Today’s computer chips are chunks of silicon that use electrical pulses to crunch data. But IBM researchers are now making chips for tomorrow: chunks of silicon that also contain pathways for light pulses.
These optical circuits can exchange information with the conventional, electronic circuits in the same chip. This could transport data inside a computer significantly faster, because light signals can transport larger quantities of data at higher speeds than conventional copper electrical wiring can. A chip could use its optical—photonic—circuits for high-speed input and output.
“We need faster ways to shuttle information around,” says Solomon Assefa, a member of the research team at IBM’s Watson Research Center in Yorktown Heights, New York. “Our main motivation is to build, in five years or so, exascale systems that will be 1,000 times faster than what we have now.”
Today’s supercomputers are dubbed “petascale” because their power is measured in petaflops, or quadrillions of floating-point operations per second. The U.S. Department of Energy has urged the development of machines capable of exaflops—quintillions of operations per second—to enable more powerful simulation-based research into climate change and renewable and nuclear energy.
Over the past seven years, IBM’s researchers have developed a chain of individual silicon components that together can convert a chip’s electrical signals into light signals and back again. Now they’ve found a way to build all of those components on the same chip without inhibiting the transistors’ performance, using the standard complementary metal-oxide semiconductor (CMOS) techniques used to build processors and other chips today.
Now that this goal has been achieved in the lab, says Assefa, “the next step is to transfer this to a commercial fab, like those making chips today.” Although the technology is not expected to be market-ready for around five years, IBM is keen to test its techniques on the production equipment for which they are designed.
This is a significant advance, says Bahram Jalali, a professor of electrical engineering at the University of California, Los Angeles, who helped kick-start silicon photonics when he developed the first silicon laser in 2004. “Integration with CMOS is a very difficult thing that has been a vision of many in the field for some time,” he says.
Other companies have been developing silicon photonics as well. Earlier this year, Intel unveiled a collection of dedicated photonic chips that can be used to carry data between conventional electronic chips. Caltech spinoff Luxtera puts photonic components on a silicon wafer after the electronic silicon components have been completed.
IBM’s technology can be more compact than either of these, says Assefa. “We’re integrating on the same chip as the electronics, using the same piece of silicon, for both transistors and photonics,” he says. “That means we’re able to make much finer features and build the much denser and power-efficient structures needed to target future high-end systems.” IBM’s technology can fit a photonic transceiver—able to send and receive optical signals—into a space 10 times smaller than has been demonstrated before.
That’s possible using new designs of photonic components that can be made at the same stage in the CMOS process in which transistors are etched, when lithography techniques precise to just tens of nanometers can be used. But it required some creative thinking to allow optical and electronic components to be built side by side.
For example, to create the last component, IBM researchers had to reinvent their photodetector, which receives incoming optical signals. “We wanted to use a layer of germanium, which is already used in CMOS processing, but had to find a way not to use too thick a layer, which would inhibit the transistors,” says Assefa. The team figured out that carefully spaced tungsten “plugs” in contact with a germanium layer thin enough not to harm nearby transistors gave it the desired electronic properties.
Finding ways to design very small photonic components is impressive, says Jalali, because they have typically been orders of magnitude larger than electrical ones, such as transistors. “They have done well to lower, if not remove, that particular barrier,” he says. “IBM has emerged as the industry leader at this stage.” However, he points out that further big leaps in miniaturization are unlikely. The light-carrying portions of IBM’s components have been scaled down to near the diffraction limit, the fundamental limit physics places on the size of optical components for any given wavelength of light. “That is a more difficult barrier to get around,” says Jalali.
Meta has built a massive new language AI—and it’s giving it away for free
Facebook’s parent company is inviting researchers to pore over and pick apart the flaws in its version of GPT-3
The gene-edited pig heart given to a dying patient was infected with a pig virus
The first transplant of a genetically-modified pig heart into a human may have ended prematurely because of a well-known—and avoidable—risk.
Saudi Arabia plans to spend $1 billion a year discovering treatments to slow aging
The oil kingdom fears that its population is aging at an accelerated rate and hopes to test drugs to reverse the problem. First up might be the diabetes drug metformin.
Yann LeCun has a bold new vision for the future of AI
One of the godfathers of deep learning pulls together old ideas to sketch out a fresh path for AI, but raises as many questions as he answers.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.