CONTACT: Stanford University News Service (650) 723-2558
Engineers will soon use 'virtual reality' to design computer chips
STANFORD -- Electrical engineers soon will be forced to use "virtual reality" to design future generations of integrated circuits.
Not only will such three-dimensional modeling become a necessity for integrated circuit design, but it also will be required for making optoelectronic devices, such as semiconductor lasers, and microscopic machines, such as the tiny accelerometers that trigger airbag inflation in today's automobiles.
Fortunately, the tools for doing this are not only feasible but practical, Robert W. Dutton, professor of electrical engineering and chief scientist at Stanford's Center for Integrated Systems (CIS), told scientists at the annual meeting of the American Physical Society on March 21 in San Jose.
"The miniaturization of electrical devices, like the semiconductor transistor, has reached the stage where two-dimensional simulations are no longer adequate. But doing three-dimensional simulations of integrated circuits is far more complicated than any of the computer graphics created for Hollywood," Dutton said.
The key to performing such advanced modeling of the electrical properties of integrated circuits is the ability to solve large numbers of special equations called simultaneous partial differential equations. Three years ago, CIS researchers received a "Grand Challenges" award in an Intel-sponsored competition at the California Institute of Technology for solving several million of these equations at a rate of more than 1.7 gigaFLOPS (1.7 billion floating point operations per second). To do so, they used a custom-designed computer, called a parallel processor, that cost tens of millions of dollars.
"It was something like proving that driving is possible with an automobile that costs as much as a Rolls-Royce. It proved the feasibility, not the practicality, of this approach," Dutton said.
Within the last few months, however, Stanford graduate student Bruce Herndon, working with scientists at IBM, has developed a software algorithm that allows even higher performance using commercial parallel computers that cost about one-tenth as much as the custom machine used just three years ago. The researchers have demonstrated the ability to solve 1.5 million equations at 9.5 gigaFLOPS using IBM's new scaled POWERparallel SP-2 system.
"This brings three-dimensional simulation capability down to a price that semiconductor companies can afford. It also runs fast enough so that designers can really use it. At these rates, 3-D will take about the same time to run as current two-dimensional simulations," Dutton said.
In the past, engineers have designed integrated circuitry by relying on vertical, two-dimensional simulations of individual transistors and the other elements that make up typical microelectronic devices. But as these elements have shrunk in size, width effects -- the interactions between the electrons and the edge of devices -- have become increasingly important. As a result, two-dimensional models have become less and less representative. According to Dutton, the new equation- solving capability will enable engineers to replace these two-dimensional simulations with full three-dimensional analysis.
"Parallel computing promises nothing less than a revolution in the way science and industry can now tackle large, complex problems," said Irving Wladawsky-Berger, general manager of IBM's POWER Parallel Division. "The exciting work at Stanford underscores the fact that high- performance parallel systems can be incredibly cost effective, and that it's entirely feasible to implement new applications with bold new visual and graphic tools that can enhance perception and lead to important new insights."
The new equation-solving algorithm has been incorporated into a new version of PISCES, a Stanford-written computer program that is commercially available and simulates the behavior of the microscopic electronic elements that are used for complex computer chips such as the PowerPC or Pentium.
The code already has been restructured to run two-dimensional simulations using parallel processors. It breaks down the millions of equations into a number of smaller problems that are solved simultaneously by the computer's multiple processors. As a result, the "parallelized" program can solve these equations more rapidly than is possible with an ordinary computer with only one processor.
Currently, Stanford and IBM researchers are working to extend the program's capabilities to full, three-dimensional analysis by incorporating techniques used in other experimental Stanford codes and IBM's well-known 3-D FIELDAY simulation code. At the same time, IBM has enlisted Stanford's help in its effort to parallelize FIELDAY.
According to industry estimates, use of even two-dimensional simulations can cut the cost of designing, testing and manufacturing a new computer chip in half, and reduce development time by as much as a year. The typical workstation that an integrated circuit designer uses has the horsepower required to solve a few thousand differential equations simultaneously, Dutton said. But even that can save millions of dollars and increase the competitive advantage of the chip designers.
The capability to do three-dimensional modeling also may increase the rate of innovation in the field. "There are some new technologies, like sub-tenth-micron CMOS, silicon-on-insulator and silicon-germanium technology, where we don't really capture all the behavior correctly in 2-D, so we really need 3-D modeling," said Ronald Knepper, an IBM researcher visiting CIS who has been involved in the project.
Not only does this equation-solving capability apply to modeling microelectronic devices, but it also can be applied to a number of other areas, including fluid dynamics and genetic engineering.
Stanford University is offering the new parallel version of PISCES under a non-exclusive license.
This is an archived release.
This release is not available in any other form.
Images mentioned in this release are not available online.
© Stanford University. All Rights Reserved. Stanford, CA 94305. (650) 723-2300.