Platform / Decoder Engine

Decoder Engine

Real-time syndrome decoding calibrated to your hardware noise model. Sub-millisecond per cycle at code distance 5–7.

Implementation

MWPM and Union-Find

The Decoder Engine exposes both decoder algorithms under a common interface. You select the decoder type at configuration time based on your latency budget and accuracy requirements.

Per-device weight tuning is the key differentiator. Before each experiment run, QECSync ingests your hardware's calibration data and computes a per-edge weight matrix for the syndrome graph. This is not a one-size-fits-all approach — every device gets its own weight configuration.

The decode pipeline accepts a 3D syndrome array (space + time across measurement rounds) and returns a correction operator in Pauli notation, ready to apply to the logical qubit.

decoder_config.py

from qecsync import DecoderEngine, NoiseModel

# Build noise model from calibration data
noise = NoiseModel.from_json("cal_2024_11.json")

# Initialize MWPM decoder, code distance 5
eng = DecoderEngine(
    algorithm="mwpm",
    code_distance=5,
    noise_model=noise
)

# Decode 10 rounds of syndrome data
result = eng.decode_rounds(syndromes, rounds=10)
→ DecoderResult(logical_correction=..., latency_ms=0.71)

Performance characteristics

Latency and accuracy by code distance

All measurements on synthetic depolarizing noise model. Real hardware performance varies.

Code Distance	MWPM Latency	UF Latency	MWPM Accuracy	UF Accuracy	Physical Qubit Count
d = 3	<0.2 ms	<0.1 ms	99.2%	98.8%	17
d = 5	<0.7 ms	<0.3 ms	99.6%	99.1%	49
d = 7	<1.8 ms	<0.6 ms	99.8%	99.4%	97
d = 9	<4.2 ms	<1.1 ms	99.9%	99.6%	161

See benchmark comparisons in detail

View Benchmarks

Integrate via API

Read the API reference for the full decoder interface specification.

API Reference →