Intel showcases 50 Gb/s silicon link

28 Jul 2010

Technology giant makes significant advance by integrating four DFB lasers based on the hybrid InP design first developed in 2006.

Intel has again pushed the boundaries of photonics, this time with the demonstration of a 50 Gb/s optical link based almost entirely on silicon devices.

The high-speed link, developed by an Intel team led by Mario Paniccia and showcased at the Integrated Photonics Research conference in Monterey, California, features four sets of lasers, modulators and detectors. Each operates at 12.5 Gb/s to give an aggregate speed of 50 Gb/s.

The only parts of that link that aren’t based entirely on silicon are the four “hybrid” lasers that are integrated within a single transmitter chip. Since nobody – not even Intel – has yet persuaded silicon to emit light at any useful level of efficiency, the hybrid design instead incorporates multiple pieces of indium phosphide (InP) semiconductor material, which are bonded to an 8-inch silicon-on-insulator (SOI) wafer.

Gratings etched into the SOI wafer act as mirrors to reflect stimulated emission back into the InP material, which also amplifies the light.

Changes to hybrid design
The hybrid laser design has changed significantly since its debut in 2006. John Bowers, now director of the Institute for Energy Efficiency at the University of California, Santa Barbara (UCSB), and a key part of the hybrid laser research team, told optics.org:

“The original laser was a Fabry-Pérot laser. The key aspect of this announcement is integration – four distributed feedback (DFB) lasers.”

And while Intel has not released specific performance details of the laser components used in the hybrid transmitter, Bowers added that their power was much higher than the original design, with a lower threshold current and a much higher maximum continuous-wave temperature. “The demonstration did not include (nor require) a thermoelectric cooler,” Bowers said.

One of the biggest changes relates to the operating wavelength of the hybrid lasers. Whereas the original hybrids emitted at 1550 nm, the latest demonstration features four lasers spaced at wavelengths around 1310 nm, with Intel saying that the different wavelengths are produced simply by altering the pitch of the gratings inside the SOI wafer.

The key challenge for Intel now will be to develop a high-yielding semiconductor process, although Bowers says that this will be aided by the switch to a coarse wavelength division multiplexing (CWDM) approach. With CWDM, the emission wavelength stipulations are less strict than with dense multiplexing, meaning that it is easier to yield a high proportion of lasers emitting within the desired wavelength.

Silicon photonics in anger
Graham Reed, head of the electronic engineering department at Surrey University in the UK, and a specialist in silicon photonics, believes that Intel’s latest work is of great significance to the entire field, in particular the integration aspect:

“This shows several things,” Reed told optics.org. “First, a technology giant like Intel is committed to using silicon photonics in anger! This is well on the way to [being] a product.”

“It is also the first time that all the Intel components including the laser have been integrated together. Even though the laser is not a “silicon laser”, this is as close as it could get today,” Reed continued.

“Luxtera were previously the only company to do any significant level of integration, but it’s different when a huge company like Intel commits, despite the excellent work of Luxtera.”

Reed believes that a key challenge for Intel now will be to develop an efficient way to test circuits at the wafer scale, to minimize production costs and maximize yields. He also thinks that although Intel might steal the headlines from others working on silicon photonics, such as the EU’s new silicon photonics cluster, the latest work will have a galvanizing effect on the field.

He cites the Euopean HELIOS project as one example of a different approach to producing transceivers based on silicon photonics, and says that some of the preliminary results from the UK Silicon Photonics project also look “very promising”.

“Historically, the Intel group has been very supportive of work around the world, so I suspect that they see work in Europe as healthy competition, and healthy for the entire field of silicon photonics.”

“What they have announced supports the legitimacy of the work that researchers are doing, because it shows that there will indeed be real applications.”

Intel envisages a radical change in the architecture of datacenters and supercomputers, once photonic links are economically viable and able to replace traditional copper cables. “Tomorrow’s datacenter or supercomputer may see components spread throughout a building, or even an entire campus,” suggested the company in its release highlighting the 50 Gb/s breakthrough, adding that likely benefits would be significant cost, space and energy savings, as well as the substantial increase in performance.