yurion.io/lab
EN
Write to the lab
YURION LAB · active research

We ship for clients. We build what's next for ourselves.

YURION LAB is the studio's research division: part of every month goes into technology that still looks like science fiction. First on the bench — a processor that computes with light.

project 01 · PHOTON-1 · optical matrix engine● active

A processor that computes with light

PHOTON-1 is our own research prototype of a photonic processor — an optical engine that performs the heaviest operation in modern AI, matrix multiplication, with photons instead of electrons.

laser
lens f₁
mask W
lens f₂
sensor
y = W·x — one pass of light, one matrix–vector product

Light is a natural medium for linear algebra: a lens physically performs a Fourier transform, interference adds numbers, a beam passing through a mask multiplies them. What a GPU grinds out in millions of transistor switches, optics does in a single pass — at the speed of light, with almost no heat.

Every serious photonic computer today is a hybrid: light does the linear math; electronics keeps memory, control and nonlinearity. PHOTON-1 follows the same honest architecture — first as a desktop optical engine, then as an integrated silicon-photonics chip.

Why light

Speed

The computation happens while light flies through the system — picoseconds per pass. No clock, no pipeline: propagation is the calculation.

Energy

Passive optics computes almost for free — interference burns no watts. The energy cost sits at the edges, in lasers and sensors, not in the math itself.

Parallelism

Beams cross without disturbing each other, and different wavelengths share one waveguide independently. One optical system carries many computations at once.

Inside the machine

01

Optical core

A free-space matrix engine: a micromirror array encodes data into light, lenses fan the beam out and gather it back, a sensor row reads the result. One pass of light — one matrix–vector product.

02

Fourier optics

A lens performs a two-dimensional Fourier transform physically, so a convolution layer becomes two lenses and a mask. The classic 4f scheme: an optical CNN layer with zero multiply operations.

03

Digital edge

Electronics does what light can't: memory, control flow, nonlinear activations. A hybrid loop — encode, propagate, read out — ordinary code around an extraordinary core.

Roadmap

01

Optical bench & 4f system

in progress

Assemble the free-space bench and get the first live optical Fourier transform — the moment the physics visibly works.

02

Optical matrix engine

up next

Static masks first, then a micromirror array: real matrix–vector products measured in light and checked against the same math in code.

03

A neural network on light

planned

A digit classifier whose matrix products are computed optically — accuracy and energy benchmarked honestly against a digital baseline.

04

Onto silicon

planned

The same architecture as an integrated photonic chip: interferometer meshes designed in open tools and fabricated on a shared multi-project wafer.

Open by design

The first Apple computer wasn't the first computer — it was the first one anyone could build. That seat is still empty in photonic computing, and we intend to take it: schematics, code, measurements and failures, published as we go.

project 02 · ???○ queued
lab next --status

In the queue. The lab doesn't stop at one experiment.

/ 04

Get involved

PHOTON-1 is funded by the studio's own R&D budget, and we're open to conversations — with engineers and researchers, with partners, and with early backers who want a seat at the optical bench.

Write to the lab Or use the studio contact form