February 11, 2004 10:34 AM PST
Intel chip aims to cut optical costs
The chip, called a 1GHz silicon modulator, essentially takes filtered laser light and chops it into a data stream of ones and zeros that then travels down a fiber-optic cable. Because the chip made of silicon it is potentially cheap to manufacture. But it can theoretically provide performance equal to existing modulator technology that's made of more exotic materials, said Victor Krutul, senior manager of silicon photonics strategy at Intel.
A paper that describes the results will appear in Nature magazine on Thursday, and Intel plans to further discuss it at its developer forum in San Francisco next week.
"The ability to process light on the silicon itself could prove to be very valuable," said Martin Reynolds, an analyst at Gartner.
To date, silicon modulators have topped out at a fairly slow 20MHz, Krutul said. At the other end of the spectrum, optical communications manufacturers now produce modulators that operate at 10GHz. But they are made of expensive materials, such as gallium arsenide.
Current modulators also require "active alignment," which means that a technician looking through a microscope has to manually attach an optical fiber to the modulator. In Intel's prototype chip, the modulator contains a trench--put in the fiber optics, and alignment occurs automatically.
"Fifty to 60 percent of the cost (of optical equipment) goes into alignment," Krutul said.
Later in the year, Intel plans to show how its chip can hit the 10GHz level.
Still, the company's chips won't likely hit the market until near the end of the decade, Krutul said.
Although Intel is typically not associated with optical technology, the company has set its sights on becoming a major participant in the market. It has acquired a number of companies in the field, including laser specialist New Focus, and has engaged in independent research projects on photonics, the science of marrying silicon manufacturing to optical technology.
Cost is a large part of the motivation to move to silicon, but it will likely bring ancillary performance benefits for computer makers. The electricity that runs through chips creates tremendous amounts of heat, which, in turn, could crimp performance improvements. Photons, which carry data in optical fiber, do not.
Optical fiber also has the ability to transfer up to 100 terabits of data per second, Krutul said, far more than wires.
According to Gartner's Reynolds: "Electrons are quite slow. They move at a fast walking pace."
As a result of these features, Intel and companies like Primarion are trying to figure out ways to use optical fiber to connect different boards or chips inside computer.
Silicon modulators work, because electrons can change the character of light, a phenomenon known as the plasma effect. In these devices, a laser beam is split into two waveguides, channels the light travels down. The two streams of light then pass through two separate phase shifters, which are used to bombard the light with electrons.
If the phase shifters aren't turned on, the light beams rejoin, and light goes down the fiber ("1," in data terms). If the phase shifters emit electrons, the two light beams cancel each other out when they rejoin, and no light is emitted. That registers as a "0."
"With electrons, you can influence light," Krutul said.
Today's 20MHz silicon modulators use a technique called current injection. To get to 1GHz with this method, the waveguides would have to be 1/100th of a micron wide. Intel's 1GHz modulator comes with waveguides that can measure about 1.5 microns. (A micron is a millionth of a meter.)
Professor Graham Reed, an expert on silicon photonics at the University of Surrey, said current injection silicon modulators will likely be able to accomodate wider waveguides. However, no one has yet done it.
"They reorganize the charge in the device rather than injecting it, so it's quicker," he wrote in an e-mail. "It's also a device that has some similarities with a transistor, so (Intel) can make them well, too, so it's a big deal, because they've got it to work experimentally."
Much of the initial research regarding the plasma optical effect came out in the mid-1990s, when Intel and other manufacturers began to plant chips upside down on motherboards. Because the connectors faced down, manufacturers had no way to directly test if the transistors inside a chip worked.
A researcher noted that silicon is transparent at infrared frequencies. By beaming infrared light on a chip region and then studying the properties of the reflected beam, testers could determine whether electrons were traveling through certain chip regions, Krutul said. Intel eventually licensed this technique to equipment manufacturers.