Simulation-validated · pre-hardware

The H-cat photonic quantum computer, where loss becomes a heralded erasure.

A hybrid CV-DV architecture: each logical qubit pairs a bosonic cat with a single photon. When a photon is lost, the cat parity flips, so loss is caught and located, not a silent unknown error. That is the advantage, made visible below.

One H-cat logical qubit, live

Mode A is a cat (amber); Mode B is a single photon in dual-rail (cyan). Inject a loss and watch.

Why H-cat wins

The same loss that breaks other photonic qubits is the H-cat's error signal.

H-cat (hybrid CV-DV)
  • Usable cats at 3.5 to 8 dB squeezing
  • Deterministic hybrid fusion, conditioned on the herald
  • Photon loss flips parity: loss = heralded erasure
  • Erasures correct far more cheaply than silent errors
GKP encoding
  • Needs 10+ dB squeezing, hard to reach on chip
  • Grid-state preparation is demanding
  • Loss degrades the grid as a silent shift error
Dual-rail / linear optics
  • Probabilistic fusion: most attempts fail
  • Heavy multiplexing to reach near-deterministic gates
  • Loss removes the photon, costing the qubit outright
Validated numbers (June 2026 circuit-level campaign)
0.79 to 0.90%
Circuit-level loss threshold per cycle, full noise
3.5 to 8 dB
Usable squeezing for cats (vs 10+ dB for GKP)
d=53
Code distance at the 0.5% per-cycle loss spec
5,617
Cats per logical qubit at 1e-9 (0.5% loss)
99%
3D fabrication Monte Carlo yield (with tuner authority)
live
pip install dyber · pip install dyberforge

Numbers reflect full circuit-level noise (noisy syndrome extraction), superseding earlier phenomenological figures. The machine is operated at a 0.4 to 0.5 percent per-cycle loss spec to leave margin below the threshold.

The toolchain

From concept to device spec, in the browser and in Python.

Credibility, honestly stated

Pre-hardware, but every claim is simulation-validated.

Cross-checked

Two independent codebases (QuTiP and Strawberry Fields) agree on cat fidelity to 4e-5.

Circuit-level EC

Full circuit-level error correction: noisy syndrome extraction with stim and PyMatching.

Fab tolerance

3D EM plus fabrication Monte Carlo: 99 percent yield with adequate thermal tuner authority.

DRC clean

Test chip QF-TC1: 5x5 mm tile, 22 structures, DRC clean (0 violations) on a foundry-neutral stand-in deck.

Packages live

pip install dyber and pip install dyberforge both work today, Apache-2.0. SDK on GitHub.

See the evidence

Honest status: hardware is in development and the hcat_qpu backend is not yet available. Circuits run on the physics-informed local simulator. Resource numbers are estimates calibrated to the June 2026 campaign, not guarantees. Dyber, Inc.