SAN JOSE, CA--The Executive Panel on “Silicon Photonics and Optical Interconnects” offered SPIE Photonics West 2009 attendees an opportunity to learn from and interact with leading silicon-photonics scientists and management to debate this daunting subject. Held on Tuesday, January 27 as part of the interactive “Industry Perspectives” sessions at Photonics West, the executive panel brought together Mario Paniccia, Photonics Technology Laboratory director at Intel (Santa Clara, CA); Ashok Krishnamoorthy, distinguished engineer at Sun Microsystems Physical Sciences Center (San Diego, CA); Ray Beausoleil, principal scientist at Hewlett-Packard Laboratories (Palo Alto, CA); Eugene Fitzgerald, professor of Materials Engineering at the Massachusetts Institute of Technology (Cambridge, MA); and Jeffrey A. Kash, research staff member at the IBM Thomas J. Watson Research Center (Yorktown Heights, NY). Although opinions differed about what technologies and methods were needed to achieve faster computing speeds, one thing was clear for all: electronic interconnects have reached the end of the road.
Paniccia said that even today’s four- and sixteen-core processors weren’t enough to meet the demands of Moore’s Law, and that the real problem was not individual processor capabilities, but the problem of moving data in and out of the central processing unit (CPU). Electrical connections simply cannot achieve the processing speeds required by today’s demanding video and image-processing applications; however, Paniccia cautioned that optical solutions need to be carefully considered in terms of cost and manufacturability. His sentiments were supported by Beausoleil, who said, “The physical limitation of wires is real.” Beausoleil believes, like the other panel members, that computer companies will hit a brick wall in a few years, despite the ongoing efforts to shrink feature sizes below 45 nm through continuing lithography innovations.
Photons trump electrons
The problem with this brick wall, said Krishnamoorthy, is that cycle times for computer-processing innovations are 2–3 years or shorter, not 5–6 years as in other optical sectors. “[CPU and computing] technology has the shelf life of a banana,” implying that optical solutions to the computer-processing speed dilemma need to be developed quickly. Krishnamoorthy is principal investigator of the $44 million dollar Defense Advanced Research Projects Agency (DARPA) Ultraperformance Nanophotonic Intrachip Communication Program (UNIC)--pronounced “unique”--that focuses on microchip interconnectivity via on-chip optical networks enabled by silicon photonics and proximity communication. The project advances will be detailed in the March issue of Laser Focus World magazine. The UNIC project has an ambitious five-year-plan to achieve a viable, optical interconnect solution that at least attacks the “wiring” problem.
While all the panel members agreed that electrical connections between chips need to be replaced by optical connections, different arguments were presented for how to build the various components needed for an all-optical scenario. Paniccia said that Intel’s 340 GHz silicon avalanche photodiodes were better devices than competing III-V semiconductor devices. Fitzgerald favors the indium phosphide (InP) integrated circuits that are compatible with complementary metal-oxide semiconductor (CMOS) processing. Beausoleil says HP is building really small (3 µm diameter) silicon microring resonators in a manufacturable process that has taken them from “insane” to “only crazy.”
The light source
But when it came to the question “Is there agreement on where the light comes from?” posed by moderator Peter Hallett, manager industry relations, exhibitions, and sales at SPIE (Bellingham, WA), there was surprising agreement that the short-term (five-year) source solution would not be a purely silicon device. Paniccia said that we all agree there won’t be a silicon laser anytime soon. He added that efficiencies and milliwatts power levels, along with the inability to electrically pump a device, were hurdles to any introduction of a truly silicon laser. Consequently, an audience member asked about a true silicon laser developed at Lucent and what happened to it, but no one on the panel or in the audience could answer his question (if you can, feel free to email me at [email protected]).
Fitzgerald and Krishnamoorthy favored a III-V on silicon, CMOS-compatible light source that could take computing companies beyond the “academic” desire for a purely silicon laser, and assist in the optical-computing revolution. Beausoleil agreed, saying that 64 flip-chipped distributed feedback lasers simply wouldn’t work as an on-chip solution. Krishnamoorthy went a step further and said that he’d rather see the source off the chip, outside the data equipment rack, to avoid having to deal with source cooling or powering difficulties.
In an interview published in the January 2009 issue of Nature Photonics, University of California at Santa Barbara professor of optoelectronics and sensors and 2004 John Tyndall Award recipient Larry A. Coldren, weighed in on the silicon laser debate. “I cannot see a purely silicon laser ever being able to catch up with the performance characteristics of other alternatives that may even be compatible with CMOS integration to an acceptable degree.” But Coldren added, “The only silicon-based laser system that is actually working well is the one developed by my colleague, John Bowers, with Intel. This one uses InP gain regions wafer-bonded onto silicon waveguides. Although this is not a pure silicon laser, it does have attractive power and efficiency metrics, as well as being compatible with CMOS integration.”
Look for a condensed version of this news story and an accompanying video highlighting the executive panel members both in our online news section and on our video player at www.laserfocusworld.com.
--Gail Overton