| Randomized benchmarking | |
|---|---|
| Name | Randomized benchmarking |
| Field | Quantum information |
| Invented | 2000s |
| Inventor | Emanuel Knill et al. |
| Applications | Quantum gate characterization, Quantum error correction |
Randomized benchmarking
Randomized benchmarking is a statistical protocol for assessing the average error rates of quantum gates on physical qubits by applying sequences of random operations and measuring survival probabilities. It provides a scalable, noise-resilient estimate of gate fidelity that is less sensitive to State Preparation And Measurement (SPAM) errors than direct quantum tomography. In the context of Quantum Physics and engineering of quantum computing devices, randomized benchmarking helps translate laboratory performance into metrics used for fault-tolerant quantum computing roadmaps and policy decisions about resource allocation.
Randomized benchmarking (RB) sits at the intersection of experimental quantum information science and statistical characterization of open quantum systems. RB interrogates quantum coherence and decoherence processes described by quantum channels and completely positive trace-preserving maps. The method exploits group theory, commonly using the Clifford group for single- and multi-qubit gates, connecting to the theoretical framework of quantum error correction and stabilizer codes. By averaging over random sequences, RB isolates gate-dependent error accumulation from SPAM noise, complementing techniques such as process tomography and gate set tomography developed by groups at institutions like National Institute of Standards and Technology (NIST) and University of Waterloo.
Basic RB protocols generate random sequences of operations drawn from a chosen group (often the Pauli group or Clifford group) of increasing length, append an inverting operation, and measure the probability of returning to an initial reference state. Fitting the decay of survival probability versus sequence length yields an average error parameter, often reported as an average gate fidelity or "error per gate". Key implementations include standard Clifford RB, simultaneous RB for cross-talk analysis, and cycle benchmarking tied to quantum processor clock cycles. Experimental platforms from IBM Quantum and Google Quantum AI commonly implement RB integrated into calibration cycles alongside quantum control techniques and pulse-level optimization.
RB assumes Markovian, time-independent, and gate-independent noise for its simplest interpretations; deviations motivate more complex error models such as non-Markovian dynamics and temporally correlated noise. Analysis often models errors via Pauli twirling to produce effective depolarizing channels, enabling simple exponential decay fits. Statistical inference employs maximum likelihood estimation and bootstrap resampling to quantify uncertainty; advanced techniques include Bayesian model selection to compare hypotheses of gate dependence or leakage. Theoretical work by researchers including Emanuel Knill, Joseph Emerson, and Robin Blume-Kohout established foundational error bounds and connections to operationally meaningful metrics like average fidelity and diamond norm bounds.
Extensions address practical complications and broaden diagnostic reach. Interleaved randomized benchmarking estimates the error rate of a specific gate by alternating it with random Clifford sequences, enabling per-gate characterization. Leakage RB measures population transfer out of the computational subspace, relevant for superconducting transmon qubits and trapped-ion leakage channels. Tomography-free or direct RB variants aim to reduce resource overhead by avoiding full reconstruction, while cycle benchmarking and mirror RB adapt the protocol for large-scale multi-qubit processors. These variants have been developed and validated by research groups at University of Maryland, Yale University, and industrial labs such as Rigetti and Microsoft Quantum.
RB has been implemented across major qubit modalities: superconducting qubits (e.g., Google Sycamore experiments), trapped ions (e.g., IonQ and academic ion-trap groups), semiconductor spin qubits, and photonic platforms. Results are reported as average gate fidelities exceeding thresholds relevant to surface code thresholds in some systems, while revealing modality-specific error sources like leakage in transmons or motional decoherence in ion traps. Large-scale benchmarking campaigns inform hardware roadmaps at companies like IBM and Honeywell Quantum Solutions and in national initiatives such as the U.S. National Quantum Initiative.
RB's assumptions can bias estimates when noise is gate-dependent, non-Markovian, or when SPAM errors vary across populations of devices. Benchmarking data often concentrates at well-funded institutions and commercial providers, producing uneven visibility and potentially skewing public understanding of progress. Equity considerations include access to calibration tools, data sharing practices, and availability of open datasets; transparent protocols and community benchmarks led by organizations like Quantum Information Science Research Centers and international collaborations can mitigate disparities. Responsible deployment of RB should account for reproducibility, publication bias, and the societal implications of concentrating quantum technological capability.
Randomized benchmarking informs error budgets for fault-tolerant architectures, helping determine required overhead for quantum error correction and logical qubit construction. RB metrics feed into resource estimates for algorithms such as Shor's algorithm and quantum simulation workloads, guiding investment decisions in hardware improvements and control engineering. Policymakers and funding agencies use aggregated benchmarking outcomes to prioritize funding toward equitable capacity-building programs, open infrastructure, and workforce development to broaden participation in the emergent quantum economy.
Category:Quantum information science Category:Quantum measurement