Hardware Integration Guide

How calibration data is used

QECSync builds a syndrome graph edge weight matrix from your device's calibration data. Each edge in the syndrome graph corresponds to a pair of adjacent stabilizer measurements; the weight represents the probability of an error causing that edge defect under your device's actual noise rates.

Correct weights are the single largest factor in MWPM decoder accuracy. A decoder using uniform weights treats all edges as equally likely; a decoder using per-device weights exploits the fact that some qubit pairs have much higher error rates than others, and routes corrections accordingly.

IBM Quantum calibration export

From the IBM Quantum dashboard, export your backend's calibration data as a JSON file. QECSync's parser reads the qubits and gates fields to extract T1, T2, single-qubit gate error rates, and two-qubit CNOT/ECR error rates per edge.

from qecsync import DeviceSpec, DecoderEngine

# Load calibration from IBM dashboard export
device = DeviceSpec.from_ibm_json(
    "ibm_sherbrooke_calibration.json"
)

# Inspect loaded properties
print(device.qubit_count)    # 127
print(device.topology)       # "heavy-hex"
print(device.mean_t2_us)     # 184.3

# Build decoder from device
eng = DecoderEngine.from_device(device, code_distance=5)

Quantinuum H-series integration

Quantinuum's H-series trapped-ion devices have all-to-all connectivity and longer coherence times. The DeviceSpec parser handles the different topology and native gate set (Mølmer-Sørensen gate instead of CNOT).

from qecsync import DeviceSpec

device = DeviceSpec.from_quantinuum_json(
    "h1_cal_20251102.json"
)
print(device.topology)       # "all-to-all"
print(device.native_gates)   # ["ZZ", "Rz", "Ry", "MS"]

Generic JSON format

For hardware not covered by built-in parsers, QECSync accepts a generic calibration schema. Contact us for the schema specification. Integration support for new device types is available on request.

Re-calibration cadence

Device calibration data drifts over time as qubit properties shift. For best results, refresh your calibration export before each experiment session and reload the DeviceSpec. The DecoderEngine initialization with updated weights takes under 1 second for code distances up to d=9.