How Randomized Benchmarking Measures Quantum Gate Performance
Randomized benchmarking estimates average gate errors by tracking how sequences of random operations lose fidelity as they grow longer.
By The Quantum Lab desk
Written in-house by the The Quantum Lab desk — an explainer, not a report of a news event.
- Randomized benchmarking separates sequence-dependent gate errors from fixed state-preparation and measurement errors.
- Its decay parameter estimates average performance over a chosen set of operations, not the error of every individual gate.
- Additional diagnostics are needed to identify leakage, crosstalk, coherent drift and other structured faults.
Quantum processors need a way to assess control operations without relying on a perfectly calibrated measurement apparatus. Randomized benchmarking addresses this by applying sequences of randomly selected gates, commonly drawn from the Clifford group, followed by a gate that should invert the entire sequence. In an ideal processor, the qubit returns to a known state regardless of the random choices.
Researchers repeat the experiment for many random sequences and several sequence lengths. As errors accumulate, the probability of recovering the expected state typically decreases with sequence length. Fitting this decay yields a parameter related to the average error per operation under assumptions about the noise.
A key advantage is reduced sensitivity to state-preparation and measurement errors. Those imperfections mainly affect the fitted offset and scale, while the decay rate is associated with errors accumulated during the gate sequence. This makes randomized benchmarking useful for comparing control calibrations even when readout is imperfect.
The method does not provide a complete description of a device's noise. Coherent miscalibrations, time-correlated noise, leakage outside the computational states and crosstalk can produce behavior that a simple exponential fit obscures. The reported average can also hide substantial differences among gates, qubits or operating conditions.
Variants extend the basic protocol for narrower questions. Interleaved randomized benchmarking inserts a target gate between random gates to estimate its performance relative to a reference experiment, while simultaneous benchmarking can probe interactions among qubits operated at the same time. These measurements are most informative when combined with techniques such as gate-set tomography, leakage detection and stability monitoring.
This story started as a thread in our Discord. Join the discussion with the researchers and engineers who work on it.
Related
Surface-code processor stays below threshold as it scales to 1,000 qubits
The logical error rate keeps falling as the code distance grows — the clearest sign yet that adding qubits now buys reliability instead of noise.
Trapped-ion shuttling record pushes two-qubit fidelity past 99.9%
Moving ions between zones used to cost coherence. A revised trap geometry makes transport nearly free.
Open-source compiler cuts logical-qubit overhead by a third
A new lattice-surgery scheduler squeezes routing overhead out of fault-tolerant circuits, and it ships under a permissive license.
