMS with CMOS on a SoC, because it had the flexible factory and the designers to do so, but not the packaging technology to package separate devices together. Rival Motorola, on the other hand, had the packaging skills, so they could better do a two chip version. McCauley noted that his company in fact has to use larger components, since contract producer needs to be able to handle them with its older generation equipment.

Also ironically sometimes mitigating against one-chip integration is the increasing trend for IC makers to add MEMS capability to their electronics systems. With the system leaning more heavily towards electronics, the electronics to MEMS area may run as much as 9 to 1 electronics. “CMOS costs 5 cents per square millimeter. MEMS costs 10 to 20 cents,” pointed out Dale Gee, global product manager for MEMS at GE Sensing. “More electronics leads to less integration. The larger IC companies see ICs as the bigger source of profit, so they look at the 5 cent MEMS as a loss leader to sell the dollar IC.”

Despite the trend to move more production to foundries, MEMS makers still need some specialty tools for making some of their idiosyncratic moving structures in silicon. Of particular interest is ways to add customized processes to some sort of standard platform.

• Processes standardized at least for a class of related applications, or standard process modules, such as for example for resonators.
• Customizable standard unit processes, tied to simulation software, with settings that can be adjusted within 5%-10%, which would cut months off development time.
• Semicustom packaging for pressure sensors, perhaps a basic standard molded package platform with different special custom openings for the different environments of oil or in which different sensors operate.
• Ability to do non-planar lithography and patterning, to be able to pattern devices after a deep etch step. Machines that work in three dimensions.
• Flexibility to handle wafers that aren’t standard thickness, whether thinner, or patterned.
• Ways to measure thicker films.
• Higher throughput tools for depositing and etching MEMS’ thicker layers.
• Improved DRIE.

For additional reporting from the MIG MEMS Executive Conference click here.