In 1965, Intel co-founder Gordon Moore noted in a speech that the speed and power of computer chips had been doubling about every 18 months. That observation became "Moore's Law," and for 36 years, chipmakers have endeavored, mostly successfully, to keep up that blistering pace.
Even in the last year, while demand for chips has sunk and semiconductor companies have seen their stock values crumble to a fraction of what they were a year ago, Moore's Law has marched on. Researchers have revealed chipmaking breakthroughs on an almost weekly basis. Here's a quick look at some of the smartest--and weirdest--innovations that will be forging the superpowered chips to power tomorrow's machines.
Taking A Circuitous Path
At its most basic level, a computer chip is just a hunk of silicon with circuits carved into it. Of course, in practice, chips are incredibly complex and powerful devices, and one of the keys to that power is how precisely manufacturers can etch the circuitry.
Chipmakers carve circuits onto a chip using a process called photolithography. Manufacturers first create a stencil, called a mask, which has the circuit designs for the chip cut out of it. They then shine light through the stencil and onto the chip, which has been coated with photosensitive chemicals. Activated by the light, the chemicals etch the circuit patterns onto the chip.
With this technique, most manufacturers can print circuits as small as 0.1 microns wide, or one-one thousandth the width of a human hair. But if they want to keep up with Moore's Law, they'll have to figure out how to work even smaller.
One way researchers are doing that is by using extreme ultraviolet light for their lithography. EUV has a shorter wavelength than the light currently used, so a beam of it can be focused more tightly; thus, more transistors can fit on a chip, making it more powerful.
At Sandia National Laboratories in Livermore, Calif., Jim Folta and his team have constructed a prototype EUV lithography machine called the Engineering Test Stand, the result of a five-year, quarter-billion-dollar venture between major semiconductor companies, including Intel, Advanced Micro Devices, and Motorola, and the U.S. Department of Energy.
"We're hoping that such machines are ready in development versions a couple of years from now," Folta says. At that point, the partnership will license the technology to chipmakers, letting them build new and faster processors.
And just how much faster are we talking about? Processors built with EUV technology are expected to reach speeds up to 10 GHz by 2005. Folta figures the office-sized machine will be able to shrink circuits as small as 0.03 microns. If circuits got much smaller than that, he says, electrons would have a hard time fitting through them.
Other researchers are looking to improve circuit precision by unmasking the process. In Massachusetts Institute of Technology's NanoStructures laboratory, professor Henry I. Smith and his team are pioneering the use of zone-plate-array lithography, which drops the mask in favor of hundreds of miniature lenses like those used in spotlights. As the chip moves slowly underneath, an array of tiny, computer-controlled mirrors flicks the light to each lens on and off, and each lens focuses a tiny beam of light on the chip surface, where it etches the circuits.
If photolithography is like making circuit patterns with spray paint and a stencil, then zone-plate-array lithography "is like dot-matrix printing," Smith says. Using a prototype machine, his lab has already etched circuits as precisely as in typical lithography, and it expects to produce circuits one-fifth the thickness of those using light with a smaller wavelength, such as X-rays. And what's more, zone-plate-array lithography has the added benefit of being far cheaper than standard photolithography, where one chip may require dozens of masks, each costing as much as $1 million.
Keeping It Cool
With all those extra circuits packed onto a chip, manufacturers have to deal with a new problem: keeping them cool. Faster chips with greater numbers of tightly packed circuits and more power flowing through them tend to overheat. That can lead to system slowdowns or even chip damage. Chipmakers have traditionally used big fans inside a computer to cool chips down, but things are getting too heated for that to remain a viable method.
One solution, under investigation by a team at the University of California, Santa Cruz, involves building miniature "refrigerators" on the surface of a chip. Ari Shakouri and his team of electrical engineers build miniature cooling towers, each only a tenth as thick as a human hair, out of 200 alternating layers of silicon and a silicon-germanium-carbon compound. When a current runs through the stack, high-energy (or "hot") electrons pass through, but cold, low-energy electrons stay inside, cooling the surface of the stack.
So far, Shakouri has used the "refrigerators" to cool a chip about by about 13 degrees Fahrenheit. To be useful commercially, they'll need to chill things out at least three times that much, but Shakouri is confident that will happen. "We hope to improve efficiency to the point where these will replace conventional cooling materials," he says.
Other scientists aren't giving up on the use of fans. At the University of Colorado at Boulder, members of the Microelectromechanical Systems team have designed microscopic rotors, each with eight blades less than half a millimeter in length. Entire fields of the microfans could sit directly on the surface of a chip, working together to drive temperatures down.
Another good way to cut chip temperatures is to reduce the amount of power they use--a strategy that has the added benefit of squeezing more operating time out of batteries.
In mid-October, IBM unveiled a chip, the PowerPC 405LP, which can conserve power by shutting off portions of the chip that aren't in use. The company says the chip could lead to battery-powered portable consumer electronics that consume only a tenth of the power they do today.
Let There Be Light
One of the wilder developments in chipmaking aims to shed some light on the inefficiencies of telecommunications networks.
Telecom systems use light to send signals down optical fibers, instead of electricity down wires. Currently, that light comes from light-emitting diodes, or LEDs, that are separate components from the silicon-based chips that drive them. That keeps optical circuits unnecessarily large and complex.
But now, researchers at the University of Surrey in Britain have found a way to get silicon to emit light, making it possible to build microchips smaller and more powerful than ever before. The team, led by professor Kevin Homewood, has found that zapping silicon with a boron laser creates microscopic loop-shaped defects in the surface. The enclosed area acts like a crystal, emitting light when a charge is applied to it.
Homewood says prototype chips with silicon-based LEDs are nearly as efficient as conventional LEDs, and should improve with time. An added bonus: Manufacturing of the devices can be performed with current technology, so chip factories would require minimal updating to produce them.
"Something of this nature is of high interest in the semiconductor industry," says Tim Bajarin, president of research firm Creative Strategies. "If it really will do what the researchers say, then it's a big deal."
So now that we can pack more and smaller circuits on a chip, get it to emit light, and keep the whole thing cool, what's next?
The answer is the same one that's been applied to computers since the days of the first room-sized behemoths: make them smaller. Tinier chips mean smaller, more desirable consumer electronics; more powerful PCs, since more chips can be packed in; and increased processor speed, thanks to the reduced distance that electrons need to travel from circuit to circuit.
In May, Intel researchers said they had built the world's smallest, fastest silicon transistor. At just 20 nanometers (or billionths of a meter) wide, they're 30% smaller and 25% faster than the previous record holder. Intel plans to build microprocessors containing more than a billion of the tiny devices and estimates they'll be able to build chips that run at speeds of 20 GHz by 2007.
"All the semiconductor companies are trying to shrink chips," Bajarin says. "The next generation of chips is going to be asked to do more. Anything in this space is important to watch."