Indeed, a handful of companies including Cisco Systems, Intel, Luxtera—and maybe IBM—hope to ramp some sort of silicon photonics offering in 2014, mainly targeting 100G opportunities.
“There is a huge bottleneck at 100G out there,” said Martin. “The old style optics are clumsy, big and expensive, but with silicon photonics you integrates all that stuff--hundreds of piece parts--into a single chip."
Intel showed in 2010 a prototype using a similar approach to Kotura, multiplexing four signals with WDM over a single fibre. But the x86 giant spiked interest in the area last week when it announced engineering samples for a 100G product using four parallel fibres carrying 25G signals.
Justin Rattner, Intel's chief technology officer, and Arista Networks founder Andy Bechtolsheim said the 100G modules will be a huge enabler for all sorts of uses inside and between server racks in a data center.
Like Intel, startup Luxtera is preparing a 100G module using four 25G fibres. It says its products could cost as little as $250 for distances of less than two kilometers. The product will consume about 2.5W even if clock-data recovery circuits are required, said Chris Bergey, vice president of marketing for Luxtera.
Luxtera considers itself ahead because it has already shipped more than half a million 40 and 56G silicon photonics chip sets for use with Ethernet and Infiniband. It is waiting for ASICs with 25G serdes to ship in 2014 before it rolls out its 100G product, said Bergey.
Work on 25G chip-to-chip interconnects is expected to be one of the hot topics at DesignCon.
Luxtera got $21.7 million in a C round of venture funding in February 2012, and took on a board member from Broadcom. It struck a deal that same month with STMicroelectronics to make its devices in a 300mm Crolles, France, fab.
The startup uses a single external laser to save cost. It claims all other functions are handled on its chip which uses MEMS techniques.
Wonderful article! What do you refer to by " there are no standards"? 100G standard is more mature than 40G, e.g. PM-QPSK modulation is the only one adopted in 100G,while there is no standard modulation method in 40G.
First of all, it is not to be confused with 100G used in telecom which used advanced modulation to achieve serial 100Gb/s. It's for datacom 100G which can be 10x10Gb/s or 4x25Gb/s, the former was adopted earlier but slower and seems being replaced by the latter as 25G electronics are getting ready. Most early advocates of 10x10G shifted to 4x25G now. No matter for telecom 100G or for datacomm 100GBASE, no 100Gb/s electronics is actually used or needed.
I think the battle of standards is in form factor. Both Cisco and Intel rolled out their own connectors to compete with existing MPO connector for highly parallel optics. Intel and IBM&Avago even wanted to completely eliminate all current panel mount transceivers module such as QSFP+, CFP, etc. which is a big threat to many component&module suppliers. We'll see how it goes.
on the chip level, some think Si photonics optoelectronic chips are treated unfairly as all the standards are written in accordance to III-V materials. Some of these standards may not be absolutely necessary for use in practice. It's why most Si photonics products are used in AOC which only have end-to-end electrical specs. Hopefully big players like Intel, IBM can change it.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.