Here’s why we don’t have light-based computing just yet

 

 In principle, communicating with light is much, much easier than communicating with electricity. We’ve been doing it for much longer, in technologies ranging from signal fires to fiber-optic networks, because photons have the capacity to move data far more quickly than electrons. Yet light also has many frustrating problems that electrons don’t — problems that have kept light from displacing electricity on the nanometer scales of modern computing. For a long time, the major impediment to a photonic revolution in computing, and an exponential increase in computer speed, has been a sort of zero sum game between three major players: size, power, and heat.

The thing about light is that by atomic standards it’s really very large. In general, the smallest useful wavelength of light for computing has been in the infrared range, around 1000 nm in size, while improvements in silicon transistors have seen them reach and even pass the 10 nm threshold. Lithography has come up with incredibly clever and complex ways of diffracting light to etch silicon wafers with details smaller than the wavelength of the light doing the etching — pretty incredible stuff — but that’s child’s play compared to the sorts of super-fast, super-complex communication that we would require inside a modern computer processor. Existing techniques in bending light waves just won’t do the job.

Optical fiber, in blue and white.

To get around the size problem and make light useful on the scales we require for next-gen computer performance, engineers have turned to something called “surface plasmons.” These are essentially electrons that have been excited so that they dance along the surface of a material, exploiting quantum weirdness to behave and travel more like a photon than an electron. It’s a bit of a halfway point between electricity and light, using many of light’s behaviors, but staying physically confined to a much, much smaller space right at the surface of the wire. If created on a regular copper wire, these surface plasmons can travel much faster than a normal electron in the same medium, and even closely approach the speed of light.

The speed at which we can communicate over a distance matters more when we have more distance over which to communicate, so the first assumed computing application for photonics is in the relatively long-distance communication between processor cores. Right now, copper wire connects these super-fast components to allow them to work together — but the communication between cores is starting to lag further and further behind the speed of any one of those cores individually. So, if we want to utilize all the potential power of, say, a 64-core processor, we’ll need to keep those cores coordinated with something much faster than electrons moving through copper wire — something as fast as light would be good.

The problem when you switch from light waves to surface plasmons, though, is that plasmons very quickly lose their power — they move real fast, but tend to peter out long before they reach their destination. To get them to maintain enough of their power all the way from source to destination, engineers can “pump” the wire into an active plasmonic component — essentially expend a bit of energy on keeping the wire in a state where the surface plasmons won’t lose a ton of energy as they travel.

But that creates its own problem: heat. Surface plasmons solve the wavelength problem, and active plasmonics solve the surface plasmon power problem, but now we’ve got to keep all these actively pumped components from overheating due to all the excess energy we’re adding. This has been a tough problem to crack, and it’s led to the assumption that any photonic computing system would need to be either cooled with some super-advanced cooling system, or made of some exotic wiring material that’s much better at maintaining surface plasmon signals without significant help.

Both areas of research are well underway, but a recent study from the Moscow Institute of Physics and Technology (MIPT) has shown that with a good enough regimen of existing cooling technologies, actively pumped copper wire could give us both the plasmon-slipperiness and the heat dissipation we need to realistically run a consumer device. That means that as conventional computer architecture gets more complex and adds more processing cores, we may actually see the associated speed increase we’d want and expect.

This complex system of conventional heat-sinks could help solve one of the biggest impediments to optically enhancing computers.

Of course, the idea of photonic computing goes beyond just maintaining coordination between processing cores made of electronic transistors. Not only is it very time- and energy-inefficient to be switching your signals back and forth between photons and electrons, but so-called optical transistors could have much higher bandwidth than electronic ones. It will require a number of additional breakthroughs, but research is underway — like this recent study looking for an affordable material that could do accurate, thin-film polarization of light signals. Graphene and carbon nanotubes have a ton of possible utility for optical computing, since they could transport surface plasmons and make the advantages of photonics work on the nano-scale.

A real optical computer is much further out than a hybrid, which uses optical tech to coordinate conventional electronic cores. Once created though, a fully optical computer could possibly allow us to restart Moore’s Law. It won’t hold a candle to some future, comprehensive quantum computer, but until we get such a thing an optical computer is one of our best bets to restart exponential growth in computing power.