Technology

Nandi Q1.

Encoding, decoding and post-selection — measured on real hardware.

See the ledger

The research stack informing Nandi Q1 produced the figures below on IBM Heron-class superconducting processors.

99.94%

logical Bell fidelity — [[4,2,2]] encoded, three bases

31/31

independent encoded pairs certified in one job

5.5×

suppression, distance 3 → 5, one basis

100,000

shots per distance point, decoded offline

[[4,2,2]] encoded pairs on IBM Heron-class hardware. Fidelity 95% CI [0.999112, 0.999608], post-selected at ~96% per-basis acceptance; the 31-pair fleet job is post-selected at 9.4% (Z) / 6.0% (X) joint acceptance; suppression measured in one basis. Conditions and job IDs on the ledger.

Built for the code it runs.

Nandi Q1 is our first-generation superconducting quantum processor, currently under development. Design targets:

  • Degree-4 lattice — the code lands with zero added SWAPs
  • Mid-circuit readout
  • Qubit reuse
  • Real-time conditional logic
  • Syndrome round and coherence budgeted together

The implementation — topology map, control architecture, placement method and decode internals — is proprietary.

Every figure behind this design — the ledger →

Validation on third-party hardware.

Three-basis logical Bell state

A [[4,2,2]] encoded pair read out in ZZ, XX and YY on separate circuits, with its confidence interval and acceptance rate published beside it.

See the ledger entry

Thirty-one encoded pairs in one job

Thirty-one independent encoded pairs — 62 logical qubits — each above the witness threshold with its own confidence interval.

See the ledger entry

A repetition-memory ladder

100,000 shots per distance, decoded offline with minimum-weight matching: a 5.5× suppression from distance 3 to 5, and the floor where it stops.

See the ledger entry

The connectivity tax

Routed onto degree-3 heavy-hex, a surface code costs roughly 270 two-qubit gates at distance 3 and 2,100 at distance 5. We measured what that does to the logical state.

Read the design brief

Every figure re-derives offline

Raw per-shot arrays are archived at submission and every number on this page is rebuilt from them on a laptop. Decode conventions are frozen before a job is sent, so the analysis cannot be tuned to the result.

See the ledger

Placement that earns its couplers

Topology-aware placement — the compiler adds no detours. The same discipline that certified one encoded pair placed the whole fleet in a single submission.

Read the note

Research in preparation.

Research papers on connectivity costs, Bell-state measurements and decoder validation are in preparation.

Explore the research programme →