Supercomputing Tools and Technology
Tony Vacca
Tony Vacca is the Vice President of Technology at Cray Research, Inc., and has responsibility for product and technology development beyond Cray's C90 vector processor. Tony has had over 20 years' experience with circuit design, packaging, and storage. He began his career working at Raytheon Company as a design engineer, thereafter joining Control Data Corporation. From 1981 to 1989, he was the leader of the technology group at Engineering Technology Associates Systems. Tony has a bachelor of science degree in electrical engineering from the Michigan Technological Institute and has done graduate work at Northeastern and Stanford Universities.
The supercomputer technologies, or more generally, high-performance computer technologies, cover a broad spectrum of requirements that have to be looked at simultaneously at any given time to meet the goals, which are usually schedule-driven.
From a semiconductor perspective, the technologies fall into four classes: silicon, gallium arsenide, the superconductor, and the optical. In parallel, we have to look simultaneously at such things as computer-aided design tools, under which is a category of elements that get increasingly important as microminiaturization and scaling of integration rise.
Also, we have to look at the packaging issues, and there are a lot of computer-aided design tools that are helping us in that area. As was
discussed earlier, the issue of thermal management at all levels is very crucial, but the need for performance still dominates; we have to keep that in perspective.
Silicon is a very resilient technology, and every time it gets challenged it appears to respond. There are a lot of challenges to silicon, but I don't see many candidates in the near future that are more promising in the area of storage, especially dynamic storage, and possibly in some forms of logic.
Gallium arsenide has struggled over the last 10 years and is finally coming out as a "real" technology. Gallium arsenide has sent some false messages in some forms because some of the technology has focused not on performance but on power consumption. When it focuses on both, it will be much more effective for us. Usually when we are applying these technologies, we have to focus on the power and the speed simultaneously, especially because we are putting more processors on the floor.
The optical technology, from our viewpoint, has been used a lot in the communications between various mediums. When people talk about multigigahertz operations, I have some difficulty because I'm fighting to get 500-megahertz, single-bit optics in production from U.S. manufacturers. When people talk about the ability of 20-, 50-, 100-, and 500-gigabit-per-second channels, I believe that is possible in some form, but I don't know how producible the concept is.
Cryogenic technology was fairly successful several years ago. Cryogenic technology is a superconductive Josephson junction technology that also needs significant help to bring it to viable production; to apply it, one needs a catalyst that the other two technologies are beginning to have.
Interestingly, there may be some people that believe that if you escape an architecture that is massively parallel, you can escape advancing technology. I think switching does not change the focus at all to a need for high-performance technology because the massively parallel entry points are the points of entrance and not the points of completion. If we lose focus across a 10-year span, we will have fooled ourselves into believing that we have defocused semiconductor technology, logic, packaging, or interconnect technology. I think we will have to have high-performance technology to stay in the leadership position that we are in.
In the interests of our position of leadership, I have been keeping what I call a U.S.-Japan technology "score card" (Figure 1). Design tools, packaging, integrated-circuit chips, and semiconductor capital equipment are the primary technology categories. The boxes under those categories in Figure 1 indicate where I think we are, relative to Japan, in
Figure 1.
USA-Japan technology "score card."
these particular technological areas. These are key areas that I think we sometimes avoid developing. However, we must concentrate on these areas and areas of technology like these because they are the basis for developing the technologies we can build products from.
We cannot select technologies "à la carte" and discount other technologies. Technologies must be selected and balanced against one another. In the past, we didn't focus so much on supercomputer technologies because supercomputers a few years ago were alive and well, and a few companies were carrying the ball. A few years ago we didn't focus on semiconductor technologies because they were alive and well, and we were doing such a good job.
Now there is the capital issue that I think is very significant. If you consider the extent to which major semiconductor suppliers in the U.S. today depend on foreign capital equipment for getting their jobs done, then you appreciate that we are facing a very crucial issue.