Technology

Peer-reviewed physics, productized.

Every capability on this page exists as a published experiment before it becomes a product. This is the science inside the box.

The platform

Silicon, deliberately.

Universal Clements meshes on 220 nm silicon-on-insulator — the most mature, most manufacturable photonics platform there is. Thermo-optic phase shifters, deep-trench thermal isolation, telecom C-band.

Silicon's density and foundry ecosystem make a 12-mode — and soon a 24-mode — universal processor affordable. The honest trade-off is higher propagation loss than specialty platforms; for classical processing, dissipative simulation, and postselected few-photon quantum optics, that trade is decisively worth making — and we will tell you plainly when it is not.

Full micrograph of the Noor-Q 12 die: 16 mm by 3 mm of silicon carrying 66 Mach-Zehnder interferometers with thermo-optic phase shifters and deep-trench isolation — The Noor‑Q 12 die — 16 mm × 3 mm of silicon: 66 Mach–Zehnder interferometers, 264 elements, deep-trench thermal isolation.

The Noor Twin

A model of your chip, not of a chip.

No fabricated mesh matches its ideal diagram, and sequential calibration fails outright when a single element does. The Noor Twin is a physics-informed model of the whole mesh, trained on your specific device; programming becomes a global optimization through it, redistributing control across every surviving element.

Published results, on an 8×8 processor:

Mean fidelity > 0.97 to 100 Haar-random targets with any 5 phase shifters permanently disabled
Graceful degradation from 0.99 to 0.81 with up to 28 elements disabled — no cliff, no bricked device
A postselected dual-rail CNOT programmed on damaged hardware, addressed entirely in software

The compile loop: the chip is measured once into its twin; every target transformation thereafter is solved globally and dispatched as phase settings — around faults, not through them.

Stylized programmable mesh with highlighted photon paths through state-preparation and evolution regions

Non-unitary programming

Loss as an instruction set.

Most platforms treat dissipation as the enemy. We program with it: exact unitary dilation embeds any lossy or non-Hermitian evolution into a larger unitary the mesh implements natively. This is how our processors realized dissipative topological invariants, exceptional-point physics, and coherent absorption of quantum light.

Four-phase interferometric readout reconstructs complex amplitudes — magnitude and phase, not mere probabilities. The platform measures what other processors can only infer.

The papers behind this →

The control stack

From target matrix to settled phases in seconds.

MZIC Studio decomposes the target, the Twin corrects it for your chip's measured reality, Koi dispatches 64 verified currents. Settling in milliseconds, a full compiled configuration in seconds — thousands of programmed matrices per campaign is demonstrated practice, not a roadmap item.

The full control stack around a 12-mode processor: laser, polarization control, optical switches, current drivers, TEC, host software

Beyond the mesh

A deeper bench than any spec sheet.

The founding team's published research spans the full quantum photonics stack — a pipeline of capability behind the current product line.

Quantum light sources

Deterministic integration of quantum emitters; nanowire quantum dots with telecom-band single and entangled photons.

Single-photon detection

Waveguide-integrated superconducting nanowire detectors with few-picosecond timing — the receiving end of a future integrated system.

Hybrid integration

Silicon nitride, thin-film lithium niobate, and III–V materials combined on-chip — the toolbox for next-generation Noor processors.

Read it before you buy it.

Every claim on this page traces to a peer-reviewed publication or a public preprint from the founding team.

Browse the publications