Silicon Chips
Harnessing optics to handle the data crunch
Photonics may hold the answer to coping with huge volume. But a big challenge remains: converting electronic data into light on silicon chips.

The take-away

  • Converting electronic data into light on silicon chips is complicated, as silicon doesn’t emit light. Scientists from Germany have developed tiny nanowire lasers that can generate and route light on those chips.
  • Lightwave Logic Inc. is creating polymer photonic integrated circuits that enable transmitters to operate up to 800 Gbps – more than the speed of data centres.

The world is on the brink of a massive traffic jam – not in vehicles, but in data. In many countries, traffic is increasing by around 40% annually, according to Beyond Fast, a report conducted by the Eindhoven University of Technology and Dutch consultancy firm Dialogic. That rate is only set to rise, with global internet traffic projected to increase nearly threefold over the next five years. Data centres bear the brunt of the strain. “Seventy-five per cent of data exchange is within and between data centres – much larger than the exchange between centres and users,” says Bert Jan Offrein, photonics manager at IBM Research.

Speeding things up

Data centres are scaling at a staggering rate, and on numerous fronts, to keep up with the rise in cloud computing demand. Every new generation of processor chip contains more transistors than the last, server boards contain increasing numbers of processors, and data centres get bigger to house more server racks. The largest data centre on Earth – the Citadel Campus in Tahoe Reno, Nevada – now covers nearly 700,000 square metres, the equivalent of 61 football pitches. The downside of storing data racks in such vast configurations is that even communicating rack-to-rack requires data to be pushed over distances and at bandwidths that current fibre-optics channels weren’t designed to support. Moving beyond a rate of ~100 Terabits per second per fibre core creates severe distortions in data signals and can lead to a phenomenon called fibre fusing, in which the fibre core melts.

Lars-Ulrik Aaen Andersen and his fellow researchers at the Technical University of Denmark (DTU) are leading the way in developing next-generation fibre optics that can handle this traffic problem. They have broken the existing transmission barrier by building optical multiplexer systems based on what’s known as a high-count, single-mode, multi-core fibre. When combined with amplifiers, the system is capable of ultra-high capacity optical transmission of 1 petabit – 1015 bits, the equivalent of around 223,000 DVDs – per second over 1,000 km. “The exciting thing about breaking this capacity crunch is that it’s not speculative theorising; we’ve been able to physically demonstrate it,” says Andersen. The system could potentially lead to a tenfold reduction in the average data centre’s cost, energy and space per bit.

Moving modulation on-chip

In order to move data through a fibre optic cable, the electronic signal on the chip must be converted to light. This is traditionally performed in-cable, meaning lasers, detectors and electronic devices that help modulate the pulses of light all sit within the cable housing. Transmission performance improves when this electro-optical conversion happens as close to the chip as possible. So research labs around the world have been looking at ways of bringing the optics ever closer to the processor. The ultimate goal is to perform the conversion on the chip itself. That would allow electrical and optical pathways to run side-by-side at the nanometre scale.

Converting electronic data into light on silicon chips is complicated, says Jonathan Finley, a researcher at the Technical University of Munich’s Walter Schottky Institut. “In computing, semiconductor physics is based on 60 years of working almost exclusively with silicon, and silicon doesn’t emit light.” Enter the new field of materials science: silicon photonics. Finley and his colleagues have developed tiny nanowire lasers 1,000 times thinner than a human hair that can generate and route light on the chip. “We’ve grown filamentary whiskers of gallium arsenide on top of silicon waveguides,” Finley explains. “The material isn’t new; it was used in the first laser ever demonstrated. However, we discovered that when you make these crystals very thin – in this case 300 nm in diameter – they perform incredibly efficiently.”

To put this performance in perspective, a typical search engine enquiry uses around 1 nanojoule of electricity per bit. The lab record using these new structures currently stands at a femtojoule per bit: “That’s a million times more energy efficient per bit of information,” Finley says. “When scaled up, the effect on IT’s global energy consumption would be drastic.” In the last four years, a number of other labs have also demonstrated optical generation on silicon chips, including Andersen’s lab at DTU. “Some of our work with on-chip laser communication is really important,” he notes. “I don’t think we’re far away from a breakthrough that will make it into a commercial product.” Finley is equally optimistic. “I’d estimate commercial rollout of on-chip lasers sooner than many people predict: around six to eight years,” he says.

Progress with polymers

In this new era of silicon photonics, polymers may make it possible to integrate photonics on both compound semiconductors and silicon platforms. Under the guidance of British photonics pioneer Michael Lebby, US-based Lightwave Logic Inc. has developed novel ridge waveguide modulators using organic polymers. The material is more temperature stable than previous designs, and provides a platform with vast potential for scaling performance and energy efficiency. And because the polymers can withstand extreme heat, they can be sprayed onto the silicon during standard chip production processes.

The modulators already have bandwidth capable of transmitting data rates greater than 50 gigabits per second (Gbps) and could be used in various 4 x 50 configurations to transmit 400 Gbps, a data rate many data centres are moving towards. The team is now working on advanced polymer structures that will enable transmitters to operate up to 800 Gbps. “Polymer photonic integrated circuits will address the challenges facing the ‘heavy data’ industry over the next decade,” Lebby says. “With high temperature stability, reliability, high performance, low power, and simple fabrication techniques, polymers are ideally suited as a vehicle to create polymer engines not just for data applications, but for healthcare, consumer, and automotive industries as well, as the drive to battery based hand-held products grows.”

A new era for photonic computing

With photonics moving deeper and deeper into computer systems, could the future of computing be completely photonic? Researchers are already imagining a future in which computing eschews electrons altogether in favour of light. “The biggest misconception around optical computing is that it would work like the computers we have today,” Finley says. “Electronic transistors perform particular mathematical operations when you input 1s and 0s. In optical computing, the transistor might differentiate something, or integrate it, or change its phase: the outputs could be far more nuanced.”

The energy efficiency of all-photonic systems is particularly interesting to engineers working on neuromorphic computing, which mimics the brain’s architecture. “Electronic computers are relatively slow, and the faster we make them the more power they consume,” says University of Exeter Professor C. David Wright. He recently helped demonstrate a fully integrated all-photonic synapse that resembled its biological counterpart. “Conventional computers are also pretty ‘dumb’, with none of the in-built learning and parallel processing capabilities of the human brain,” he says. “We tackle both of these issues – by developing not only new brain-like computer architectures, but also by working in the optical domain to leverage the huge speed and power advantages of the upcoming silicon photonics revolution.”

Sources

Bert Jan Offrein (IBM), Lars-Ulrik Aaen Andersen (DTU), Jonathan Finley (TUM), Michael Lebby (Lightware Logic Inc.), C. David Wright (University of Exeter)