News & Analysis
Comment
R_Colin_Johnson
R_Colin_Johnson
I believe the photodiode is made from germanium.
Intel demos 50-Gbit/s silicon optics
R. Colin Johnson
7/27/2010 1:30 PM EDT
PORTLAND, Ore. -- Silicon photonics will replace copper connections in everything from supercomputers to servers to PCs, according to Intel Corp. researchers who demonstrated 50-Gb/s optical transmitter and receiver chips that it plans to scale up to terabit-per-second speeds prior to commercialization.
"This milestone marks the beginning of silicon photonics in the high-volume marketplace, in applications from [high-performance computing] all the way down to the client" PC, said Mario Paniccia, director of Intel’s Photonics Technology Lab. "We call it a concept vehicle, but we've done the key things that would need to be addressed to commercialize it," added Paniccia. "We see a clear development path from 50 Gbits per second today to a terabit in the future."
Optical connections can operate over longer distances than copper wires, according to Intel, and could eventually replace not only the copper connections between systems, but those between boards in the same system and eventually between cores on the same board. The chip maker already has a 10-Gbit/s Light Peak chip that uses conventional optical technologies. Intel's Photonics Technology Lab is leveraging its silicon manufacturing expertise to build photonic components. Paniccia estimated that the first commercial applications of silicon photonics will begin appearing in as little as five years in data centers and supercomputer facilities.
Intel's concept vehicle is based on technologies developed since 2004. The modulators required to encode optical information using signal waveguides and photodiodes were cast in silicon on custom chips designed by Intel. The transmitter chip uses Intel's hybrid silicon laser technology that bonds a small indium phosphide die to on-chip silicon waveguides. It is then patterned into a connected optical laser, four of which were placed on the transmitter chip.
"We combined our silicon manufacturing techniques with our hybrid laser, and demonstrated an integrated transmitter using four lasers each operating at a different wavelengths and four silicon modulators each operating at 12.5 Gbits per second, then combined them together into an aggregate 50 Gbits per second into the optical fiber," said Paniccia.
The optical fiber output on the receiver chip was filtered into separate colors, then diverted by waveguides into four separate photodiodes, each of which recovered one of the four separate 12.5-Gbit per second channels.
Next, Intel plans to add more lasers per chip, increase the number of channels and focus on optimization, power reduction and improved efficiency. It then hopes to commercialize the optical connection technology.
|
|
|
|
|
Navigate to related information
R_Colin_Johnson
7/27/2010 2:23 PM EDT
How much weight do you think can be shaved from mobile devices by switching to internal optical connections instead of copper? If you compare the tiny size of current mobile devices to non-electronic devices, their heavier weight is arguable there most distinctive quality. The heavy weight of mobile devices comes from all the copper used internally for interconnections, but by switching to plastic optical connections, their weight should be greatly reduced. How much weight do you think can be shaved from mobile devices by switching to internal optical connections instead of copper?
Sign in to Reply
KB3001
7/27/2010 2:34 PM EDT
Intel is banking on silicon-level innovations like these to address the Von-Neumann bottleneck, keep Moore's law going and keep their main competitive advantage (x86 compatibility). I worry however about the timelines announced "... begin appearing in as little as five years...". Anything can happen in such period.
Sign in to Reply
Luis Sanchez
7/27/2010 5:06 PM EDT
Even so the efforts in reaching higher speeds, the limit is at the frontiers. The conversion from light to electrical signal and vice versa will be the bottleneck until this hurdle is bypassed there is not much speed that can be overcome.
Unless of course the all light-electronics are achieved. Or may we say, “lightronics”?
Nevertheless, it's a great job the work being done by Dr. Paniccia and his team. And looks like before quantum computers are here, silicon optics will take a moment in the industry. I suppose this will get all the way to our mobile devices... will USB cable become an optical fiber?
Sign in to Reply
chanj
7/27/2010 7:20 PM EDT
If the innovation is indeed cost down enough, it will give a great push to today's computing, not only in the weight reduction but also in the speed improvement. It improves the speed from CPU to peripheral. The cost of using fiber optics as the last mile will be a lot more reasonable. The cost of fiber optics will lightly come down as the shipping volume is higher. I can't wait to see the commercialize of the innovation.
Sign in to Reply
alfred95
7/28/2010 12:10 AM EDT
The laser itself is InP based. It does not say what is made of photodiode. As long as it is involved III-V semiconductor, I worry about its yield as well as size shrinking. Therefore, I have some doubts about the future of this technology.
Sign in to Reply
R_Colin_Johnson
7/28/2010 4:49 PM EDT
I believe the photodiode is made from germanium.
Sign in to Reply
goafrit
7/28/2010 7:03 AM EDT
There are issues with polarization and I am very happy that Intel is leading on this. Optical Interconnect is not a slam dunk, there are many ways to go before commercialization. But if they pull this out, it could be the end of AMD and a one-microprocessor world. That is what it looks like as Intel is pushing really hard into the future than AMD. During my PhD I worked on this technology, it has many promises, but it is not yet for prime time because you have to look cost also.
Sign in to Reply
gronk
7/28/2010 9:50 AM EDT
Are the InP lasers VCSELs?
Sign in to Reply
R_Colin_Johnson
7/28/2010 4:54 PM EDT
No. Read about Intel's laser here:
http://bit.ly/NextGenLog-c8aB
Sign in to Reply
Silicon_Smith
7/28/2010 2:16 PM EDT
Interesting read and comments! I am not too sure how or if at all the Von-Neumann bottleneck can be addressed through this technology. Also, it seems highly unlikely that this will actually percolate to the more ubiquitous applications, for instance PAN connections!! Essentially, this will really make the supercomputing and "supernetworking" much more efficient.
Sign in to Reply