Chips that steer light under software control.

Our processors work with light instead of electricity. Here is how they work, without the physics degree.

How it works

A photonic processor is a network of tiny light paths on a chip. Where paths meet, atunable interferometer decides how much light goes each way.

Set all of them from software and the chip performs a chosen operation on the light passing through. Change the settings and it performs a different one, on the same chip, with no new manufacturing.

Light enters, is mixed by a grid of tunable interferometers, and leaves as the result, at room temperature. An illustration, not our chip layout.

Why it matters

Room temperature

The processor chip needs no cryostat. Single-photon detectors, where an experiment uses them, still need a compact cooler, and we say so openly.

Programmable

One chip, many experiments, reconfigured in software.

Made with chip-industry methods

The chip uses wafer-scale lithography on a low-loss material that is well established for photonics.

Our idea: a memory layer for light

Today’s programmable photonic chips hold each setting by continuously heating a small part of the chip. That costs power, produces heat, and makes the chip drift as its temperature changes.

Q-Memory’s patent-pending memory layer is designed to hold settings at near-zero power, so the chip keeps its configuration without constant heating. The aim is processors that are more stable and lower-power, which matters when many chips share a rack.

Status: this is our patent-pending concept. It is not part of the first test chip. The first chip proves the processor platform the memory layer will sit on.

Designed simulation-first

Every component is simulated before fabrication, including under worst-case manufacturing conditions, not only ideal ones. Three things we learned:

A tiny length change can ruin a light splitter

A splitter must divide light 50/50. Our simulations showed that the splitter length carried over from an earlier design would have sent about 93% (simulated) of the light one way instead of half. We caught it on the computer, not in the cleanroom.

The chip now also carries a small grid of splitter variants, so the first run pins down the exact value for every later chip.

A splitter’s behaviour swings strongly with its length, so we simulate before we build. Schematic, without design values.

Designing for the worst case, not the best case

Real manufacturing never hits the target exactly. We simulate each component at the nominal design and at the weak edge of what the process might deliver, and the chip is dimensioned to work at that weak edge. The result is a small cost in chip area in exchange for a much better chance that the first run works.

Heat is the hidden enemy of photonic chips

Tunable interferometers are set by tiny heaters, and heat spreads. Our design work showed that a stable chip needs three things:

precise temperature control of the whole chip
a measured map of how each heater affects its neighbours, corrected in software
heaters sized so they operate well inside their safe limits

This is exactly the problem our memory layer is designed to remove in later generations.

The limits, stated plainly

No chip has been fabricated yet. The first test chip has completed design review and fabrication is being scheduled.
Detectors need cooling. The processor runs at room temperature, but single-photon detectors need a compact cooler.
The memory layer comes later. It is not part of the first test chip.
Goals are goals. We will publish the measured numbers, whatever they are.

See what we are building

Marks beside the figures on this page:

Simulated: A result from our simulations. Not yet confirmed on hardware.