Moore's Law meets Gore's Law at processor forum
Microprocessor makers say they can meet demand for greater processing power and energy efficiency with research aired at an industry conference Tuesday in San Jose, Calif.
Processors made from new materials or that can reduce power to individual cores as needed were among the innovations presented at Microprocessor Forum 2007, hosted by the technology research firm In-Stat.
For years, chip makers focused on making faster processors, following Moore’s Law, named for Intel cofounder Gordon Moore, that processor power would double every two years. More recently, chip makers have tried to improve energy efficiency, both to lengthen battery life in portable devices and reduce electrical use in servers and other computers. Electricity not only costs more but generating it causes pollution. Pressure to decrease power use and related carbon emissions is regarded by some as “Gore’s Law,” so called for environmental activist and former U.S. vice president Al Gore.
Intel says it can increase performance of the processor but reduce leakage — electricity that is wasted — by adding a new layer to the silicon.
It is using a material called “high-k metal gate” instead of silicon dioxide to provide better insulation to reduce leakage. The high-k process, to be used in Intel’s upcoming 45-nm processor, codenamed Penryn, is expected to use 30 percent less power, operate 20 percent faster and leak five times less electricity than a 65-nm processor, said Mark Bohr, a senior Intel fellow. The performance claims could not be verified.
“Moore’s Law is alive and well with Penryn,” added Stephen Fischer, senior principal engineer at Intel.
Penryn will also improve power management, said Fischer, allowing the computer’s operating system to reduce power to individual cores if they are not needed at a particular moment.
Intel is also researching tera-computing, in which processors could have as many as 80 cores. That could increase the flexibility to do parallel processing, or perform multiple tasks simultaneously, said Jim Held, director of tera-scale computing research at Intel.
The benefits of tera-computing, still a ways off, depend on the software and the internal memory keeping up, said Held. Another Intel researcher, Ravi Iyer, explained hierarchical memory, in which the chip could assign certain data being computed to the chip’s memory based on its importance relative to other data flowing through.
Advanced Micro Devices is researching mobile processing platform technology. Its Griffin processor, in the Puma processor package slated for introduction in mid-2008, features processors that power down when not needed, said Maurice Steinman, an AMD fellow. During his slide presentation, Steinman pointed to an image that was on a screen for several seconds. When the image is still, power can be reduced to the processors and can be powered up when he changes to the next slide.
“You want to be able to operate at the most power-efficient level and not pay for performance that you do not need,” Steinman said.
AMD is also developing Fusion, a chip expected in 2009, so named because it is intended to combine a central processing unit (CPU) with a graphics processing unit (GPU), Steinman said.
This research, while innovative, may end the era of processors as commodity products, said Max Baron, principal analyst at In-Stat.
“[This is] fresh intellectual property and very high-end physics,” Baron said, and some are years away from production.
“We’re looking at a process in the making and while the process is in the making the prices are higher,” he said. “When we see a ramp-up in technology you get a lot of advantages you never had before, but you have to pay for them.”