Cogitan

Fluxus · the surrogate design suite

One engine.
Two products.

Fluxus is the surrogate engine for superconducting hardware design. It ships as two products — Fluxus-Q for qubit chips and Fluxus-S for SFQ logic — on a shared verification kernel: a stated held-out error behind every engine, a conformal band where one is fitted, and a refusal outside the box it was measured on. Buy either. Buy both for co-design.

Request accessSee the public demo Every surrogate, with its validation card

Qubit · cQED

Fluxus-Q

Live in production

Design qubit chips.

Resonator frequency at 0.48% median error against SQuADDS HFSS on a grouped held-out split, TransmonCross capacitance → Hamiltonian, inverse design, frequency-collision yield, and radiation-correlated errors. Each engine names its reference and refuses outside its envelope.

RESQEMINVYLDRAD

For transmon / cQED chip designers.

Explore Fluxus-Q

SFQ logic

Fluxus-S

Deployed today

Verify SFQ logic.

Margin, timing, yield, and design-rule verification for RSFQ single-flux-quantum cells — from the netlist, JoSIM-free at serving time, measured against JoSIM on a held-out split, targeting MIT-LL SFQ5ee+.

marginstimingyieldDRV

For RSFQ / superconducting-digital designers.

Explore Fluxus-S

The shared engine

What both products stand on.

Superconducting hardware splits into two worlds — analog cQED qubits and classical SFQ logic — that almost never share a tool. Fluxus puts both on one engine: typed functions, a verification kernel, and a process registry. Swap the ground truth — closed-form physics, HFSS, JoSIM — and it changes the data the models train on, not the interface you work in.

Calibrated trust
Split-conformal intervals where an engine is calibrated, a measured held-out error where it is not, source labels telling you which backend answered, and abstention when a surrogate leaves its trained envelope. The model tells you when not to trust it.
Escalation
Where the conformal set is ambiguous, the SFQ head defers the design to JoSIM instead of answering. Every engine names the reference it was measured against — JoSIM for SFQ, SQuADDS HFSS for the resonator and TransmonCross engines — and none is validated against silicon.
Per-process calibration
Both products retarget to your fab node from a fixed-scope engagement: we train on your reference simulator and hand you the validated endpoint with its validity report.

Co-design · the bundle

Both halves, one design.

The two products are independently useful — but owned together they unlock what neither can alone. Co-design (COD) simulates SFQ control logic against qubit Hamiltonians: flux crosstalk, timing, drive fidelity, and heat load at 10 mK. It is the direction the field is moving — SFQ-controlled qubits at the cold stage — and Cogitan owns the only deployed SFQ surrogate to anchor it.

Fluxus-Q

the qubit half — live: resonator, Hamiltonians, yield

Fluxus-S

the SFQ half — deployed, measured against JoSIM

COD

co-design — needs both. Planned

Before the solve · Corpus

The cheapest solve is the one you skip.

Fluxus makes the expensive solve fast. Corpus checks whether you needed it — scanning a design against a rule database and returning violations and near-boundary warnings in seconds. The two meet in the engine room: Corpus’s cQED rules are verified against Fluxus’s own FEM and analytic solvers, never against a surrogate.

Explore Corpus

Designing superconducting hardware?
Let's talk.

Qubit chips, SFQ logic, or both for co-design — if your design loop is bottlenecked on simulation, we'd like to understand it and train Fluxus on your process.

Get in touch