Copper boosts chips' speed Graphic: Moore's Law (60 KB) Published: August 13, 1997 BY TOM QUINLAN Mercury News Staff Writer A deceptively small change in chip-making techniques is poised to reassert Silicon Valley's promise of providing the world with ever-more powerful semiconductor chips at ever-lower prices. The breakthrough announced Tuesday by the Austin, Texas-based Sematech consortium involves replacing a chip's microscopic aluminum wiring with copper -- a far better conductor of electricity. That means that digital data -- the 1s and 0s that represent information -- will travel even faster through chip circuitry than is possible with today's technology. The advance is likely to help extend the life of Moore's Law, the now-famous pronouncement made by Intel Corp. co-founder Gordon Moore in 1965 that chip performance will double every 18 months. Moore's law, and its unwritten corollary that chips will at the same time become cheaper to produce, has been for three decades the locomotive of Silicon Valley's explosive growth. Over the years, the assertion has been challenged by more than a few skeptics who predicted its demise. Each time, scientists have been able to seemingly shove aside what some said were unbeatable laws of physics, shrinking again and again the minuscule lines that form transistors on silicon. And as engineers cram more transistors onto ever-tinier chips, the chips operate faster -- fulfilling seemingly insatiable demand for speedier electronic gear. Whether it's a supercomputer that helps predict hurricanes or a desktop PC that serves up a gory game of ``Doom,'' a computer's performance depends on how dramatically scientists can shrink transistors. Limits of aluminum In recent years, some in the semiconductor industry have raised the specter of a new obstacle to Moore's Law -- the limitations of aluminum used to wire together millions of transistors found in the latest generations of chips. These aluminum ``interconnects'' loomed as an especially knotty problem for makers of microprocessors, the brains of personal computers from which consumers expect lightning-fast performance. Intel's most recent microprocessor is crowded with 5.5 million transistors, but its next-generation processor, code named Merced, is expected to have nearly three times that number. Aluminum may not be able to keep up. ``When you're talking about very fast processors, like Intel's fastest next-generation designs, for example, using smaller and smaller (designs) you come to the point where aluminum is no longer an efficient conductor,'' said Dan Rose, president of Rose Associates, a Los Altos research firm that specializes in the metallurgy of semiconductor design. Copper's difficulties Although companies have been investigating the use of copper for wiring for more than 10 years, until recently the metal has just been too difficult to work with, Rose said. Unlike aluminum, copper tends to fuse with the silicon, making it nearly useless as a conductor unless it's covered with an insulating material. Using a variety of technologies drawn from different segments of the semiconductor industry, Sematech on Tuesday announced that it had been able to develop a working wafer -- the silicon disk from which chips are cut -- using copper. Sematech didn't develop any of the technology itself, but instead took different products from multiple vendors and proved that using copper was feasible. ``This probably speeds up the adoption of copper by between six and 12 months,'' said Jon Dahm, Sematech's director of interconnect research. ``It should also give our consortium members a significant technological lead in using this process.'' Sematech is a 10-year-old research consortium funded by 10 U.S. chip companies, including such Silicon Valley firms as Intel, Advanced Micro Devices Inc., National Semiconductor Corp. and Hewlett-Packard Co. With Sematech providing a proof of concept, products necessary to begin the move to copper should be in place by the end of this year, Dahm said. Intel and AMD are already planning to use copper, although exactly when and how is undetermined. Intel's current time table calls for the Santa Clara company to implement copper wiring early in the 21st century, spokesman Howard High said. Experts say that even if copper doesn't live up to its promise, some other change in chip fabrication will have to be made to compensate for aluminum's shortcomings. The rubber will meet the road, they say, when -- in their craving to cram more transistors into ever smaller chips of silicon -- companies try to construct transistors with dimensions of 0.15 microns or less. A human hair measures 75 to 100 microns across. Moore's Law rules Virtually every segment of the computer industry has staked its economic future on the successful continuation of Moore's Law. ``Every company around here is dependent on Moore's Law,'' said Dan Hutcheson, president of VLSI Research Inc., a market research firm in San Jose. ``Companies like Oracle (Corp.) and Netscape (Communications Corp.) need faster and faster chips to run their software. If Moore's Law ends, they could end too.'' But even as Sematech trumpets its achievement, chip makers have numerous hurdles to overcome before copper can be substituted for aluminum. The chip-making process is extremely delicate, and new equipment means an investment of tens of millions of dollars, so chip companies don't change their manufacturing process on a lark. Novellus Systems Inc., a San Jose maker of semiconductor-production equipment, has been developing techniques for working with copper for the past 18 months and plans to ship products to facilitate using copper next year, according to Jeff Benzing, vice president of engineering and product development. But, he said, ``even after the products are available, it will still take (chip) companies a year or two before they are ready to start implementing something like this. And even then, it will be one or two companies willing to take a chance -- and the rest will follow later.'' The addition of copper to the wafer has a multitude of ripple effects. There will need to be a new way of adding the metal to the silicon, new machines to imprint diagrams on chips and even new methods for cutting the grooves into the wafer for holding the wiring, Hutcheson noted. The problem is, none of that technology is available yet. ``You can't buy any of the products you need to do this with today. The process still has to prove itself,'' he said.

| Mercury Center Home | Index | Feedback |
©1996-7 Mercury Center. The information you receive on-line from Mercury Center is protected by the copyright laws of the United States. The copyright laws prohibit any copying, redistributing, retransmitting, or repurposing of any copyright-protected material.