[HN Gopher] Getting an all-optical AI to handle non-linear math
___________________________________________________________________
Getting an all-optical AI to handle non-linear math
Author : LorenDB
Score : 85 points
Date : 2025-01-12 13:33 UTC (3 days ago)
(HTM) web link (arstechnica.com)
(TXT) w3m dump (arstechnica.com)
| 4b11b4 wrote:
| Hmm... NNs will be at light speed..?
|
| Are we still in the "glass era"?
| momoschili wrote:
| light speed isn't really all that good when you remember that
| information in your electrical circuit is traveling as fast (if
| not faster) than light in the medium in this case. Sounds good
| in marketing, but the key here is bandwidth, not 'speed'
| 4b11b4 wrote:
| certainly -- not necessarily the raw velocity of light, but
| the ability to augment other properties of light to increase
| the bandwidth.
|
| Is it correct then to say that augmenting the other
| properties of light increases overall information
| density/capacity of light as a medium? whereas with
| electricity we only have 2D: amplitude & freq?
| momoschili wrote:
| definitely
| aoeusnth1 wrote:
| faster?
| momoschili wrote:
| Information in an electric circuit travels along that
| circuit at ~0.9c. The physical light pulses that are
| running through the optical waveguides are more or less
| traveling at around ~c/n, where n is the refractive index
| of the material (in this case silicon, so n ~ 3.5).
|
| the actual optical "packet" of information is traveling
| slower than the electric "packet". The key here is that the
| electric packet can a few bits, while the photonic packet
| in theory has a much larger bandwidth.
| LorenDB wrote:
| Also keep in mind that photonics will generate practically
| zero heat and use less power overall, which will be great
| news for data centers.
| stormfather wrote:
| Layman here. Isn't the big speed penalty in digital
| electronics the clock speed of the transistors changing
| state? Not the actual signal going down the wire from gate to
| gate?
| danpalmer wrote:
| This sounds impressive, but this bit stood out to me:
|
| > This process works by sending a tiny bit of the optical signal
| to a photodiode that measures how much optical power is there.
|
| It seems that the benefit of the approach in general is to keep
| compute in optics, because crossing the optical to electrical
| boundary takes too long. But then in the middle of their
| described process is a boundary transition.
|
| How is this so different to the CMOS/CCD boundary? Is a
| photodiode that much quicker to activate that it doesn't matter?
| jcims wrote:
| I'm sure you'll get a better answer eventually but yes
| photodiodes are widely available that have sub-nanosecond
| response time, and the output could potentially be used in its
| raw analog form do whatever modulation they are describing.
|
| Edit: Turbo encabulator description from the paper linked at
| the bottom:
|
| >To realize a programmable coherent optical activation
| function, we developed a resonant electro-optical nonlinearity
| (Fig. 1(iii)). This device directs a fraction of the incident
| optical power |b|2 into a photodiode by programming the phase
| shift th in an MZI. The photodiode is electrically connected to
| a p-n-doped resonant microring modulator, and the resultant
| photocurrent (or photovoltage) detunes the resonance by either
| injecting (or deplet-ing) carriers from the waveguide.
|
| ... and a couple of notes on the observed latency later in the
| paper
|
| >We experimentally characterized the computational latency of
| the NOFU in this mode, finding that the response time for
| carrier injection was shorter than 100 ps and that 75 mA of
| photocurrent was sufficient to detune the resonator by a
| linewidth, corresponding to a static power dissipation of 60
| mW.
|
| >As our architecture computes entirely in the optical domain
| and is integrated onto a single photonic circuit, inference
| latency is limited only by the optical time of flight through
| the chip
| momoschili wrote:
| It's a really smart idea to try to leverage the inherent
| scalability of semiconductor photonics. I think the use of a
| linear optical resonator to amplify a weak optical nonlinearity
| is quite genius, and something the relatively small nonlinear
| photonics community has been trying to do forever. That they
| showed this kind of 'all-optical-ish' nonlinearity on a
| relatively mature process in a foundry is nothing to scoff at,
| and likely one of the biggest results in semiconductor photonics
| in a while. At the single device level I think it makes so much
| sense, but what concerns me in general is how well this scales
| from a few perspectives:
|
| 1. resonators and device-to-device variance: in general it's
| pretty hard to get these resonant effects to line up with each
| other from a production POV, especially with large arrays.
| Silicon photonics has come far, but I don't think it has
| approached the level of uniformity as electronics. They have
| demonstrated some level of electro-optic tunability, which is the
| traditional solution, but they still need to leverage that for
| their nonlinear effects too.
|
| 2. area and space: the 'minimum' trace size of these planar
| photonics circuits is still quite large (~200 nm minimum feature
| size typically for these waveguides). This is essentially due to
| a minimum size needed to confine light within a waveguide which
| depends generally on the waveguide's refractive index and target
| wavelength. These are currently all integrated on a planar
| manner, so each channel becomes quite large, especially if now
| you also need a relatively large ring resonator, which in this
| case is at least ~100 micrometers or so in diameter
|
| 3. the combination of 1 and 2: high device-to-device variation,
| along with a large planar footprint means that these things are
| quite expensive and difficult to manufacture, without some kind
| of miniaturization benefit that you would typically get with
| electronics (at least not yet). This effect appears to be more
| than the sum of 1 + 2.
| singularity2001 wrote:
| The planar footprint can be relatively easy be mitigated by
| just using a mirror and sending the next layer back to the same
| plane a few nanometers higher?
| momoschili wrote:
| It is a solution to increase density, but this introduces
| significant integration complexity in addition to just
| increasing the overall cost. I think if this optical analog
| computer takes off, this kind of routing will likely be
| necessary
| adrian_b wrote:
| The preprint of the research paper:
|
| https://arxiv.org/abs/2208.01623
| looofooo0 wrote:
| Interesting that they don't mention event based sensors at all.
|
| https://www.sony-semicon.com/en/technology/industry/evs.html
___________________________________________________________________
(page generated 2025-01-15 23:02 UTC)