Yesterday in AOCs Make Faster, Lighter Connections, I discussed how active optical cables (AOCs) are, in some applications, replacing copper in high-speed links used in datacenters.
While AOCs may be able to move more data, are lighter, and produce less heat than copper cables, they require complete optical transceivers -- E/O and O/E converters -- on each end. The optical components drive up the cost of using fiber as compared to copper and thus limits fiber deployment. Figure 1 from TE Connectivity shows that as speeds increase, optics move closer to the protocol chips found in networking equipment.
A possible solution to reduce the cost of the optics is silicon photonics, where the optical components of a transceiver are fabricated on silicon. That carries the "Optical" lines in Figure 1 a step further to the left. Several companies are now producing photonics on silicon, with the most information coming from Intel and IBM.
Intel has been working with the University of California, Santa Barbara (UCSB) to develop the devices that produce, modulate, and steer light on the transmit side with laser-light sources, optical modulators, amplifiers, multiplexers, and waveguides. Optical receivers need photodetectors, amplifiers, demultiplexers, and waveguides. All of the components are fabricated using CMOS processes, which, because of their high volume, produce devices as relatively low cost
On July 27, 2010, Intel announced that it had produced a 50Gbit/s photonic link that uses four 12.5Gbit/s lanes, each lane transmitted on a unique wavelength. The company claimed that a test had a bit-error rate of ‹3e-15.
Figure 2 shows the optical interconnect between a board and an optical connector with alignment pins. The shielded area on the board contains an optical transmitter that's a bonded flip-chip. The document linked above shows this photograph annotated that points out the components.
The video below shows a demonstration of the link.
Around the same time as the Intel announcement, researchers from Sun, Luxtera, and Kotera published "Ultralow-Power Silicon Photonic Interconnect for High-Performance
Computing Systems," where they developed optical components using several processes and bonded them to silicon devices. The intent was not to produce a production product, but to prove how the photonic components could work with silicon electronics.
In the 2010 presentation linked above, Intel predicted that the data rate per lane would increase to 25Gbit/s per lane. That's now happened as described in a December 2012 IBM paper, "A 90nm CMOS Integrated Nano-Photonics Technology for 25Gbps WDM Optical Communications Applications." The paper shows the structure of the optical components fabricated on silicon and presents test results showing their performance characteristics.
Silicon photonics could prove to be the technology that puts fiber in places where only copper now goes. It could signify a cost structure that not only lightens the physical load, but runs cooler and is less susceptible to EMI than copper.
This is going to rock the semicon party. I believe the technology currently is meant for short distance communication like hand held devices or a PC to PC connectivity. Are there any long distance communication possible (I understand optical cables can carry with less loss), but still?
The mass production of Soc systems is really a key point for the success of this technology . The testing equipment has to test both the optical and the silicon part of the system, so it's an hybrid ATE.
The flip chip containing the optical transmitter in figure has to be accurately tested because this type of assembly is very sensitive to heat and if soldering has not been performed properly the whole system reliability might be affected.
@Martin: Copper is the most cheap solution for low frequencies, if the frequency increases it's necessary to switch to fibers,more expensive , but there will be an effect related to the scaling of the technology. At 90 nm the capacitive coupling between data lines on silicon will be a primary source of error on data transmission.
Martin Rowe 1/25/2013 2:05:12 PM User Rank Blogger
Re: How far back will photonics go?
eafpres, I just spoke with Finisar and mentioned RF over fiber. The person I was speaking with isn;t the right person, but we will have another call later.
Martin Rowe 1/22/2013 1:48:18 PM User Rank Blogger
Re: How far back will photonics go?
eafpres, Copper will always be there to carry power and signals that don't need the speed that fiber can bring. So for example, Even at speeds of, say, 1Gbps, you can still use copper and therefore get the PoE. Where cost really matters, copper will win as long as the cost is less than fiber.
@Martin--one more possible case for copper. Power over Ethernet (PoE) has become really popular for many applications. If that ethernet cable is replaced by an optical cable, then what?
Hi Martin--another thought. In some RF applications, like GPS receiver antennas, the LNA is located at the antenna and so DC power is required at the antenna end. In many cases this is provided by puttng the DC on the coax going to the antenna; so the coax provides power and transmits the RF signal back. This would be impossible with fiber, unless somebody makes a fiber + DC bundle. This would take away an advantage for photonics at the receiver level in this particular case.
Martin Rowe 1/21/2013 10:21:12 PM User Rank Blogger
Re: How far back will photonics go?
eafpres, and everyone else.
You can make live links. Just use the link icon under the text entry field. note that I dont see the option buttons on my iPad, but I see them on my pc.
To save this item to your list of favorite The Connecting Edge content so you can find it later in your Profile page, click the "Save It" button next to the item.
If you found this interesting or useful, please use the links to the services below to share it with other readers. You will need a free account with each service to share an item via that service.
Tweet This
