Do you mean as opposed to operating in an analog fashion? My guess is because keeping it digital means it is as close a possible to a drop-in replacement for today's electronic digital computers, including all the firmware, OS, and software that has already been written for them.
"The first thing to understand about Akhetonics’ optical computer is that it’s not one thing yet." Great, that is such a good explanation. Thanks ChatGPT I guess.
This has been tried by Cray like 15 years ago and the university of Colorado 11 years ago; there were also other initiatives, all failed.
I'm not excluding that now we have more budget/time/AI to make it work (if it can), I'd just say exercise caution reading this (very AI slop) article (and anything that looks too good to be true)
> Typically, in order to observe nonlinear phenomena, an intensity of the electromagnetic field of light larger than 108 V/m (and thus comparable to the atomic electric field of ~1011 V/m) is required.
I don’t know anything about this field. Can you address this point? How close are we to overcoming this in academia?
10^8 V/m, yeah that is quite a lot, 100V per micron or 1MV/cm. Even if it works it's going to have reliability issues, especially since the structures are not going to be perfect.
It's just as if he had the brilliant idea to do photonics as an outcast even though there is already a 6" wafer semi-production level fablab being built for it in The Netherlands. There is even a version of TinyTapeout for Photonics.
When you say pointer heavy, just how heavy is that pointer?
In traditional brainfuck, a single < suffices to move The Pointer one cell to the left.
For a dialect of brainfuck where The Pointer is 3x as heavy as is traditional, you'd have to apply 3x the effort, i.e. <<<, in order to move one cell to the left.
Cursed bogocomputing idea of the day: a variant of brainfuck where The Pointer is physically simulated, with mass & momentum. < & > apply impulses. The Pointer continues to move across The Array with momentum after you apply an impulse.
References of Akhenaten is making rounds in modern pagan/occult circles as well. The "Law of One" group is really pushing ancient Egyptian 'pyramid power' stuff.
I didn’t read but I’d also like to point out photonics are 1000x larger than leading edge electronics with minimal path to scaling smaller. I don’t see a free lunch
Basically the density of early CPUs (1970s), but 1000x faster than those, or current CPUs? Like a 1GHz 6502, or a 1THz 6502? The former is not really better, but the latter might interesting.
As a total non-expert I understand most serious work on optical computing to not be aimed at a general replacement for CPUs.
The application that comes to mind is packet switching.
When a packet is transmitted from router to router over the Internet it has to be decoded and decisions determined by the header. That necessitates optical-electronic conversion and then back again at each step. This has inherent latency. And a lot of expensive parallel electronic hardware. Optical systems are much faster than electronic so conversion between the domains involves extremely complex optical manipulation already with the time and frequency multiplexing handled optically.
So given that, what if we could read packet headers and make the routing decision in the optical domain? Truly switching at the speed of light (almost).
If you're looking for the limits of this approach: it's about 1000x faster than electric BUT the individual components have to be way larger (and can't be shrinked down)
... which hopefully means a LOT of fabs will be able to produce these (because a 2nm node makes no sense for this)
It also means this will be incredible for radio chips, interconnects, that sort of thing and will do exactly nothing for the first 20 years for cpu's, never mind gpus.
39 comments
[ 154 ms ] story [ 98.3 ms ] threadI'm not excluding that now we have more budget/time/AI to make it work (if it can), I'd just say exercise caution reading this (very AI slop) article (and anything that looks too good to be true)
So also 1000x smaller integration
Sounds incredible, as in: can't believe it.
By the way, isn't Nvidia researching this?
Very real stuff (I have a PhD on this)
> Typically, in order to observe nonlinear phenomena, an intensity of the electromagnetic field of light larger than 108 V/m (and thus comparable to the atomic electric field of ~1011 V/m) is required.
I don’t know anything about this field. Can you address this point? How close are we to overcoming this in academia?
for a laser the intensity in the wavelength / frequency domain is very high, because it is coherent radiation
it just means you can use a (low power) laser to produce nonlinear optical effects.
In traditional brainfuck, a single < suffices to move The Pointer one cell to the left.
For a dialect of brainfuck where The Pointer is 3x as heavy as is traditional, you'd have to apply 3x the effort, i.e. <<<, in order to move one cell to the left.
Cursed bogocomputing idea of the day: a variant of brainfuck where The Pointer is physically simulated, with mass & momentum. < & > apply impulses. The Pointer continues to move across The Array with momentum after you apply an impulse.
The application that comes to mind is packet switching.
When a packet is transmitted from router to router over the Internet it has to be decoded and decisions determined by the header. That necessitates optical-electronic conversion and then back again at each step. This has inherent latency. And a lot of expensive parallel electronic hardware. Optical systems are much faster than electronic so conversion between the domains involves extremely complex optical manipulation already with the time and frequency multiplexing handled optically.
So given that, what if we could read packet headers and make the routing decision in the optical domain? Truly switching at the speed of light (almost).
... which hopefully means a LOT of fabs will be able to produce these (because a 2nm node makes no sense for this)
It also means this will be incredible for radio chips, interconnects, that sort of thing and will do exactly nothing for the first 20 years for cpu's, never mind gpus.
https://www.decisionproblem.com/paperclips/