Mass market CPUs have reached the four-core stage, and there are 64-core and 80-core processors in labs right now, but for IBM, that’s still not enough. Big Blue has designs on a processor with a whopping 1,000 (or more) cores. But don’t get any purchase orders ready; this is a long-term effort that might take as long as 15 years to complete.
The breakthrough, published by a group of IBM researchers in the journal Optics Express, details how IBM used optical interconnects between CPU cores instead of copper wires to transmit data between the cores through pulses of light rather than electrical charges.
Yurii A. Vlasov, a research scientist at IBM’s T.J. Watson Research Center and co-author of the paper, said it’s not so much how fast you transmit the data but the power you have to burn to transmit it.
“When you transmit power over wires, there is some voltage, which causes chips to be hot. That is one of the problems with the ability of conventional CMOS
The technology — known as a silicon Mach-Zehnder electro-optic modulator — is 100 to 1,000 times smaller in than any previously demonstrated modulators of its kind, which could allow for complete optical routing networks to be integrated onto a single chip.
By replacing the copper wires that connect the cores with pulses of light, IBM estimates information can be transferred 100 times faster and use 10 times less power. The modulator converts the digital signal into a series of light pulses and then back to digital again on the other end of the transmission.
The process is not unlike another IBM announcement in March, of a similar breakthrough in an optical-based networking chip, and for the same reasons: bandwidth, heat and power. Vlasov said there are some similarities, at least in what they are trying to accomplish, which is moving data around.
Fred Zeiber, president of Pathfinder Research, believes IBM is on the right track in using optics for data connects. “One of the problems you have is not just throwing a lot of cores together but how you hook them up,” he said. “Clearly you get rid of some of the parasitic problems you have with going through wires, so that gives you significant power savings as well as speed.”
Don’t expect to see this technology any time soon, though. “This is definitely long term research, running five, ten, fifteen years,” said Vlasov. “This is only one piece of the puzzle we are showing. We need a lot of different devices, not only the modulator to support it.”
