The basics of digital logic are pretty easy to master, and figuring out how the ones and zeroes flow through various kinds of gates is often an interesting exercise. Taking things down a level and breaking the component AND, OR, and NOR gates down to their underlying analog circuits adds some complexity, but the flow of electrons is still pretty understandable. Substitute all that for photons, though, and you’ll enter a strange world indeed.
At least that’s our take on [Jeroen Vleggaar]’s latest project, which is making logic gates from purely optical components. As he himself admits in the video below, this isn’t exactly unexplored territory, but his method, which uses constructive and destructive interference, seems not to have been used before. The basic “circuit” consists of a generator, a pair of diffraction patterns etched into a quartz plate, and an evaluator, which is basically a pinhole in another plate positioned to coincide with the common focal point of the generator patterns. An OR gate is formed when the two generators are hit with in-phase monochromatic light. Making the two inputs out of phase by 180° results in an XOR gate, as destructive interference between the two inputs prevents any light from making it out of the evaluator.
All seven basic gates are possible using variations of this technique, but of course, actually building such things poses challenges. The microscopic manufacturing techniques, including photolithography and etching patterns on thin metal coatings, are pretty non-trivial. And while [Jeroen] was able to construct a few basic gates and test them, there’s still a long way to go. He says that he needs to add both input beam controls — LCD shutters, perhaps — as well as output detection.
The potential applications of this are pretty exciting, and we’ll be staying tuned for details. For deeper background on the dual nature of light and how interference patterns form, check out [Jeroen]’s explanation of the double-slit experiment.
There is one problem with photonic elements, that makes photonics impractical as a replacement of electronics. Electronic elements are more or less square, while photonic elements are extremely rectangular with one dimension (length) orders of magnitude than two other. This makes VLSI photonic elements infeasible.
In a universe without mirrors, that is a problem.
It is a hard problem, but there are plenty of people working on it such as the startup lightmatter https://www.eetimes.com/optical-compute-promises-game-changing-ai-performance/
Very interesting! I wonder about his last remark though about performing complex analog calculus with such a setup. It seems to me that it would be sensitive to the same conditions as (electrical) analog computers; variance in production, temperature, .. and maybe also age. Would be interesting to know how many use cases such a technology would have and where it could beat other solutions.
A few years back HaD featured an article on a mechanical targeting computer.
Perhaps optical processing could take a leaf from that book and scan across three dimensional topologies to do complex mathematical operations and then combine them optically via interference. Perhaps some sort of magneto optic effect with several polarised beams or maybe a single beam of light polarised by a varying amount in regards to the thickness of a medium or saturation of a magnetic field at a given distance from a magnet.
I don’t know the article in question, but you can do some pretty clever stuff with interference effects, e.g. a 4f correlator. Basically a lens performs a Fourier transform of an object, so place the object next to a reference object, and if the object and it’s reference match, there will be a bright spot in the Fourier plane.
Optical lock? Wonder how resistant it would be to Ramset.
I’ve been following this guy’s work for some time. This is pretty neat, if unintuitive (light is linear, XOR is not). In order to make complicated circuits out of this you need to make restorative logic – the output of that XOR gate is going to be too weak to drive another gate, but the same optical amplification used in e.g. fiber optics could be employed here.
The nonlinear part is in the detection as it acts on the modulus square (intensity) of the wave and requires to select a region in space (the photon will always end up somewhere). This selection can be made perfect in principle so that a 1 in the truth table really means 100% probability to detect an photon and 0 means 0%. The fact that photons can be seen as particles makes this also intrinsically discrete unlike an analog computer as long as you can work with single photon detectors. The more energetic the photon, the easier this is.
One downside for visible light photons is their wavelength which would force this device to stay in the 100’s nm range at least, being far more bulky that current semiconductor technology. Plasmonic tricks could overcome this. Replacing the gratings by spatial light modulators would allow for a fully programmable gate that can be reconfigured.
Does it decohere if you attach a debugger?
I jest, but the similarity to the double slit experiment is interesting.
I’m sure I saw this in the 1980’s and someone said they were patenting it. It didn’t get any traction and people thought it was pretty ho-hum. I do recall the photos and diagrams and it was very much like what is seen here. Same interference concepts and same logic and truth tables. But I can’t find anything about it from that far back.
At the time there were several magazines like Electro-Optical Systems Design and Laser Focus but I think I saw it in something much more obscure, like Forth Dimensions or Bergenholm Engineering. The earliest I see on the Interwebs is https://uncw.edu/phy/documents/raphael_06.pdf. So maybe not. This was also the time of the great but brief hype in Silicon Valley about “The Laws of Form” revolutionizing computer hardware.
I follow ‘Huygens Optics’ and always learn new things. I’m pretty sure the ideas I saw in the past was all proposed with slits and straight diffraction elements. Huygens Optics use of Fresnel Zone Plates to focus the energy to points is very cool.
Working with optics in a binary fashion is interesting, but is that really the “best” that optics have to offer?
Considering that just like in RF, we can produce fairly clean tones, that we can filter, amplify and also mix with each other.
Where we can gate unwanted tones by simply mixing them out of band as to filter them away.
Computing can be done in similar fashion, where we can shift a tone to a new frequency. In this new frequency it can either be filtered away or used to shift another tone. Though, we will likely have filters after every mixer, since unwanted byproducts will exist in plenty amounts…
Downside of using the frequency domain for computing is that we suddenly have an analog computer, where manufacturing tolerance, drift and accuracy of both filters and tone generators will introduce important offsets to keep in mind. Unless we sufficiently quantize our values for this accuracy and drift to not matter. Though, our filters will have to be fairly sharp as well.
But I see it as an interesting field of computing that seems to garner little attention in general.
Though, to return to the largely binary logic gate in the article above, it is still interesting.
Read up on Mach–Zehnder devices on Lithium Niobate.
You might want to read up on waveguides build out of photonic crystals. There are a lot of weird quantum effects involved but it essential boils down to interference but just with a single wave.
Why bother with an LCD & CCD, diodes in diodes out.
Even if optical is limited in terms of circuit scale, I wonder if it could be produced in a way comparable to stamped holograms, which could let them just be so cheap to make as to have them be worth using for certain embedded system purposes. Not only cheap on a per-device basis, but also but also maybe in terms of baseline cost compared to a custom silicon-based AISC, which could help organizations that want to make secure systems produce the logic components for more sensitive devices in-house.