Gate-to-pulse calibration reference implementation for neutral-atom devices

[hiq-lab]

Context

We maintain the Arvak quantum compilation platform, which includes a gate-to-pulse integration for Pasqal neutral-atom hardware via Pulser. As part of this work, we developed:

1. A verified gate-to-pulse mapping for all standard single-qubit gates (H, X, Y, Z, S, T, SX, RX, RY, RZ) targeting the raman_local channel, and a CZ entangling gate via Rydberg blockade (rydberg_local).

2. An empirical calibration pipeline that validates pulse parameters by comparing measurement distributions from a gate-model simulator against Pulser's QutipEmulator, using Bhattacharyya coefficient as fidelity metric.

3. Pre-calibrated pulse parameters with verified fidelities (e.g., CZ via Jaksch/Lukin pi-2pi-pi protocol at 99.96% Bell fidelity, 4µm atom spacing).

Relation to PR #182 and Issue #173

This directly complements the pulse representation types proposed in #173 / #182 (QDMI_PULSE_PARAMETER, QDMI_PULSE_SHAPE, QDMI_PULSE_OPERATION_IMPLEMENTATION). Our implementation provides concrete values for what those types would contain in a real neutral-atom device:

QDMI type	Our concrete data
QDMI_PULSE_PARAMETER	area (Rabi rotation angle), phase (XY-plane axis), post_phase_shift (virtual Z), detuning
QDMI_PULSE_SHAPE	BlackmanWaveform with duration auto-selected to respect channel amplitude limits
QDMI_PULSE_OPERATION_IMPLEMENTATION	Full gate decompositions, e.g. H = Ry(π/2) + Rz(π) via (area=π/2, phase=π/2, post_phase_shift=π)

CZ protocol detail

The two-qubit entangling gate uses a pi-2pi-pi Rydberg blockade sequence on rydberg_local:

1. π pulse on control qubit (|g⟩→|r⟩)
2. 2π pulse on target qubit (blocked if control in |r⟩, free rotation if in |h⟩)
3. π pulse on control qubit (|r⟩→|g⟩, restores state)

The conditional 2π geometric phase (-1) in SU(2) creates the CZ entanglement. CNOT is then (I⊗H)·CZ·(I⊗H).

What we'd like to contribute

We'd be happy to contribute:

1. A reference QDMI device implementation for neutral-atom hardware using the pulse types from ✨ Pulse representation specifications #182, populated with our calibrated parameters.
2. The calibration pipeline as a testing/validation tool — it could serve as a template for how device vendors verify their QDMI_PULSE_OPERATION_IMPLEMENTATION values.

The full implementation is Apache-2.0 licensed and available at:
https://github.com/hiq-lab/arvak/tree/main/crates/arvak-python/python/arvak/integrations/pulser

We're opening this issue to discuss how this would best fit into QDMI's architecture before submitting a PR.

Calibrated parameters (excerpt)

{
  "single_qubit": {
    "h": {"area": 1.5708, "phase": 1.5708, "post_phase_shift": 3.1416},
    "x": {"area": 3.1416, "phase": 0.0, "post_phase_shift": 0.0},
    "y": {"area": 3.1416, "phase": 1.5708, "post_phase_shift": 0.0},
    "z": {"area": 0.0, "phase": 0.0, "post_phase_shift": 3.1416}
  },
  "entangling": {
    "cz": {
      "protocol": "pi-2pi-pi",
      "channel": "rydberg_local",
      "fidelity_at_4um": 0.9996
    }
  }
}

/cc @mnfarooqi (PR #182 author)

[burgholzer]

Hi @hiq-lab 👋🏼

Moving this to the discussion tab, as it is not really an "issue" per-se and much better suited here.

There are two things to consider here:

• The pulse extension for QDMI is far from being fleshed out completely and it will most likely require several further changes to QDMI itself to add proper support. As such, it feels premature to try to implement an interface that is not consolidated (yet).
• Any such implementation of the interface would not be submitted to this repository as the QDMI repository itself mainly maintains the specification of the interface; not the actual implementations. Such implementations may eventually be maintained out of tree in your own GitHub organization, for example.