OIDA Photonic Integration Forum

The next speaker, Meint Smit, of the Technical University of Eindhoven, brought a touch of Dutch pessimism to the conference. He said that neither InP PICs nor silicon PICs could get beyond 1000 devices on a chip before the power consumption of the lasers would make the devices impractical. He presented an alternative, based on "digital nanophotonics," two tiny "microlasers" working in tandem to generate a single lightstream with far less power dissipation. He has labeled this technology "indium phosphide membrane on silicon" or IMOS, and said it allows for a much larger number of lasers per chip. His prediction was that IMOS technology could push PIC complexity to 100,000 devices on a chip.

Both IBM's Clint Schow and Sun Microsystems' Ashok Krishnamoorthy focused on the same problem: the growing power consumption of high-performance microprocessors in today's multithreaded, multicore computer servers. According to Schow, IBM's Cell processor (developed in partnership with Sony and used in Sony videogame controllers) consumes 100 W of power, with 30% of that total going to moving data either on and off the chip, or between the cores, buffers, and memory on the chip. Krishnamoorthy said that the growing power of servers in the data center has created a situation where power consumption has reached levels of 10 kW per square foot. Adding more threads and more cores simply adds to the need to move data around the processor chip or between chips, adding further to power consumption. "This is not sustainable. We are in a vicious thermal cycle," he said.

Both men said substituting optical interconnects for copper can make a big contribution to reducing power consumption. Krishnamoorthy said Sun has built a chip with 256 VCSEL lasers, each one operating at 1.25 Gbit/s and consuming just 12 W. Cost is also critical, so an optical interconnect needs to be able to move 1 Gbit/s of data 100 meters for a cost of around $2.

Schow presented an equally ambitious vision, not just optical interconnects based on PICs, but an "optical printed circuit board," based on polymer waveguides, onto which PICs and other optics devices can be placed on a volume production line just the way electronic boards are built today. He said IBM has demonstrated transceiver chips with 16 lasers and a data rate of 160 Gbit/s, based on hybrid technology. He said III-V materials are needed for the optical reach, and better manufacturing techniques need to develop, including better packaging techniques to align arrays of fiber to arrays of lasers. "Integration is key, and we need the right mix of hybrid, and monolithic," Schow said.

Eli Yablonovitch of UC Berkeley, and an adviser to California start-up Luxtera, described Luxtera's first product, the Blazar fiber optic cable, and showed that this company is very focused on cost. Blazar looks something like a USB cable, but inside the connector at each end of the cable is an optoelectronic transceiver based on silicon photonics technology that converts 4 x 10 Gbit/s of data from electronics to optics and back again. It has low power consumption and cost approaching that of copper. To develop this product, which is scheduled to begin volume shipments this quarter, Luxtera made significant innovations in grating couplers and developed a high-yield, high-volume manufacturing process. Applications are initially in high performance computing, but according to one conference attendee, the optical cable also has consumer electronics applications. If you need to get that HDTV signal from the cable box out to the backyard, this could be your answer.

NOTE: The views in this blog are the author's own and do not represent those of Infinera or any other organization.