Akhetonics and against-the-grain thinking

Munich in early June was warmer than I had packed for. The address, when I finally stood in front of it, was an ordinary building with an ordinary stairwell, no glass atrium, no sculpture in the lobby, a buzzer I had to press twice. I had expected the company trying to rebuild the processor to announce itself a little more.

Akhetonics wants to build a computer that runs on light. Light as a courier between chips is already ordinary, the stuff of every data center. What this company means is further in: light doing the computing itself, the logic, photons standing in for the electrons that every processor since the 1950s has pushed around a circuit. It started in 2021 and works out of Munich and Berlin. People reach for the word optical to describe it and then, more often than not, get it slightly wrong. I got it wrong.

What I had assumed, walking in, was the fashionable version of the idea: optics bolted onto a normal chip to speed up the matrix multiplications that AI models chew through, an analog shortcut for one narrow kind of math. One of the engineers set me straight before I had finished the coffee a machine in the hallway had dispensed for me. That is not what they are after. The claim is that a processor could run entirely in light, from the logic gates up, and never borrow an electron to make a decision. General purpose, digital, the sort of machine that could in principle run ordinary code.

The claim is that a processor could run entirely in light, from the logic gates up, and never borrow an electron to make a decision.

The two people behind it come at the problem from opposite ends. Michael Kissner, the chief executive, trained in physics and mathematical engineering, picked up a doctorate in artificial intelligence somewhere along the way, and did a stretch in cybersecurity before this, which is not the background you would draw up for a man rebuilding the processor. Leonardo del Bino, the chief technology officer, is an Italian physicist who works in nonlinear photonics, the finicky craft of coaxing light into behaviors it does not naturally offer. Kissner described his method to me as running the logic backwards. You decide what an optical processor has to be able to do, and only then go hunting for the physical effects that might do it, instead of building up from whatever the physics happens to make cheap. He went looking for someone whose optical transistor designs fit that wish list, and found del Bino.

※

The hard part, and everyone here tells a version of it, is that light is antisocial. Two beams crossing in empty space pass straight through each other and carry on as though nothing happened, which is wonderful for fiber and useless if you need one signal to change another. Electrons are the opposite. They shove. A transistor works because electrons repel, and that shoving, for all the heat and resistance it wastes, is easy to turn into a yes or a no. To compute with light you have to manufacture the interaction that light will not give you for free.

This is where most people stop, and roughly where Akhetonics started.

In early 2024 they showed a general-purpose all-optical two-bit CPU. A demonstrator, laughably small next to anything in your phone, and also a real processor running real logic without dropping back into electrons partway through. I was walked past a bench that, to my untrained eye, looked like very tidy plumbing for light, fiber and mounts and boxes with the lids left on. I understood maybe a third of it. I said as much, and the person showing me around seemed relieved rather than irritated, as though an honest question were less work to handle than a fake nod.

One detail stuck with me more than the optics did. The software used to design electronic chips is a decades-deep industry, and almost none of it applies here, so they have been building their own design tools from scratch, the photonic equivalent of the programs that lay out transistors. You cannot buy the thing that lets you draw the thing. That struck me as the quiet center of the whole effort, the part that does not photograph well and that no funding announcement mentions.

They talk about manufacturing in Europe, at a 130-nanometer scale that sounds coarse next to the newest nodes fighting over a few more nanometers, and they argue that optical chips can be made with far less ultrapure water and fewer steps than electronic ones. There is money behind this now, a round of a few million euros in late 2024 and backing from a German agency that funds long-shot science on purpose. I am not here to tell you whether it pays off, and Reverie does not do that anyway. What held me was the willingness to be wrong slowly, in the open, on a question most of the field treats as closed.

※

On the way out someone mentioned, almost as an aside, that the name comes from Akhet, the Egyptian hieroglyph for the sun rising between two hills, the horizon. I nodded as if I had known. I had not. I spent the tram ride back trying and failing to picture a logic gate built out of light. I still cannot. The people who can work in an ordinary building on an unremarkable street, and they buzzed me out through the same door, two tries.