To download the latest issue
Jul 21st, 2010
MEMS CTO Meeting: challenges of small markets with big demands
Get 60-some senior MEMS technologists from around the world together and what do they talk about?
Everything from how to work to make the world a better place, to how to work themselves out of a job. But the real hot button issues at this first MEMS CTO gathering, organized by AMFitzgerald Associates, ChipWorks and Yole Developpement, were how to fund development, and how get products to market faster in their tiny idiosyncratic niche.
How to fund development of the components that enable big markets?
One of the world’s largest MEMS companies thinks MEMS components is a tough business-- unless it enables much bigger business opportunities.
HP Labs’ inventor Peter Hartwell estimated that there was a roughly order of magnitude difference in market size at each level up in the value chain –a MEMS component supplier may sell 20 million accelerometers for a smart phone over the last three years at, say, $1.50 each, to earn $30 million, but the phone maker took in $2.2 billion, while the wireless carrier took in $25 billion for its service. “Thank goodness some MEMS guys think that’s a good business,” he said. “I’m afraid we won’t know how to fund the little components that enable the big systems anymore.”
HP had developed a micromechanical, phase-change memory technology that used a stepper-like micromover system to move 10,000 read-write tips to different positions for ultra high density storage. But when market researchers said there would be never be demand for more than 2GB of portable storeage, HP cancelled the project.
Hartwell then tried to figure out what else he could do with the technology, and stripped out much of the complicated structure to use the system as essentially a large proof mass with precise position sensing, for a 2-axis in-plane accelerometer with a large dynamic range from 1 µm g to >10g. But HP had no interest in developing an accelerometer, since it would never be a $2 billion business. But it did see a big business in building the sensor network, and an even bigger business in analyzing the massive data gathered by the network to give useful answers to questions. “The money is in the information,” says Hartwell. “The network information market may be a $300 billion market.”
The challenge of differentiating products using other people’s standard processes
One result of limited market size, the gathered technology executives concluded, is that the MEMS sector is likely too small for the economies of scale needed to drive standards, though, interestingly, standards were still an issue that provoked considerable discussion. “It’s just not a wafer volume business,” said one audience memberarguing that the entire MEMS market probably consumes only a month or two worth of wafer starts from a good-size 8-inch fab Lack of standards, and even of some important support infrastructure like some kinds of electronic design automation, is clearly to the advantage of the established players, who have invested in developing their own internal systems.
But it’s a problem for new players—it’s difficult to differentiate their product with a standard process, but they can’t get venture funding to develop a unique process anymore. Instead, they’ll increasingly have to figure out how to design, or re-design, their products to use someone else’s standard process. Startups will have to make their products with the existing process at one of the foundries, unless they have potential for large enough volumes to get a foundry to invest in adjusting its process for them.
And increasingly, these could be standards set by the IC industry, as it becomes the major user of MEMS’ core etching and bonding processes in its TSV ICs. IC companies are interested in MEMS technology mainly to master the TSV process, pointed out Yole CEO Jean Christophe Eloy. He noted that TSMC has been working on MEMS for ten years but still has only a $10 million business. But it has grown a $150 million 3D packaging business at Xintec in only three years. “I think they don’t care about the MEMS,” he said.
While the MEMS industry now uses the equivalent of about 3 million 8-inch wafers a year, compared to only 1 million used by the IC sector for TSVs, the IC side is fast becoming the bigger customer for these deep etching and wafer bonding technologies. Within several years, Yole figures the IC sector will run more TSV wafers than the MEMS industry uses in its total production. (Though putting number estimates on this makes this a much more compelling argument.) One equipment supplier noted that with their bigger volumes, IC companies would dictate equipment design, and MEMS companies would just have to use that—and IC companies’ practices of paying new equipment prices and buying service contracts will likely push MEMS companies towards that same model as well.
Small market means limited EDA tools, but sector doesn’t use the ones it has
The industry also unfortunately may be too small to support development of the better electronic design automation it really needs to ease development time and cost. Not only does the sector have no reference designs or IP blocks, it has only limited verified physics models, few foundry-specific design rules, little process simulation, though some design simulation and rule checking software tools. “There’s a real problem with underutilization of simulation,” says Alissa Fitzgerald, founder and managing member of AMFitzgerald, a MEMS product development firm based in Burlingame, CA. “This is killing us in time- to- market, and really hindering our ability to address new markets.”
The gaps in design automation have to be filled by experienced MEMS engineers, another problem. In the U.S., most of these come out of PhD programs, and about half of this small set of students are foreign and will have trouble getting work visas. She noted the tendency of even some big companies to just go build the device to see if it works– leading to a series of 8-week iterative fabrication cycles to develop a device.
Of course without good material properties data, the models don’t give reliable results, so fab experiments are often the only option. Physics models can’t predict stiction, electrical charging, fractures, or time-dependent materials behavior, so the simulations are often not very helpful. “If you can’t characterize the failure mechanisms, simulations don’t do you much good,” noted one participant. “But that idea has permeated the industry, so we don’t simulate the stuff we can,” countered another. “20% of the problem will still be hard, but we can do the other 80%.”
This lack of a complete EDA solution is a major hazard to early stage development, and a major advantage to the more established and bigger companies who already have developed their own proprietary reference designs and IP blocks. “This is why the big will stay big, and the small will have problems,” said Fitzgerald.
CSO Aaron Partridge noted that SiTime had developed its own simulation tools and sees them as a core competitive advantage. Another startup reported developing its own simulation tools that worked great to develop its prototype, but were of no use for the transfer to the foundry because of different tools, different processes, and even different process parameters that were controlled and measured—the foundry couldn’t tell why some runs worked and some didn’t because the key factors turned out to be parameters that the foundry didn’t typically measure. But using simulation on film characteristics from the foundry later saved the 3-4 months of experiments.
Of course in the IC world, the major EDA software was developed as a matter of necessity because there were too many gates to possibly design any way but by computer, and the EDA developers had a large market for their product—which can still cost $2 or $3 million per seat. For a total MEMS industry that may total 100 companies, and only 20 of those of significant size, the total market of companies who can afford EDA is tiny, and realistically probably not large enough to ever support development of good tools. Worse, MEMS is harder to design than ICs, since MEMS developers are varying not just the design, but both the design and the process.
But some intriguing first steps towards possible improvements were suggested. SoftMEMS CEO Mary Ann Maher noted that one major MEMS company had commissioned a wafer mapping program, then had SoftMEMS offer it to the market so others would contribute improvements. Process Relations CTO Dirk Ortloff suggested his company’s software worked as a kind of MES for the development process, providing a common platform for handling data from the different major available EDA tools, so information from across all programs could be searched and queried.
Box or sidebar: Notable and quotable
The main reason to work is to help Humanity, argued SiTime CSO Aaron Partridge, reminding his colleagues that “Accelerometers and gyros have saved 100,000 lives from airbags and ESC systems, to say nothing of the impact of medical applications.”
What does a CTO really do? Partridge sees it as a continually evolving process. When the company is in the idea stage, the CTO’s role is to focus on selecting the right idea: “Yes this is cool.” In the startup stage it is to guarantee the technology: “Yes, it will work.” In the development stage it is to be the jack of all trades: “Yes, I can help design that.” Then in the mature company it is to be the sage advisor: “Yes, that is a good direction.” And finally, the job may be to realize when it is time to quit and move on to something else, to avoid becoming the useless founder parked in the CTO position without anything to do.
Paula Doe for Yole Développement
For more information, please contact Sandrine Leroy (firstname.lastname@example.org)
More MEMS CORPORATE/FINANCE news