But what does that mean for the companies producing the chips? Mastering lots of arcane technology and lots of headaches for the research department.
It's not just about transistors anymore: Tantalum oxide chip gates, extreme ultraviolet (EUV) lithography, new microarchitecture and better insulation are some of the developments that will come to the microprocessor arena in the next decade so that chips can continue to increase in performance according to Moore's law, Intel researchers said this week.
The oft-quoted Moore's law states that microprocessors double in power approximately every 18 months, the prediction of Intel cofounder Gordon Moore.
In addition to technology hurdles, Intel is also working on the other problem with multi-GHz chips: namely, what to do with them. The company is increasing its investments in applications, such as visual recognition software and other input devices, so that people will be able to take advantage of chip power, said Fred Pollack, director of measurement, architecture and planning for Intel's microprocessor products group. More chip families will likely result as well.
"We're going to have different microprocessors designed for different market segments," Pollack said. "Intel is already going in a number of new directions."
But dealing with these increasingly complex methods won't be easy, and more problems could emerge. "The physics of driving (chip) clock rate is fundamentally getting harder," said Nathan Brookwood, principal analyst at Insight 64.
The research currently under way largely comes as a logical by-product of the microprocessor business, according to Sunlin Chou, vice president and general manager of Intel's technology and manufacturing operations. To stay on top of Moore's law, semiconductor manufacturers have continually increased the speed and shrunk the size of chip transistors.
Annually, chip speeds in terms of megahertz have increased by 50 percent a year, Brookwood said. That could mean 4-GHz chips by 2003 and 20-GHz chips in 2008.
Unfortunately, today's technology won't allow for that to happen. The largest roadblock right now lies in lithography, the process of etching circuitry onto chips, Chou said. The light waves currently used to inscribe images into silicon are too thick to be used when microprocessor makers will have to move to the 0.07-micron manufacturing process approximately five years from now. It's like trying to sculpt eyelashes with a club.
EUV is still under development but is leading x-ray as the next candidate. "EUV will start to become important," Chou said, adding that similar roadblocks have been swept away in the past. A decade ago, after all, researchers thought they would be hitting the 250-MHz level sometime this year.
"The frequency road map has gone (further) up than the forecasts," he said. "There is at least another decade worth of scaling."
Power consumption and heat stand as other problems. Although chips have been shrinking in size, their power requirements have been increasing because of added transistors and faster speeds.
"This creates a multiplicity of problems. Cooling is one of them but it is not the dominant problem. Actually supplying power to the chip is the primary problem," said Pollack. Another problem: as chips get smaller, they begin to "leak" electricity, which can scramble signals.
To get around the issue, the company is examining different insulating materials. Silicon oxide transistor gates, for instance, may get replaced with thinner aluminum, titanium or tantalum oxide gates within four to six years, said Mark Bohr, director of process architecture and integration at the company.
Microprocessor designs, and market segmentation, will also be rethought with an eye toward efficiency and low-power requirements, added Pollack. In the future, transistors could be activated by a few electrons, speculated Dean McCarron, principal at Mercury Research. Right now, transistors get bombarded.
But does anyone need a 20-GHz chip? Not at the moment, but efforts are under way to get software developers to figure out how to harness the power. Voice recognition server applications that would let keyboard-less devices such as cell phones connect to databases remain a possibility. The company is also working on Mandarin-English voice recognition tools in Beijing, he added.
Still, "There aren't as many apps as we'd like," said Pollack.