A science discipline that studies the largest of things has given a leg up to microprocessor researchers concerned with the smallest of devices.
University of North Texas scientists have developed new computer simulation software, based on an algorithm used in computational astronomy, that they claim will allow computer chip manufacturers such as Intel and IBM to design chips 100 times more quickly than they can with current software.
"Simulating chip design is a gigantic problem today. There's just no way you can overstate the challenge, and there's no question this could have a significant impact on purchasers of microprocessors," says Keith Diefendorff, editor in chief of The Microprocessor Report, an industry newsletter based in Sebastopol, California.
Simulation software is crucial to speedy chip design because the process of designing a new chip and manufacturing a prototype can be extremely lengthy in an industry where time-to-market is considered paramount. Much like the way computer simulation of a building speeds the time it takes architects to complete their schematic, integrated circuit simulation software allows chip companies to greatly decrease the time it takes to move a chip from the drawing board to the manufacturing plant to the store shelves.
Also, slow simulation software is often behind undetected flaws in chips. The new simulation software unveiled last week specifically helps engineers pinpoint which individual transistors "switch" power from one transistor to another. Thus, if chip designers fail to detect the one transistor out of 100 that switches when it shouldn't, eventually computers housing the chip will feel the repercussion down the road in the form of an occasional, unexplained system crash.
"If someone made a substantial progress in simulating chips, there definitely could be major gains that would come from that, either from getting the chips out on time earlier or on getting them more compact or in getting the circuits running faster," says Diefendorff. "It depends on how much improvement [the University of North Texas researchers] made. If they made a significant improvement on one of the key areas of simulation that is more difficult, then the impact could be huge."
The inventors of the new software -- assistant professor of computer science Weiping Shi, assistant professor of mathematics Jianguo Liu, and Naveen Kakani, a graduate student and research assistant in computer science -- are currently in the process of applying for a national patent for their research, which took two years to develop.
The team developed the new software borrowing an algorithm "that came to our attention sort of by accident," according to Shi. The algorithm was used by astronomers to tackle what is known as the N Body problem. For astronomers seeking to understand the varying forces affecting all the different particles in the universe, the billions of variables needed for an equation that takes all the particles into account is a decimal-point nightmare. But astronomers discovered a technique based on a particular algorithm that makes the N Body problem mathematically manageable.
Researchers at MIT made some of the first advances in chip simulation software about 10 years ago with a benchmark called FastCap, which the university made available free to chip manufacturers and which several CAD software companies have since used to develop commercial products. FastCap improved the speed of the chip design process by some 60 to 100 times, but the new method is "a big step forward," says Jacob White, a professor of electrical engineering and computer science at MIT and the adviser to the graduate student who produced FastCap in 1989.
White predicts microprocessor companies will begin to use the new simulation software almost immediately, while consumers will begin to benefit from the improved chips made possible by the software "a year or two down the line." Other observers of the semiconductor industry are less bullish. Diefendorff, for example, predicts it will be three to four years before consumers realize any benefit.
Such predictions aside, Shi's work in adapting the algorithm to microprocessors is just the beginning for N Body cross-pollination, according to White, who points to fields as disparate as applied physics and biomechanics as benefactors.
For his part, Shi hopes commercialization is just around the corner. "We have had some discussions with different research labs, but right now we're in the process of establishing joint research projects with companies. The wider participation the better, but we haven't reached an agreement with anyone yet."
