I was under the impression that electrons moved at the speed of light, but according to [1], they "only" move at 2200 km/s, which is less than 1% of the speed of light (~300000 km/s).
I thought that too. My mistake was in confusing the speed of the "wave" or signal (which travels at about 50% to 99% the speed of light) and the speed of the electron, which is quite slow.
I think I'm confused about waves because of early childhood text books that had person A's mouth, then some lines increasing in size (like the lines on WIFI symbols) and then person B's ear. I now think of waves as moving forward, pushing forward, rather than up and down.
Is there a version of "Maths for mums and dads" that covers science?
The drift speed of an electron in metal is actually much, much slower, somewhere on the order of millimeters per second. The electromagnetic wave in metal propagates a lot faster, and it's usually pretty close to the speed of light in a vacuum. The link you posted was talking about the speed of an electron in an atom or something.
If you imagine a pipe filled with marbles and you add a marble to one end, a marble will pop out the other end instantaneously (i.e. the speed of light).
A fun little caveat of that fact is that if you were able to make a 100% incompressible material you would violate causality. If the glass in your marbles did not compress at all then the "information" that you inserted your marble would reach the other end before it actually happened (faster than the speed of light)
Actually, that particular hypothetical would happen at the speed of sound in marbles (plus some corrective factors due to the interfaces between the individual marbles).
But your intuition about propagation vs drift velocity is correct.
If you touch a marble at the other end, you will feel the vibration at the other end with the delay equal to the speed of sound (and also hear the vibration when it's transferred to air). The speed of sound is the propagation speed of mechanical energy in a matter.
The speed of sound is the speed at which any disturbance in the material propagates. When you push on a marble, that's a disturbance. The other end of the marble begins to move when the disturbance reaches that end. In this scenario, the disturbance is then transmitted to the next marble, and so on in that fashion, all at the speed of sound.
A mechanical impulse travels through a solid via atoms literally pushing and pulling each other. Pushing something is a mechanical impulse; sound is a series of mechanical impulses. The propagation speed is the same because they are the same thing.
Perhaps it's simpler to think of the other way - why the speed of sound is in marbles is the same as the speed that a marble would pop out of the tube when another was inserted.
Imagine that instead of pushing a whole marble in you merely wiggled the last one rapidly.
The drift speed of the electrons is actually not that important at all. If we communicated over the atlantic ocean by gently pushing on very long sticks, the atoms in the stick might not move faster than a few mm/s, but the signal could be transmitted at the speed of sound for the material composing the stick.
So your real question is "what is the speed of sound for electrons in a conductor?", i.e., how fast do disturbances travel? Wikipedia calls this the "speed of electricity", and it's of order half the speed of light:
This is a good example of asking the wrong question.
How fast does an electron move? 2200 km/s.
How fast does an electrical signal propagate from one end of a circuit to the other? This is a much more relevant, but also more difficult, question. The unit used to measure this performance is called Velocity Factor [1]. Unfortunately, I don't know what the VF would be within a modern microprocessor, but in a wire, it varies from 50% to 99% of the speed of light in a vacuum.
As is typical for mass media news, this story has been summarized in a way that glosses over a lot of the detail.
Photonic computers have a very, very long way to go. Even this tiny beam beam splitter is thousands of times larger than current transistors. Intel uses a 14 nanometer process for Broadwell. The University of Utah beam splitter is 2400 nanometers square. Let's not ignore the fact that a beam splitter and a transistor aren't even close cousins. A beam splitter can't perform any logic.
All of this technology is still very much in the realm of basic science. A lot of the assumptions about the speed improvements are based on napkin math involving theoretical numbers. If anyone ought to know that the difference between theory and reality is often muddied by the trip through the physical world, it would be Photonics researchers. Fiber optic (probably the most widespread application of Photonics research) performance is significantly impacted by real-world factors. Far more so than electromagnetic mediums.
> Photonic computers have a very, very long way to go. Even this tiny beam beam splitter is thousands of times larger than current transistors. Intel uses a 14 nanometer process for Broadwell.
The POC is a major milestone. Miniaturization will come. Don't forget that at one time, we had to use flatbed trucks to move around 5MB of RAM.
Absolutely. A 2.4 micron beam splitter is a hell of an achievement. I don't know what the next closest is in size, but the vast majority of beam splitters I'm familiar with are measured in millimeters. I'm sure there are others doing work on miniaturization, but 2.4 micron would seem like genuine landmark progress.
What we need next is a practical photonic transistor at these scales!
The main benefit optical interconnects aim to provide is not about the speed with which photons propagate, which as you point out is not all that much faster than electrons in copper, but it is about bandwidth.
Propagation speed of course affects latency, which as you point out is already running up against relativistic limits, but the physics of photonics vs. electronics affects the maximum bandwidth achievable in a modulated digital signal. At a high level, one issue is that the capacitance and inductance present in copper limit the ability of discrete pulses to remain discrete voltage levels that still are distinguishable between logical high and logical low when modulated at a very high frequency due to the capacitance of the wire which effectively blurs the pulses together as they propagate. The other fundamental issue with electronic interconnects is the issue of shot noise, which for information theoretic reasons can further limit the theoretical maximum bandwidth logical high/low voltages in copper. Photonics does not have these limitations, but as you say, is still a very long way off from practical application. It is important to point out, though, that the motivation (unlike the article's slightly clickbaity title) is not speed but bandwidth.
It's worth clarifying that I do believe photonic computers are incredibly exciting. I was just a little annoyed at how the news report appeared to focus on the wrong attributes. I have little doubt that the researcher interviewed spoke at length about the subject, but there's an incredible challenge involved in boiling this kind of information down to a 60 second news spot.
I guess that's what it takes to keep people excited about science though. Not everyone is going to be interested in the nitty gritty.
What I don't understand about light computing is how to you use photons to do work? EM radiation is great at carrying information, but it seems that you need electricity to actually do the work of physically pushing gates and transistors around. How do you actually replace the work load with light?
what I'm thinking is that you don't really need the light to do work, as such. If you can build a microscopic, say, XOR gate based on light, which accepts two input light streams, and outputs a single stream that is the XOR of the inputs, you can start building a CPU using light. Perhaps there is some material that, when shined with the light from a single source, it emits light, but when shined with twice the light, it blocks the light. That would be an XOR gate, as far as I can see, and it would not be running on a clock -- the two input streams combine inside the gate to produce the output.
More interesting would be a computer that moves at relativistic speeds or is placed in a deep gravity well. At light speed or in a black hole the computer would seem to make any computation instantly.
If you shoot a computer into a black hole, its signals will come to you slower and slower, because they will take more and more time to climb out of the gravity well. Or am I missing something?
Damn, you're right. So, I should leave my computer on Earth and make a long trip at relativistic speed. When I come back, my computation would be much further than if I had waited on Earth.
Thanks, that's interesting! It seems to imply that I would need to jump into a black hole to collect my solution, which is a major drawback (I was planning this for a Bitcoin miner).
What a terrible article, and it's on a University' official site. The headline is highly misleading. Making a smaller beam-splitter has almost no effect on making computers faster, let alone "at the speed of light".
This would be news-worthy and revolutionary if they somehow came up with a photonic equivalent of a transistor. Something that doesn't exist yet, at least in the traditional room-temperature equivalent, that's easy to integrate on a chip.
All these passive photonic components, such as modulators, beam-splitters, wave-guides and others are all just there to interface to the real work-horses: the electronic transistors. And the best of these we have currently have a maximum oscillation frequency of 1-2 THz (InP HEMTs). Anyway, saying you can compute at the 'speed of light' is a meaningless statement. We measure compute performance in cycles per second.
46 comments
[ 2.9 ms ] story [ 91.7 ms ] threadExciting times indeed.
[1] http://education.jlab.org/qa/electron_01.html
I think I'm confused about waves because of early childhood text books that had person A's mouth, then some lines increasing in size (like the lines on WIFI symbols) and then person B's ear. I now think of waves as moving forward, pushing forward, rather than up and down.
Is there a version of "Maths for mums and dads" that covers science?
http://www.amazon.co.uk/Maths-Mums-Dads-Mike-Askew/dp/022408...
There are both kinds of waves.
Sound waves, propagating from mouths and into ears, and waves in a spring along the direction of the spring, are longitudinal: http://en.wikipedia.org/wiki/Longitudinal_wave
Electromagnetic waves (in an appropriate physical model) and waves on the ocean are transverse: http://en.wikipedia.org/wiki/Transverse_wave
But your intuition about propagation vs drift velocity is correct.
Imagine that instead of pushing a whole marble in you merely wiggled the last one rapidly.
http://physics.stackexchange.com/a/4730/81783
So your real question is "what is the speed of sound for electrons in a conductor?", i.e., how fast do disturbances travel? Wikipedia calls this the "speed of electricity", and it's of order half the speed of light:
https://en.wikipedia.org/wiki/Speed_of_electricity
So the propagation speed of signals for electrons versus photons is not a very good explanation for the potential advantages of photonic computers.
How fast does an electron move? 2200 km/s.
How fast does an electrical signal propagate from one end of a circuit to the other? This is a much more relevant, but also more difficult, question. The unit used to measure this performance is called Velocity Factor [1]. Unfortunately, I don't know what the VF would be within a modern microprocessor, but in a wire, it varies from 50% to 99% of the speed of light in a vacuum.
As is typical for mass media news, this story has been summarized in a way that glosses over a lot of the detail.
Photonic computers have a very, very long way to go. Even this tiny beam beam splitter is thousands of times larger than current transistors. Intel uses a 14 nanometer process for Broadwell. The University of Utah beam splitter is 2400 nanometers square. Let's not ignore the fact that a beam splitter and a transistor aren't even close cousins. A beam splitter can't perform any logic.
All of this technology is still very much in the realm of basic science. A lot of the assumptions about the speed improvements are based on napkin math involving theoretical numbers. If anyone ought to know that the difference between theory and reality is often muddied by the trip through the physical world, it would be Photonics researchers. Fiber optic (probably the most widespread application of Photonics research) performance is significantly impacted by real-world factors. Far more so than electromagnetic mediums.
[1]: http://en.wikipedia.org/wiki/Velocity_factor#Typical_velocit...
The POC is a major milestone. Miniaturization will come. Don't forget that at one time, we had to use flatbed trucks to move around 5MB of RAM.
What we need next is a practical photonic transistor at these scales!
Propagation speed of course affects latency, which as you point out is already running up against relativistic limits, but the physics of photonics vs. electronics affects the maximum bandwidth achievable in a modulated digital signal. At a high level, one issue is that the capacitance and inductance present in copper limit the ability of discrete pulses to remain discrete voltage levels that still are distinguishable between logical high and logical low when modulated at a very high frequency due to the capacitance of the wire which effectively blurs the pulses together as they propagate. The other fundamental issue with electronic interconnects is the issue of shot noise, which for information theoretic reasons can further limit the theoretical maximum bandwidth logical high/low voltages in copper. Photonics does not have these limitations, but as you say, is still a very long way off from practical application. It is important to point out, though, that the motivation (unlike the article's slightly clickbaity title) is not speed but bandwidth.
I guess that's what it takes to keep people excited about science though. Not everyone is going to be interested in the nitty gritty.
https://xkcd.com/678/
Is that actually feasible?
Also I'm not sure if I'm interpreting your question wrong, but nothing actually physically moves in a CPU. "Gate" is just a metaphor.
Too bad the observer can't share the result with anyone, and a couple of moments later s|he expires.
This would be news-worthy and revolutionary if they somehow came up with a photonic equivalent of a transistor. Something that doesn't exist yet, at least in the traditional room-temperature equivalent, that's easy to integrate on a chip.
All these passive photonic components, such as modulators, beam-splitters, wave-guides and others are all just there to interface to the real work-horses: the electronic transistors. And the best of these we have currently have a maximum oscillation frequency of 1-2 THz (InP HEMTs). Anyway, saying you can compute at the 'speed of light' is a meaningless statement. We measure compute performance in cycles per second.