Advertisement

Blog

Interconnect & Photonics: Do EEs Need to Go Back to School?

Back in the middle of August, Intel teased the communications and data industries by revealing some talks to be given at the Intel’s Developer Forum (IDF) in San Francisco. The talks that caught my attention had to do with silicon photonics and interconnect.

I reached out to Victor Krutul, Intel's director of photonics strategy, who deftly handed me to Robert Manetta in Intel's PR department. At the time, Manetta's response was “We aren't talking right now,” but he promised that I was on the list for further information. Once the IDF opened in September, we all got the updates. Intel, in cooperation with Corning, launched a multi-fiber interconnect they call MXC, along with a fiber called ClearCurve. Manetta was true to his word, and I received updates almost simultaneously to the release at the IDF. As part of my ongoing interest in the impact of optical communications on the EE profession, as well as my interest in silicon photonics, herewith is an update.

Anyone involved in datacenter design or communications is aware that interest in optical interconnect has been growing for many years. The commonly cited reasons for the less than ubiquitous conversion to optical include cost, reliability, and power consumption. Intel has been at the forefront of pursuing so-called silicon photonics — as a possible answer to all the barriers to universal optical interconnect in datacenters.

In case you missed it, way back in January, the company announced a collaboration with Facebook and Quanta Computer to develop a “disaggregated” server architecture that would separate compute, storage, and network components by using optical interconnect between the subsystems. Intel said the companies would base the program on a system on chip (SoC) Atom processor code-named Avoton.

By June, Intel’s updates were touting a 100Gbit/s IC that was a “complete integrated module that includes silicon modulators, detectors, waveguides, and circuitry.” Intel also said it believed “this is the only module in the world that uses a hybrid silicon laser.”

Putting all this together, Intel took the approach that, since its hybrid on chip laser solution runs at 1,310nm, it would design an interconnect system optimized for that wavelength. The new system offers several innovations. If you want even more details, you should read the joint Intel-Corning whitepaper, as well as the IDF presentation.

Here is one of the connectors, clearly displaying the multi-fiber interface.

Figure 1

The MXC connector will support up to 64 fibers, each operating at 25 Gbit/s. Corning says cable assemblies will be available in 8-, 16-, and 24-fiber counts in the first half of 2014 and in 32- and 64-fiber counts later.(Source: Intel)

The MXC connector will support up to 64 fibers, each operating at 25 Gbit/s. Corning says cable assemblies will be available in 8-, 16-, and 24-fiber counts in the first half of 2014 and in 32- and 64-fiber counts later.
(Source: Intel)

A simplified internal view of the connector is shown below.

Figure 2

The connector recesses the multi-mode fibers and sends the light through a beam-expanding lens to increase the diameter, shown near the exit edge of the connector body where the light beams grow in diameter. Intel and Corning claim the larger beams offer 10x better immunity to dust particles and eliminate the tolerances required in traditional connectors. They call this arrangement a lensed ferrule.(Source: Intel)

The connector recesses the multi-mode fibers and sends the light through a beam-expanding lens to increase the diameter, shown near the exit edge of the connector body where the light beams grow in diameter. Intel and Corning claim the larger beams offer 10x better immunity to dust particles and eliminate the tolerances required in traditional connectors. They call this arrangement a lensed ferrule.
(Source: Intel)

The claims for the connector/cable combination are impressive: total aggregate bandwidth of 1.6Tbit/s for the 64-fiber version. Also, having demonstrated 25Gbit/s per fiber over 300m by the IDF in September, soon afterward the company more than doubled the distance by sending 25Gbit/s per fiber over 820m. In its press materials, Intel says the “low dispersion and low attenuation enable the fiber to carry 25Gbps data” over 300m with good margin. It also says the fiber will be bent around a 7.5mm radius, making routing easier in crowded datacenters.

In the whitepaper, Intel and Corning cite Cisco’s Virtual Networking Index, which has become almost a de facto standard citation for datacenter and traffic growth. In the 2013 VNI, Cisco said mobile data traffic will exceed 10Eb/month by 2017. Eb are exabytes, 1,024 x 1,024 x 1,024Gb; the mobile data traffic is expected to exceed 10 billion gigabytes per month in 2017.

Intel says that, since most mobile traffic will be on IP networks by that time, nearly all this traffic will pass through datacenters, where it will be joined by all the nonmobile Internet traffic. Forrester Research said in a July blog post that mobile traffic may exceed nonmobile Internet traffic by this year's holiday shopping season, so we can double that figure (about 2Eb/month by end of 2013). Intel says that large datacenters have more links than several small countries' entire telecom networks, with as many as 50,000 servers and 200,000 network connections.

What does all this mean for EEs and analog designers in particular? There is no doubt that more and more optical devices will be used in computing and communication. As these devices and interconnects move on to processors, the whole game will change for all the circuits around the processors. Have you struck up a friendship with photons yet?

Related posts:

12 comments on “Interconnect & Photonics: Do EEs Need to Go Back to School?

  1. etnapowers
    October 25, 2013

    Hi Blaine, this processor is very revolutionary, did Intel's communication team say something about its qualification status?

  2. etnapowers
    October 25, 2013

    “Putting all this together, Intel took the approach that, since its hybrid on chip laser solution runs at 1,310nm, it would design an interconnect system optimized for that wavelength”

     

    This dimension of 1,310 nm is really challenging because many relativistic effects on hybrid systems on silicon may arise, so the R&D on this system involves the state of the art of nanoelectronics technology. This is really a big challenge!

  3. etnapowers
    October 25, 2013

    The volume of data cited in this blog is so high that I believe that only a photonics technology can guarantee a so great data traffic.

  4. etnapowers
    October 25, 2013

    @Blaine:I think that the question you ask in this blog is a good question but it should be updated according to the consideration that physics experts and electronics engineers have to work very closely to engineer this technology and to  make it wide diffused, reliable and modular. 

  5. eafpres
    October 25, 2013

    @etnapowers–thanks for all the good comments.  Let me try to address a few of them.

    As far as the Si Photonic hardware status for Intel; they have demonstrated completely integrated chips using what they call a hybrid silicon laser.  The light source is the key technology they are pursuing, as it is important to have a way to integrate low power lasers with the chip.  They have demonstrated this from at least June of 2013.  I'm not sure when they will ship production modules but would guess in the 2nd half of 2014.

    One point of clarification, the wavelength is 1310 nm (= 1.31 X 103 nm).  I think you might be interpreting the comma as the decimal point based on your concern with reletavistic effects.  The main effect that must be address is low transmission loss.  Most designs for Silicon Photonics are using various forms of silicon waveguides built in CMOS.  Of course there is also the change in propagation velocity in Si vs. free space, which in principle would have to be accounted for, but at chip-scale dimensions I don't think this is important.

    The predicted increase in internet data traffic is a key driver.  Some data centers use photonic modules today to convert to optical for long-haul transmission.  The current modules are expensive and consume too much power.  So inside of the data center most data are moving between servers in traditional backplanes or coaxial cable etc.  Intel is looking at the next couple of generations where (a) data from server to server will be optical vs. copper, and (b) when even the server archtecture is split up, and the communication between processor and memory, or from processor to storage, is also optical.  They are working with Facebook on (b).

  6. eafpres
    October 25, 2013

    @etnapowers–I agree with you that the near-future, while advances in photonics are first being adopted in new architectures, will require a lot of multi-disciplinary teams.  There are lots of challenges that even someone fully expert at IC design for high-end processors will have to learn new ideas to design processors with embedded photonics.  Also, the simulation and design tools will evolve, so you have the physics expertise, the electronic expertise, packaging (i.e. chip packaging may change for photonic modules), and software; and probably a few more, all have to come together.  This is potentailly an exciting time for the industry and for technology watchers.  It may also geneate new work opportunities for those with the right skills; that is part of my question as I think soon there will be higher demand for engineers with understanding applicable to photonics.

  7. yalanand
    October 27, 2013

    INTER CONNECT is the photonic integrated circuit design software platform that permits for the design, replication and analysis of incorporated circuits, silicon photonics, and optical interconnects enclosing such strategies as Mach Zehnder modulators, arrayed waveguide gratings and coupled ring-resonators. 

  8. eafpres
    October 27, 2013

    @yalanand–I agree that Lumerical's software seems to be at the forefront providing a simulation and design platform.  Of course, I expect that the top companies, like Intel, Cisco, Mellanox, Teraxion, etc. probably have their own proprietary models.  Imec in Europe also has a complete platform from simulation to fab as well.

    Compared to today's circuit model and semiconductor chip level models I think that photonic models at the chip level are very new.

    Can you tell us anything about INTERCONNECT from Lumerical?

  9. etnapowers
    October 28, 2013

    @Blaine: thank you very much for your clarification, I'm happy for you agreeing with me about considering the predicted increase in internet data traffic as a key point that will determine the success on silicon photonics technology. So Intel's idea for the next generation of data center is to utilize light waveforms in the parts of the data network that need to have a very high trasmission rate, I find this is a pretty good strategy.  

  10. etnapowers
    October 28, 2013

    @Blaine, you're absolutely right! That's a great challenge for experts of many disciplines and I believe that also the University courses should be adapted, to form a new generation of engineers having strong skills in photonics.

  11. eafpres
    October 29, 2013

    @etnapowers–Education surely is critical.  One good thing is that the possibility of Silicon photonic and other photonic-electronic integration paths has led to a large surge in University work in these areas.  This is providing a lot of top quality graduates, although probably not enough!

    You might be interested in the US National Photoinics Initiative.

    Of course, it appears to me the Europe combined efforts are years ahead with lots of EU funded work and even the governments working to get all the manufacturing for silicon photonics in Europe.  Imec in Belgium is the center of it all.  I think the programs related to all the EU work will produce a lot of PhDs who will populate the top positions in the industry for a while.

  12. etnapowers
    October 30, 2013

    @Blaine, that's the point: “This is providing a lot of top quality graduates, although probably not enough!”

    In the near future this kind of experts will be very requested but the university  courses have to be focused on mixed contents typical of various disciplines: engineering, physics, informatics…

    A direct experience into the laboratories of the companies working on silicon photonics should be a really good integration of the universities courses , I believe that this would form the experts  “who will populate the top positions in the industry for a while” as you correctly said.

     

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.