Optical Computer Prototype | Hacker News Reader

Optical Computer Prototype

optalysys.com

50 pointsfenrissan26 comments

This appears to basically be an analog optical computer - it inputs some 2D data using an LCD, puts it through a bunch of optical transforms, and then captures the output with a camera. This probably does have some kind of obscure use cases where it makes sense, but it's very hard to compete with modern silicon for raw compute using some kind of hybrid method - moving data in and out of the optical part is just painful compared to keeping it all digital/electrical. The main problem is that computers today aren't actually slow =)

Today's computers are very slow. Try modeling something complex, let's say 1 billion water molecules interacting. Then realize that 1 billion molecules is orders of orders of magnitude far from modeling a cup of water.

> 1 billion molecules is orders of orders of magnitude far

Understatement. There are 50 trillion atoms in a cell, 50 trillion cells in a human body (give or take an order of magnitude or two for definitions and caveats). Furthermore, big swaths of chemistry/biochemistry are inherently quantum mechanical (classical mech + E&M doesn't explain why molecules snap into little geometric shapes, let alone how those shapes interact) which has god-awful asymptotic complexity on account of the "present state" of the system (wavefunction) being a probability for each possible configuration of the system rather than a description of a single configuration.

A purpose-built silicon supercomputer will struggle to simulate a single small protein using classical-mechanics approximations for a millisecond (and there are millions of those per cell and trillions of cells per body). There's a lot of room for improvement.

Right, but they didn't build an optical computer that is even close to competitive with something like this:http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/...

I think the submission here on HN has overdone it on the title.

It doesn't compute "flops" like a traditional computer. The relevant text from the article:

"The prototype achieves a processing speed equivalent to 320 Gflops and it is incredibly energy efficient as it uses low-powered, cost effective components."

I am as interested as anybody in switching out for photons instead of electrons/holes. But please use the original title, unless it is misleading or linkbait.

Ok, we took "32O Gflop" out of the title. If anyone suggests a better title we can change it again.

If photons and electrons travel at the same speed am I right in thinking that the benefits of optical computing would be limited to parallel processing?

>If photons and electrons travel at the same speed

"If" - they do not! https://en.wikipedia.org/wiki/Electron_mobility

Quoting from the above source...

>Typical electron mobility for Si at room temperature (300 K) is 1400 cm2/ (V·s) and the hole mobility is around 450 cm2/ (V·s).[2]

However this is besides the point since electric fields in a conductor do move at the speed of light.

The article refers to lower power requirements. One of the problems of higher clock rates is heat dissipation; if this runs a lot cooler, then it might be able to clock much higher.

Electrons have mass they can't move at the speed of light. Photons also have other nice properties such as wavelength which open a whole suit of possibilities e.g. like having a logic gate which can operate in different mods based on the wavelength and polarization of the light.

While this is technically possible with electronics as well by setting a different voltage limit it's much more effective with photonic computing and doesn't not increase the complexity of your base components as much.

The current designs for a photonic computer are also much more parallel most of them basically layers of LED's and detectors with a very fast LCD matrix which serves as a mask between them. if you have a 256x256 pixel screen you can perform an operation on 65536 bits in a single clock, if you stack them up you basically getting 1 order of magnitude with each layer this isn't something you could ever achieve with current solid state electronics.

What fraction of the speed of light are electrons moving at in the most advanced silicon chip we have? Trying to understand how much speed is left on the table for us to pick up in serial processing?

It's not that easy to define because were talking about semi-conductors after all;)

They don't move at a constant speed like say in a conductor (which isn't the case either because electric field causes resistance e.g. eddy currents, but lets say a super-conductor)

You have quite a few concepts with semiconductors primarily saturation velocity (which in most cases is the peak velocity, but not necessarily attained in actual operation) which is also affected by the drift velocity due to any electric fields in your components.

https://en.wikipedia.org/wiki/Saturation_velocity

Light speed in vacuum is 29979245800 cm/s Saturation velocity in Si based SC is 10000000 cm/s

So there's quite a bit of difference there ;)

So does this imply an opportunity to improve computing speed by 2997?

No because the speed of the components with the current designs is still limited by the switching speed of the masking matrix. And in any case you will still need to transform photons back to electrons twice. The speed of electrons or EM wave propagation vs photons isn't a player in why photonic computing might be better. The ability to encode more data into the photons, a much more power efficient and inherently parallel design are.

Also because light has much more shorter wavelength than "electricity" electromagnetic waves in conductors (kilometers vs nano-meters) you can have a much higher throughput with light even if you use it to replace the carrier wires in electronics (which is what Intel is working on they want to keep electronic switching but make all carriers photonic).

https://en.wikipedia.org/wiki/Speed_of_electricity

Light moves 3*10^8 m/s in a vaccuum, which is 7.5cm per clock cycle of a 4GHz processor. The processor is about 2cm wide, so not much room for improvement I figure.

Information we move around using electrons moves much faster than any single electron - because electrostatic field changes move at speed of light obviously, so the electrons at the end of wire start moving as soon as the electrostatic field change gets to it.

You don't have to wait with processing for the electron from the beggining of the wire to get to the end?

So, the difference between photons and electrons speed as measured by the time it takes electronic and optical signal to move through the same distance - is insignificant.

I didn't say that we need to move electrons (although we do for some things like Flash Memory), but you are confusing simple wave propagation in conductors with electron mobility in semiconductors which is quite a bit more complicated.

You also need to remember that when we talking about waves then the wavelength ties directly to your data throughput, and with electrons the wavelength in conductors at say a frequency of 1GHZ is still around 3 kilometers, visible light has much much shorter wavelengths which allows you to pass more data per given amount of time.

Yeah I was thinking about conductors, didn't realized it's different in semiconductor.

Regarding the wavelength - it's another of these things I never quite understood about physics. I've been told that every particle can be thought of as a wave with a frequency, and that it corresponds to the energy of the particle. Photons have lower energies than electrons or protons, so the wavelength of electrons should be shorter, not longer?

You've mixed and matched some of the stuff, leave wave-matter duality aside. You can treat it as any other wave in that regards best analogy will be acoustic waves because they are very simple. http://farside.ph.utexas.edu/teaching/em/lectures/node102.ht...

Might be worth reading up to understand how much the speed Electrons can move at matters: https://en.wikipedia.org/wiki/Speed_of_electricity

TL;DR: "The speed at which energy or signals travel down a cable is actually the speed of the electromagnetic wave, not the movement of electrons. Electromagnetic wave propagation is fast and depends on the dielectric constant of the material. In a vacuum the wave travels at the speed of light and almost that fast in air."

While that's true for simple conductive cables that's not the case for semi-conductors. https://en.wikipedia.org/wiki/Saturation_velocity

Especially since there are some semi-conductors which actually require you to move electrons, like e.g. flash memory ;)

The better we understand EM radiation effects on all kinds of life the more important this technology will become.

Isn't 320 GFLOPs kinda slow? An NVIDIA GeForce GTX Titan Z can crank out 8122 SP GFLOPs for comparison.

That's the proof of concept prototype. If they achieve their stated targets, it would be a leap in computing capability.

http://www.hpcwire.com/2014/08/06/exascale-breakthrough-weve...

"The analysis unit works in tandem with a traditional supercomputer. Initial models will start at 1.32 petaflops and will ramp up to 300 petaflops by 2020.

The Optalysys Optical Solver Supercomputer will initially offer 9 petaflops of compute power, increasing to 17.1 exaflops by 2020."

Apparently hype has the right of way on this post.

Oh no, not this stuff again.

Reply on news.ycombinator.com