Corkscrew optics yield direct line to electronics(spectrum.ieee.org)
spectrum.ieee.org
Corkscrew optics yield direct line to electronics
https://spectrum.ieee.org/optical-computing-picosecond-gates
24 comments
Actual paper: https://www.science.org/doi/10.1126/sciadv.abq8246
- "The new gates performed at speeds of less than 100 femtoseconds, which is roughly 1 million times as fast as electronic gates"
Electronic gates switching at 10 MHz?
Electronic gates switching at 10 MHz?
Yeah, they're off by a factor of 1000.
There are other serious technical errors in the piece too. For example they say "because photons move at the speed of light and electrons don't" ... which is irrelevant because electronic signals do move at the speed of light.
The most serious problem is missing context. It's been known for decades that you can build shockingly fast optical switches and even amplifiers. A hundred fs is fast but not unique. The problem is how to connect these devices in a complicated circuit without putting nanoseconds of waveguide in between. Unfortunately 1 ns is 1,000,000 fs.
So in telecom, where you're just going a long way in one direction, optical amplifiers with multi-THz bandwidth have dominated transport since the 1990s. But in computers, no.
There are other serious technical errors in the piece too. For example they say "because photons move at the speed of light and electrons don't" ... which is irrelevant because electronic signals do move at the speed of light.
The most serious problem is missing context. It's been known for decades that you can build shockingly fast optical switches and even amplifiers. A hundred fs is fast but not unique. The problem is how to connect these devices in a complicated circuit without putting nanoseconds of waveguide in between. Unfortunately 1 ns is 1,000,000 fs.
So in telecom, where you're just going a long way in one direction, optical amplifiers with multi-THz bandwidth have dominated transport since the 1990s. But in computers, no.
Electronic signals also do not move at the speed of light: https://en.wikipedia.org/wiki/Velocity_factor
I think you're missing the bigger picture. While they're both EM but they're significantly different i.e. photon are much more desirable than electron.
For starter, for electron you have contention in the form of collision but there's no collision for photon. Imagine you have vehicles that can go pass through each other without any collision, they will be exponentially faster than the one's without this capability.
For starter, for electron you have contention in the form of collision but there's no collision for photon. Imagine you have vehicles that can go pass through each other without any collision, they will be exponentially faster than the one's without this capability.
In both cases it's EM waves that are traveling. In 'electronics,' mobile electrons normally guide the EM waves, which are typically 0-10GHz in frequency. In 'optics,' bound electrons normally guide the EM waves, which are typically hundreds of THz in frequency. In either case, you make a switch by somehow reversibly spoiling the waveguide that connects point A with point B.
Collisions between electrons don't directly affect either picture. You can multiplex lots and lots of completely independent 'electronic' EM waves on one conductor, they pass right through each other exactly as you described for optics, no problem.
The advantage of optics is that the low-loss bandwidth of optical waveguides is enormously wider than the low-loss bandwidth of waveguides made using copper or other conductors. And with optics we have enormous bandwidth available.
One disadvantage of optics is that we don't have a low-lag way to interconnect switches. We can connect switches, but so far the interconnect delays are much, much longer than the switching times. So a fast optical switch is useful, but so far only in systems that are solely "feed forward," like communication.
Collisions between electrons don't directly affect either picture. You can multiplex lots and lots of completely independent 'electronic' EM waves on one conductor, they pass right through each other exactly as you described for optics, no problem.
The advantage of optics is that the low-loss bandwidth of optical waveguides is enormously wider than the low-loss bandwidth of waveguides made using copper or other conductors. And with optics we have enormous bandwidth available.
One disadvantage of optics is that we don't have a low-lag way to interconnect switches. We can connect switches, but so far the interconnect delays are much, much longer than the switching times. So a fast optical switch is useful, but so far only in systems that are solely "feed forward," like communication.
For transport, yes light is more desirable. For computing, no. The fact that electrons interact with each other is the basis of computational circuits. The article is about how to get photons to do the same.
Pardon me, but I don't think you get the big picture either. Chip industry is now seriously considering photonics based on-chip interconnect for better speed and efficiency [1][2]. However having this hybrid system of electronics and photonics required conversion overhead that's being lamented in the posted article. Ideally, having native photonics based computing and communication systems is going to be a game changer for better speed and efficiency since no conversion is necessary.
[1]Two Startups Are Bringing Fiber to the Processor:
https://spectrum.ieee.org/optical-interconnects
[2]Power and Latency Analysis of the Memory Tiering Pyramid: The CXL Effects:
https://news.ycombinator.com/item?id=34037321
[1]Two Startups Are Bringing Fiber to the Processor:
https://spectrum.ieee.org/optical-interconnects
[2]Power and Latency Analysis of the Memory Tiering Pyramid: The CXL Effects:
https://news.ycombinator.com/item?id=34037321
Probably just a miscommunication. The actual improvement is more like 10K times, but this moves from the 'gigahertz range' to the 'petahertz range' which differ by 1e6.
> "Since light oscillates so fast (roughly a few hundred million times per second), using light could speed up electronics by a factor of roughly 10 000 as compared to computer chips,” says Tobias Boolakee, a laser physicist in Peter Hommelhoff’s group at the FAU and the first author of a study in Nature on the new gate." (2022)
https://physicsworld.com/a/logic-gate-breaks-speed-record/
> "Since light oscillates so fast (roughly a few hundred million times per second), using light could speed up electronics by a factor of roughly 10 000 as compared to computer chips,” says Tobias Boolakee, a laser physicist in Peter Hommelhoff’s group at the FAU and the first author of a study in Nature on the new gate." (2022)
https://physicsworld.com/a/logic-gate-breaks-speed-record/
Light oscillates much, much faster than a few hundred million times per second. Take 1 micron wavelength. At a speed of 3e8 m/s, 1 micron light oscillates 3e8/1e-6 = 3e14 times per second. Three times ten to the fourteenth power.
For anybody who doesn't juggle large numbers every day, a hundred million is 1e8. Ten to the eighth.
For anybody who doesn't juggle large numbers every day, a hundred million is 1e8. Ten to the eighth.
Maybe they meant trillions? The 10,000 times faster number seems correct though, as fiber optic carries light at ~200-300 THz and 10 GHz is the electronic standard at present.
1 femtosecond = 1e-15 seconds.
Right, so 100 fs = 1e-13 seconds. And a million times slower than that would be 1e-7 seconds, or (10 MHz)^{-1}.
I always found circular polarisation with its "natural" complexity one of the most fascinating aspects of generally available optics. Really happy about this practical use case and application of its properties.
Might sound a bit odd, but sometimes there is some kind of beauty in a certain design or basic approach.
Might sound a bit odd, but sometimes there is some kind of beauty in a certain design or basic approach.
The problem I see is that at these types of speeds, timing skew is going to be a crucial issue. Latency will dominate the design. I can see a design much like early computers with delay lines used as RAM. You could keep a series of packets in a fiber, and then operate on it at the appropriate time.
Perhaps we can use fewer optical gates and serialize the use of gates, using pipelining at gate level.