1

Why is it that optical interconnects require so much less power than electrical ones?

update from comments:

This question was prompted by this video. The speaker shows a table at 50:43 that indicates that optical interconnects require much less power for the same port speed. He also mentions at other points in his talk that photonic integrated circuits require less power.

enter image description here

Description of the video:

Published on Dec 27, 2013

(Visit: http://www.uctv.tv/) Photonics has transformed our work and, indeed, our lives, by enabling the Internet through low-cost, high-capacity fiber optic transmission. In data centers, photonics is replacing electrical cables, thereby allowing for higher and more economical performance. In his talk, John Bowers, professor of electrical and computer engineering and materials, discusses the evolution of photonics and what the future holds for this technology. Recorded on 10/3/2013. Series: "Institute for Energy Efficiency" [12/2013] [Science] [Show ID: 25995]

Community
  • 1
okarin
  • 69
  • 4
  • 3
    What is the basis of your supposition? Where did you get this idea? – Michael Karas Dec 01 '14 at 01:41
  • What @MichaelKaras said. Why do you believe this? – Connor Wolf Dec 01 '14 at 01:53
  • 3
    With current technology, optical interconnects require less power at certain data rates and certain link distances, but it's certainly not generally true that optical interconnects require less power than electrical ones. – The Photon Dec 01 '14 at 02:18
  • @ConnorWolf The speaker in [this video](https://www.youtube.com/watch?v=GcNyyeAeoJ8) shows a table at 50:43 that indicates that optical interconnects require much less power for the same port speed. He also mentions at other points in his talk that photonic integrated circuits require less power. – okarin Dec 01 '14 at 02:51
  • @MichaelKaras See my response to Connor Wolf above. – okarin Dec 01 '14 at 02:51
  • 1
    I'm sure that, from a theoretical standpoint, you can design a hypothetical optical interconnect that impressively outperforms *current* planned copper interconnects. That assumes there are no improvements in copper interconnects. Presumably, this requires more research, plz support our next grant cycle, etc.... – Connor Wolf Dec 01 '14 at 03:49
  • @ConnorWolf Do you know where I could find more information on the theory behind this hypothetical optical interconnect? – okarin Dec 01 '14 at 03:50
  • @okarin - Beats me, I was just being cynical about academia (everything is realizable on one *more* grant cycle!). Maybe dig up papers the professor in that lecture has published? – Connor Wolf Dec 01 '14 at 05:17
  • 2
    okarin - it's pretty basic physics. The transmission of a signal through effectively zero resistance can go a LONG way with little energy through optical material, especially if the light conductivity/optical permeability is very high (which is why fibre-optics is used). To push a communications signal through 1km of copper cable would take far more power (read: current, voltage) than a little light blinking through a glass strand. – KyranF Dec 01 '14 at 07:18
  • 1
    @KyranF - The greater efficiency of fibre-optics in communications transmission lines is not just down to lower "direct" losses (attenuation over distance), it's also very significantly helped by the fact that fibre-optics are not subject to interference, crosstalk, inductive coupling, etc. etc. all that stuff that plagues electrical signals. FO does have its own issues with distortion of course, but certainly over short-to-medium distances it's vastly "cleaner" then electrical transmission. – John U Dec 01 '14 at 12:24
  • @KyranF Isn't there still high loss in fibre-optics? – okarin Dec 01 '14 at 16:18
  • @JohnU Do you know where I could find more information about this stuff? – okarin Dec 01 '14 at 16:20
  • 1
    @okarin, fiber losses for single-mode fiber in the 1550 nm band tend to be about 0.1 dB/km. – The Photon Dec 01 '14 at 17:01
  • 1
    As for a source for more information, try Jeff Hecht's *Understanding Fiber Optics*. Or start with the Wikipedia page on [Fiber Optics](http://en.wikipedia.org/wiki/Fiber_optics) – The Photon Dec 01 '14 at 17:02
  • 1
    As well as Wikipedia etc., I've posted on this before in a little more detail: http://electronics.stackexchange.com/questions/111126/what-limits-the-speed-at-which-information-is-sent-received-over-a-fibre-optic-l/111145#111145 – John U Dec 01 '14 at 17:12

1 Answers1

3

It helps to watch the presentation starting at least as far back as https://www.youtube.com/watch?v=GcNyyeAeoJ8#t=2785 to understand what the premises are for that awesome slide in the question. The assumption, but not fact is that in 10 years time there will a reduction of 1000x in the power used by hybrid microring, silicon lasers; essentially these are lasers built straight onto the Si wafer and their hope is that in 10 years these lasers will be much smaller and use 1000x less power than the current experimental devices. Assuming that happens, then you'll see the awesome power reduction in supercomputer/cluster interconnects using optical interconnects based on such lasers (from the slide posted here). I'm no expert on lasers (in general), so I can't say if that 1000x reduction in power in 10 years is a reasonable prediction for such devices, but their hope appears predicated on an analogy with other silicon photonics (non-laser) device improvements, which apparently have reached that level of efficiency; that is made more clear on the subsequent slide https://www.youtube.com/watch?v=GcNyyeAeoJ8#t=3259. Presently these lasers appear to be not much better than using copper if the black lines are for copper in that slide (which is unclear). Frankly hoping for more depth of answer here on SE is a bit unreasonable, given that there only a few experts on hybrid silicon lasers in the world (the researchers actively engaged in this) and even they are making predictions rather than going by present day tech for those improvement claims...

Although I know little about lasers, I do have hands on experience in supercomputing [which is frankly why I got interested in this question], and based on my experience I'm rather skeptical of 1000x speed/power improvements in interconnects 10 years from now...

Fizz
  • 14,355
  • 2
  • 43
  • 97