Robert Kahn on nanotechnology research

The Corporation for National Research Initiatives has also been an important player in the area of Micro-Electro-Mechanical Systems and Nanotechnology with its MEMS Exchange. In an extended interview with Government Computer News chief editor Wyatt Kash, Internet pioneer and CNRI chairman, chief executive officer and president, Dr. Robert E. Kahn, talked about how CNRI got involved in the supporting MEMS research and that work might impact the government technology market.

Dr. Robert E. Kahn: I think this work almost surely will have an important impact on research and prototyping activities in the country and ultimately innovation more broadly. Most of what comes into the MEMS and Nanotechnology Exchange consists of proprietary designs that organizations and individuals wish to have fabricated into devices. This is an effort that DARPA funded because they wanted to help groups that might otherwise have difficulty getting designs fabricated to get them done. It’s very hard for an organization to work with foundries to fabricate one device or even very small lots. But if you want, say, a million devices, they’ll work with you. The overhead of training a new user to use a facility for generating very small lots isn’t usually worth it.

Back in the late 1970s when I was at DARPA, I was very concerned that if we didn’t get the university computer science community to learn how to deal with LSI (large scale integration) technology, and the ability to design integrated circuits, they would be marginalized going forward. In addition to being out of touch with the latest technology for building computational devices, industry would find a shortage of adequately trained personnel. Most importantly, we wouldn’t benefit from the innovative ideas that might have resulted from the burgeoning computer science community.

So we ended up establishing a service, hosted at USC-ISI called MOSIS (which stands for MOS Implementation System). MOSIS was actually based on an earlier prototype system developed by researchers at Xerox Parc, which was based on the work of Carver Mead at CalTech. Both the initial efforts at Parc and the subsequent efforts by MOSIS showed that multiple designs from different designers could be put on a single wafer and all the designs could be fabricated in a single run. The type of process run was announced well in advance for the designers to plan their submissions accordingly. Standardized design rules made it possible for any design that adhered to those design rules to be fabricated at any of the several foundries available to MOSIS.

The economics were roughly as follows. If the total cost of a process run was $50,000, and if you had, say, 100 separate designs/projects on that run, each project would cost $500. With 10 projects, each of which took more chip area, and assuming none were replicated on the wafer, each project would cost $5,000. Either way, that’s a lot less than $50,000. And because everything was streamlined, this task could be carried out fairly rapidly, over a span of a few weeks or months. You could be a student in the school or a researcher in a research project, and you didn’t have to perform the fabrication steps, and you also got your chips back in a timely enough fashion to be useful class-wise.

For the world of MEMS, it became clear that something similar was needed, but the problem was quite a bit different. There were no standardized processes that work for all kinds of MEMS devices. Existing commercial services may not offer a process that fits the needs of a given user. For many projects, therefore, job shopping was required. In the world of MEMS, multiple process steps need to be applied to a single device in fabrication and that entire set of steps might not be available from any one commercial supplier. Changing commercial process lines for small jobs on demand is also not very practical. To make the point more explicit, you cannot take a factory designed to manufacture a refrigerator and have it manufacture a jet engine instead, without effectively rebuilding the entire process line.

Much of what was going on in the MEMS and Nanotechnology world also involved proprietary processes as well as proprietary designs. With DARPA’s support, we endeavored to help the (by now) burgeoning MEMS community gain access to innovative processes by making it possible for them to design a specialized “virtual foundry” using the MNX services over the Internet. The current (and founding) director of the MNX, Michael Huff, developed a way to do that such that people could design a process on-line, which the MNX would enable through electronic means with various fabrication sites (“fab sites”) around the country. We bid the process steps out electronically, and make sure the fab sites can do it. We come back with schedules and costing. If the user likes the price and schedule, then he typically agrees to proceed, ships us the design for that process, after which we get the masks made. Then the MNX controls the farming out of the masks to the different fab sites and eventually gets them working chips back. And then the MNX enables the wafers to be shipped from fab site to fab site, as each subsequent process step is performed, until fabrication of the proprietary design is completed.

Now, we don’t necessarily know what these designs do, but we can tell whether the process that they propose to use is a reasonable process or not. For example, if they’re going to deposit a layer of some volatile material and then put it in an oven at several thousand degrees, it’s probably going to vaporize. The MNX will generally point this out, or in certain cases refuse to proceed where it knows that damage to a fab site may occur; but, in most other cases, the primary decision to proceed would be the responsibility of the customer.