Source-verifiedLevel 3

Calibration

Calibration is the continuous, sequential stack of experiments that tunes each qubit's control pulses; its data defines which gates exist on which qubits and is the QPU's source of truth.

What it means

A QPU is not a deterministic machine but a sensitive analog physics experiment.Three things make calibration unavoidable: manufacturing inhomogeneity means every qubit is a snowflake — each has slightly different resonance frequency, anharmonicity, and coupling strength, so Q1's NOT pulse is not Q2's NOT pulse; environmental drift — millikelvin temperature fluctuations, stray electromagnetic fields, substrate defects — makes properties wander over time; and cross-talk lets a pulse aimed at one qubit leak into its neighbor.Calibration is therefore not a single experiment but a sequential stack, each step built on the last: spectroscopy sweeps microwave frequencies to find where the qubit responds (superconducting qubits resonate at 4–8 GHz); Rabi oscillation varies pulse amplitude and duration while plotting P(1) — the first peak is the π-pulse, i.e.the calibrated X gate; Ramsey experiments refine phase and frequency and measure T2; two-qubit tuning (CZ/CNOT) is the hardest and slowest; even measurement and active reset are operations that need calibrating.Initial calibration of a large QPU takes hours, and because T1/T2 drift daily, recalibration is continuous.The resulting calibration data structure is the source of truth for which gates exist where — the transpiler consults it.Quality is graded by Randomized Benchmarking: survival probability decays exponentially with sequence length, and the decay rate yields the average gate fidelity (target 99.9%+).

Everyday analogy

A quantum computer is a violin that goes out of tune when the temperature shifts even slightly — and every string (every qubit) is a unique snowflake. You must retune before the performance and even during it, because the exact push needed to go from do (0) to re (1) differs per string and per hour. The tuning stack: spectroscopy sweeps frequencies to find where the qubit 'wakes up' (like finding the note that makes a wine glass ring), Rabi oscillation finds the exact push for a perfect flip (the π-pulse), Ramsey fine-tunes phase and frequency, two-qubit tuning is the hardest, and Randomized Benchmarking is the report card.
Etymology hook: calibration traces back to Arabic qālib, 'mold', which became Latin/French calibre, 'the bore of a gun' — measuring and correcting against a standard. Fidelity comes from Latin fidelis, 'faithful' — it is the Fi in Hi-Fi (high-fidelity audio): how faithfully the physical operation reproduces the ideal one.

Common misconceptions

  • Calibration data and its quality metrics are different things people confuse: some mean the pulse numbers themselves (amplitudes, durations, frequencies), others mean how good those numbers are (fidelity). The data says what the gates are; fidelity says how well they work.
  • Calibration is not a one-time setup — T1/T2 and other properties drift daily due to temperature fluctuations, stray fields, and substrate defects, so recalibration is continuous, not once.
  • RB is not a lightweight check — Randomized Benchmarking stress-tests the ENTIRE stack at once (transpiler, compiler, pulse scheduler, data transfer, job scheduler, control electronics, and the QPU itself), which is exactly why it is the heaviest job.

Key takeaways

  • Superconducting qubit resonance frequencies lie at 4–8 GHz — spectroscopy sweeps this band to find each qubit's transition.
  • Initial calibration of a large QPU takes hours; T1/T2 drift daily, so recalibration is continuous, not once.
  • Target gate fidelity is 99.9%+ (two-qubit gates are harder), graded by the exponential decay rate of the Randomized Benchmarking survival curve.
  • The calibration data structure is the source of truth for which gates exist on which qubits and pairs — if a CNOT between two qubits was never calibrated, the transpiler must reject or reroute it.

Check your understanding

In a Rabi oscillation experiment, what does the first peak of P(1) versus pulse amplitude/duration give you?

  1. A.The qubit's resonance frequency
  2. B.The coherence time T1 of the qubit
  3. C.The π-pulse parameters — the exact amplitude and duration of a complete 0→1 flip, i.e. the calibrated X gate
  4. D.The readout error rate of the measurement channel
Show the answer

Answer: C. The π-pulse parameters — the exact amplitude and duration of a complete 0→1 flip, i.e. the calibrated X gate

Why: Sweeping the drive pulse's amplitude/duration while plotting the probability of measuring |1⟩ traces a sinusoid (the Rabi oscillation). The first maximum is the π-pulse — a complete 0→1 flip — and its amplitude/duration pair is stored as the calibrated X gate. Resonance frequency comes from spectroscopy, done earlier in the stack.

In Randomized Benchmarking, what does the decay rate of the exponential survival curve measure?

  1. A.The qubit's resonance frequency drift per hour
  2. B.The average gate fidelity — a gentler decay means higher fidelity (e.g. 99.92%)
  3. C.The number of SWAP gates the router inserted
  4. D.The temperature of the dilution refrigerator
Show the answer

Answer: B. The average gate fidelity — a gentler decay means higher fidelity (e.g. 99.92%)

Why: RB runs random gate sequences of increasing length, each ended with an inverse gate that should return the qubit to |0⟩. Errors accumulate with length, so survival probability decays exponentially — and the decay rate translates into the average gate fidelity. A specific gate's fidelity (e.g. CNOT) is measured with Interleaved RB.

Builds on

QubitDecoherence

Primary source: Krantz et al., Appl. Phys. Rev. 6, 021318 (2019), doi:10.1063/1.5089550; Magesan, Gambetta & Emerson, Phys. Rev. Lett. 106, 180504 (2011), doi:10.1103/PhysRevLett.106.180504

Calibration experiment stack (spectroscopy/Rabi/Ramsey) per Krantz; RB per Magesan 2011; 2026 tutorial arXiv:2606.03815 verified 2026-07-03.

Learn it hands-on

This concept is part of a 46-level curriculum with an interactive simulator and Lumen, a tutor whose answers are verified before you see them. Levels 1–5 are free.

Start freeSee the curriculum