EET: Why is Intel interested in MEMS?

Rao: Because MEMS is basically a silicon technology. It uses back-end process steps. It is CMOS-like with a few additional steps. MEMS plays to our strengths as a silicon manufacturer.

EET: Although not all MEMS require silicon?

Rao: Well, that depends on your definition. I think of micromechanical systems built in silicon. Some of the biological applications I don't necessarily view as MEMS. I am interested in micromachines in silicon.

EET: Why after 40 years has there been a seemingly sudden explosion of applications for MEMS?

Rao: Well, at the core is the technology; the process, the packaging, the design. But what you do is governed by the application. Always the question is, "How can I do something better?" or, rather, "How can I make money by doing something better?" So with MEMS we've had a lot of technological progress but the applications have been slow in coming, there were few of those things that needed to be done better.

You need to find miniaturized versions of macroscopic applications. The airbag [MEMS accelerometer] was one.

The big advantage that microelectronics has enjoyed has been the existence of a universal building block, the transistor. It is replicable across a die by the millions. In MEMS there has been no such universal building block; every application has a different building block. It is the same with packaging, which quickly became very standard in microelectronics but you require different packages for every application in MEMS.

The trending of MEMS is not as clean as VLSI and so MEMS have been successful only in niche areas.

Optical MEMS have been pursued because optical networking emerged as a major force in the last 10 years. It became clear that MEMS could have a role avoiding optical-electronic-optical conversions. Optical was a compelling application that spawned a big area of development.

Similarly, the 1990s saw an explosion in a lot of biological applications such as genome-mapping.

But for all that I don't think there has been a thresholding effect yet in terms of markets reaching critical mass or third-party design tools. The volume markets for MEMS haven't happened yet.

Right now we've got applications but low volumes. Manufacturing becomes a lot more seamless when you have volume.

EET: What about the scalability of MEMS compared with microelectronics?

Rao: The scalability is different. Halving the size of a mechanical structure does not make it better; it will probably make it worse. Generally in MEMS the lateral lithography is very loose, say at 1 micron. But the vertical lithography can be very tight, thin films, thin gaps. That is where the scaling is coming. You can reduce voltages by bringing things closer together.

EET: So are there advantages from integrating MEMS and electronics?

Rao: MEMS do have a scalability but it is akin to the scalability of analog and mixed-signal electronics. Therefore, putting MEMS components on a digital chip doesn't necessarily make sense. It's not cost-effective to put the two together unless there is a compelling reason. The area you take up with a MEMS device is an issue. It may be less costly to build MEMS in an older process technology and integrate within a package.

EET: What are the barriers to deployment?

Rao: As I said, finding end applications that can really benefit. And then the thing that is lagging is manufacturing experience.

EET: Intel could have made use of microfluidic cooling for its processors?

Rao: The need hasn't arisen yet. It's definitely very interesting to us but at the same time it raises issues of reliability. Applications drive what you do, but getting the manufacturing reliable in high volume needs to happen at an appropriate cost.

EET: Which route will MEMS follow? Foundry-plus-fabless or self-contained specialization?

There are already fabless MEMS companies. There are MEMS foundries. This will continue. As the demand for MEMS emerges it will drive the creation of fabs.

EET: but there few standard processes and third-party design tools.

Rao: In general, MEMS is in an immature state right now. The number of variables in a design is much greater than VLSI. Every time you do a design you are facing a finite element analysis simulation. In a way MEMS is the merging of mechanical design and VLSI design. One thing that has changed is that you can do this finite element analysis on a relatively simple personal computer these days but still....

The thing about MEMS chips is that although the integration is low the testing is like analog circuit testing.

EET: There seems to be as many processes as there are applications. Does diversity mean fragmentation?

Rao: The number of process technologies is not that large--if you restrict yourself to silicon. The release step, the etching away of any sacrificial layers to free up structures, that's the key step.

EET: Yes, but do you have structures in metal, in polysilicon, do you have multiple layers of polysilicon? A lot of MEMS processes seem to have been written around a particular application, which is not the way in microelectronics.

Rao: That's true, but if you modularize the process steps: a deep-etch module, a bulk-etch module, a surface-etch module. If you modularize like that then it's not too bad. I can imagine a single fab with all these things available. The difficulty is the integration of the modules.

Release deep-etch, thick metal deposition, double-sided wafer alignment, wafer bonding. I don't think it is anarchy out there.

EET: What are the opportunities for cross-over among plastics, glass and silicon substrates and structures?

Rao: There is some closeness between glass and silicon but perhaps not between silicon and plastic. So packaging will be the key. But new materials are coming into silicon technology.

EET: What are the most promising areas of MEMS research?

Rao: We are keeping an eye on sensor networks through a "Lablet" at University of California Berkeley under David Culler.

Also the storage area is very interesting. RF work we are pursuing internally. Then there's fuel cells, which can fall into the MEMS area.

Also, optical networking has matured greatly. I think optics and MEMS play together well because optics relies on the precision motion that MEMS can give.

EET: What is the earliest date that you think Intel could be deploying MEMS technology?

Rao: That's difficult. For us MEMS is an early-stage research project.

EET: Within five years?

Rao: Maybe. It's too early to say.

EET: Thank you for your time.