Integrated 3-D chips combining logic, memory and optical interconnects are on the horizon, but won't be available for at least 10 years, according to Bert-Jan Offrein, manager of photonics at IBM's Zurich Research Lab.
"Things will come together, eventually," Offrein said.
Speaking at a panel discussion at the SPIE Photonics West 2011 conference last week, Offrein said the development of optical interconnects is being driven by advances in supercomputing. The performance of supercomputers is increasing 10 fold every four years, Offrein said, requiring new technologies that deliver exponential increases in communications at all levels of the system.
Optical fibers are now starting to be used for off-chip communications across circuit boards in some applications. Offrein said optics would continue to be used for shorter and shorter links to fulfill bandwidth and power efficiency requirements. But to meet projected supercomputing requirements, Offrein said, optics will have to become more efficient, less expensive and simpler.
Prior to the development of monolithic 3-D chips with integrated optics, Offrein said he expects the industry to develop multimode optical printed circuit board technology and then a hybrid interim step whereby a processor and optics are packaged together in a single multi-chip package.
"I think it will be exciting to see how that will happen," Offrein said.
Other panelists agreed with Offrein, though they noted there would be challenges to overcome.
"If we can do that with the right energy and cost, we potentially have a windfall solution with silicon photonics," said Ashok Krishnamoorthy, a distinguished engineer with Oracle Labs (formerly Sun Labs).
But Michael Hochberg, an EE professor at the University of Washington, said he did not believe that it was a necessity to integrate in CMOS a photonics laser within a chip. Hochberg said the electronics industry has become "stunningly good" at multi-chip integration and suggested it would be far less challenging to integrate silicon photonics into a package with a processor at the back end of production.
"It would be beautiful to have a truly integrated silicon photonics laser," Hochberg said. "But it's a very challenging project."
Because silicon is not the right material for lasing applications, other materials would have to be integrated that enable light amplification. Direct growth of such materials on silicon is difficult because of lattice mismatch issues and the layer stress and dislocations. Therefore, most results so far are based on hybrid integration of complete laser structures or the gain material onto the silicon, according to Offrein.
Hochberg added that while much of the focus of silicon photonics commercially is for data communications, "there has been a ton of beautiful academic work on silicon photonics" for other applications.
Also participating in the panel discussion was Peter De Dobbelaere, vice president of engineering at Luxtera Inc. Dobbelaere briefly outlined the challenges overcome by Luxtera to create the company's OptoPHY line of printed circuit board-mountable optical transceivers, which combine electronics and optics on a single CMOS chip.
Dobbelaere said it was hard work for Luxtera to bring the technology "from a lab concept to something that could ship in volume." Among other challenges, Dobbelaere noted that the single chip solution had to comply with standards set by JEDEC and others.
Luxtera (Carlsbad, Calif.) has been shipping the OptoPHY devices in volume since last year. The devices are made for Luxtera by Freescale Semiconductor Inc. through a foundry supply agreement.