Calibration and Drift
Qubit parameters drift over hours, so quantum computers are continuously recalibrated to keep gates and readout accurate.
Nothing stays put
Qubit frequencies, gate amplitudes, readout thresholds, and coupling strengths are not fixed constants. Two-level-system defects switch, flux and charge environments wander, temperatures and magnetic fields drift, and materials age. A gate calibrated perfectly this morning may be visibly worse by afternoon. Keeping a processor performing near its best is a continuous, automated task, not a one-time setup.
What gets calibrated
- Qubit transition frequencies and anharmonicities
- Single-qubit pulse amplitude, phase, and DRAG coefficient
- Two-qubit gate parameters (flux pulse shape, drive detuning)
- Readout frequency, power, and discrimination thresholds
- Crosstalk and cable-distortion compensation
Calibration hierarchies
Because parameters depend on one another, calibration is organized as a dependency graph: coarse spectroscopy first, then finer tune-ups that assume the earlier results. Automated systems such as the Optimus scheme decide which nodes need re-measuring based on how stale or drifted they are, avoiding a full recalibration when only a few parameters moved.
Drift and its causes
Drift ranges from slow aging to abrupt jumps when a nearby defect switches state. Two-level-system defects can drift into resonance with a qubit and sharply lower its T1 for minutes to hours. Cosmic rays cause brief bursts of correlated errors. Monitoring these fluctuations is part of both operations and research, since understanding drift points to the underlying noise sources to fix.
At scale, calibration becomes a bottleneck of its own: recalibrating thousands of qubits and gates fast enough is an active engineering problem, pushing toward machine-learning-assisted and always-on calibration.