09-22-2003, 07:10 AM
|
#1 (permalink)
|
|
Loser
|
Very cool advance: No More Boards
http://www.nytimes.com/2003/09/22/technology/22SUN.html
Quote:
September 22, 2003
New Sun Microsystems Chip May Unseat the Circuit Board
By JOHN MARKOFF
OUNTAIN VIEW, Calif., Sept. 19 — Written off lately by the computer industry as a has-been, Sun Microsystems may still have a few tricks up its engineers' shirt sleeves.
On Tuesday, Sun researchers plan to report that they have discovered a way to transmit data inside a computer much more quickly than current techniques allow. By placing the edge of one chip directly in contact with its neighbor, it may be possible to move data 60 to 100 times as fast as the present top speeds.
For the computer industry, the advance — if it can be repeated on the assembly line — would be truly revolutionary. It would make obsolete the traditional circuit board constructed of tiny bits of soldered wires between chips, familiar to hobbyists who hand-soldered connections when assembling Heathkit electronic projects.
"It could represent the end of the printed circuit board," said Jim Mitchell, director of Sun Laboratories here. "It makes things way, way faster."
Sun, an icon of Silicon Valley, has been losing market share and laying off thousands of workers as corporate computing customers turn increasingly to Microsoft and Intel for their software and hardware. Sun is in desperate need of a technical advance that can differentiate it from the others.
The new technology is being developed as part of a military-financed supercomputer effort. But Sun executives said they were seeking ways to find commercial uses quickly for a future generation of computer systems.
Sun has not decided whether to license the technology to other manufacturers or reserve it exclusively for Sun's own systems, Dr. Mitchell said. Analysts, though, say they believe that the company is moving toward a more liberal technology licensing policy.
"This is a big thought project," said Vernon Turner, vice president for global enterprise servers at the International Data Corporation, a market research firm. "It will give them some leadership if they can pull it off."
The recent resignation of Sun's co-founder, William Joy, a leading software designer and developer of the Java programming language, has been seen as evidence that the company is struggling to remain innovative. Still, Sun has maintained its research spending despite corporate cutbacks.
The new breakthrough is based on an insight by Ivan E. Sutherland, a Sun vice president and research fellow who is a pioneer of modern computing. Dr. Sutherland, 65, was a co-founder of Evans & Sutherland, an early maker of high-performance computers. He is also the inventor of interactive computer graphics.
In a paper to be presented at the Custom Integrated Circuits Conference on Tuesday in San Jose, Calif., Dr. Sutherland, Robert J. Drost and Robert D. Hopkins plan to report that they were able to send data at a speed of 21.6 billion bits a second between chips in a scaled-down version of the new technology. By comparison, an Intel Pentium 4 processor, the fastest desktop chip, can transmit about 50 billion bits a second. But when the technology is used in complete products, the researchers say, they expect to reach speeds in excess of a trillion bits a second, which would be about 100 times the limits of today's technology.
Currently, computer data is moved in and out of an integrated circuit through tiny wires soldered to the surface at special pads that ring the edge of each chip. While the pads are small, they are vastly larger than the transistors and wires that make up the chip's circuitry.
A typical gold or aluminum wire might be 25 microns in width and soldered to a pad that is 100 microns wide, about the width of a human hair. Compared with the internal circuitry, this passageway requires relatively large amounts of power. Also, the size of the pads and wires necessarily limits the number there are to ferry information in and out of the circuit.
The new Sun chip has tiny transmitters that are only a few microns in width. In addition to having many more connecting points, the chip should consume far less power. The chip's additional channels increase the processing speed, like adding lanes to a highway; being able to eliminate the pads is another benefit of the chip's design, like getting rid of a series of tollbooths.
Chip-to-chip bottlenecks have long been a vexing challenge for computer designers, who have explored many ways of increasing the overall speed of systems that are composed of hundreds of chips.
Other potential technologies have included optical lasers and even the idea of quantum entanglement of electrons, which holds out the possibility of moving huge amounts of data instantaneously.
Transmitting data between chips by placing a transmitter next to a receiver, along the lines of the Sun design, employs an effect known as "capacitive coupling" to send electrical pulses at high speed. The idea came to Dr. Sutherland when he was visiting Steve Jacobsen, a robotics expert based in Utah, who has developed a technique for ultraprecise mechanical alignment.
This technique might be applied to connect large arrays of ultrasmall transmitters and receivers, Dr. Sutherland decided.
While the concept has yet to be validated fully, Sun researchers have already received an important vote of confidence from the Defense Advanced Research Projects Agency of the Pentagon.
In July, Sun was a surprise winner of a $49.7 million award from the agency to work on supercomputer designs. Cray and I.B.M. also won contracts, but Sun was chosen over two competitors, Silicon Graphics and Hewlett-Packard.
The choice of Sun surprised many supercomputing researchers because both Silicon Graphics and Hewlett-Packard have larger supercomputer businesses.
"This is one of those things that could have great potential if they can work out the details," said William J. Dally, a professor of electrical engineering and computer science at Stanford University and a consultant for Cray on its supercomputer project financed by the Pentagon agency.
The Sun technique could pack hundreds of chips in face-to-face checkerboard fashion far more densely than is possible today. The technique holds out the hope of attaining what had been one of Silicon Valley's far-off dreams: a computer packaging technique known as wafer-scale integration.
Today, chips are manufactured in wafers that contain hundreds of identical circuits. The individual chips are cut apart and each chip is wired into a separate package. The chips are then laid out on printed circuit boards and connected to other packages by wires that are thousands of times thicker than the chip circuits.
For decades, computer designers have tried to figure out how to make computer systems out of single large wafers. But designers have stumbled over the fact that it is virtually impossible to create large wafers that are free of defects.
Now the Sun researchers may have surmounted the hurdle with a simple mechanical solution — having a bunch of small chips work together with the computing properties of a single wafer.
"This is a very novel idea that could give you a way to make a very compact computer," said David Patterson, a computer scientist at the University of California at Berkeley who is a Sun consultant. "From the very beginning people have been making circuits on wafers and then chopping them up and then wiring them back together again."
As a graduate student in the 1970's, Dr. Patterson worked on a Pentagon-financed wafer-scale integration research project at Hughes Aircraft. Even though it was not cost-effective, he said, it was one of the few successful efforts to build such a computer.
Since then, the industry has tried unsuccessfully to commercialize wafer-scale circuits aimed at avoiding chip-to-chip communication bottlenecks.
All of them have failed and several have collapsed in spectacular fashion. Gene M. Amdahl, the designer of I.B.M.'s 360 mainframe computer, founded Trilogy Systems in 1980 to build an advanced mainframe computer based on wafer-scale technology. He was able to raise $279 million from computer partners, venture capitalists and a public offering, before going under.
Dr. Sutherland acknowledged that Sun has more to do before it could determine if its proximity communication technology was viable. One issue is potential interference between the tiny transmitters and receivers. A second issue is cooling. As chips are moved closer together the challenges in removing heat increase sharply.
At the same time, Sun's computer designers said they were optimistic about the technology and were eager to consider ways of using it in future Sun computers.
"It's pretty exciting in what it has enabled," said Marc Tremblay, a Sun microprocessor designer. "As you cross boundaries between chips, that's where the pipe has been narrow."
He said that faster chip-to-chip speeds might also lead to a rethinking of the internal layout of computers in ways that would enhance performance even more.
Dr. Sutherland said that he was uncertain where the new technology might be applied first commercially, but that there was already great interest from the company's computer division.
"The news we hear from product-land is, `We want this yesterday,' " he said.
|
|
|
|
)
