Monitoring, calibrations, and benchmarking

Calibrating a quantum computer requires optimizing a plethora of parameters defining the electrical signals that actuate the quantum gates and readout operations. Before a quantum computer is released, its initial calibration process involves thoroughly tuning each parameter to achieve the best possible performance based on benchmarks relevant to the quantum computer's expected workloads. Once a quantum computer has been released, the primary goal is to maintain consistent performance for the lifetime of the device. The optimal values of many of the calibrated parameters remain stable indefinitely, but some vary over time as a result of factors such as changes in the environment of two-level systems (TLS) on the quantum processor chip, changes in the ambient conditions (for example, temperature) in the data center, or instability within the control systems.

To ensure consistent performance, IBM® quantum computers are frequently monitored to track parameters that may drift over time, run calibrations when needed, and perform daily benchmarking. This page details three processes — monitoring, calibrating, and benchmarking — that work together to ensure that the IBM fleet of quantum computers remains as stable, predictable, and available to users as possible.

Monitoring

Parameter monitoring

Brief parameter monitoring jobs are executed approximately once an hour, automatically interleaved between user jobs, using the full Qiskit Runtime software stack. The results of these jobs are analyzed to check whether any parameters have begun to deviate from acceptable ranges, ideally catching issues before they become significant enough to affect performance to a noticeable level.

Some of the parameters monitored include the following:

• Readout angles, amplitudes, and discriminator threshold, ensuring accurate state discrimination, low leakage and stable operation. This includes the operational parameters of our quantum limited amplifiers.
• Single- and two-qubit gate operations, confirming they behave as expected to maintain correct rotation angles and minimize phase and amplitude errors.
• Signatures of TLS activity.

If the results of these monitoring jobs indicate modest deviations from their expected performance, then appropriate calibration jobs are launched. If severe TLS activity is detected, then the calibration strategy for gates associated with affected qubits might be automatically modified (and could include pausing calibrations) until such TLS activity decreases back to acceptable levels.

Holistic monitoring

In addition to jobs monitoring individual parameters are jobs that monitor the performance of the quantum computer more holistically, such as tests that look at the fidelity of generated Bell states, as well as tests of fractional gates and dynamic circuits on quantum computers that support those features. The goal of these tests, which also run through the full Qiskit Runtime stack interleaved with user jobs, is to efficiently validate the overall behavior of the hardware and software. If these tests detect a significant drop in performance, the quantum computer will automatically pause the job queue until the issue is resolved, ensuring that user jobs won't run until the device is performing as expected again.

Calibrating

Calibrations are triggered whenever monitoring jobs indicate that parameters such as pulse amplitudes or angles have deviated from their ideal values. They run throughout the day in between user jobs, and therefore there is no set time period at which calibrations start and finish. These run only on the qubits/gates for which parameter monitoring has identified specific issues, along with any qubits required to run at the same time according to specific batching rules. On Heron QPUs, the total time spent on calibrations is typically under two hours per day.

Single-qubit operations

These calibrations ensure accurate implementation of the single-qubit gates: sx, x, rx (fractional). We adjust:

• Qubit frequencies
• Pulse amplitudes and phases

These calibrations are batched across the affected qubits and executed concurrently where appropriate, with batching strategies tailored to each calibration type.

Two-qubit operations

• CZ and RZZ gate amplitudes and phases (for Heron and Nighthawk processors)
• ECR gate amplitudes and phases (for Eagle processors)

These calibrations are run in batches of non-nearest-neighbor qubits to minimize crosstalk.

Readout

• Readout pulse angles
• Measurement discrimination parameters

These calibrations are run simultaneously on qubits that require calibration.

How calibrations are scheduled

• A calibration job cannot run simultaneously while a job or session is running.
• Therefore, during long sessions, the quantum computer may experience reduced effective stability due to delayed or infrequent recalibration.
• Two jobs submitted at the same time might run under different calibration sets, depending on timing.

Benchmarking

Daily benchmarking gives a comprehensive view of quantum computer performance and generates the metrics sent to users through Qiskit. They help users choose qubits, optimize compilations, and better anticipate expected circuit performance. You can view reported numbers either programmatically or on the Compute resources page (click any QPU to open its detailed information card). Find more details about each metric in the documentation.

Single-qubit performance

• Randomized benchmarking (RB) in batched groups
• Coherence times for $T_1$ and $T_2$
• Measurement-fidelity metrics

Two-qubit performance

• CZ and RZZ fractional gate EPG (Heron), ECR (Eagle) as measured by RB on those gates

System-level metrics

• Layer fidelity (EPLG), for the best 100-qubit length string