[论文解读] Benchmarking near-term quantum computers via random circuit sampling

Yunchao Liu, Matthew Otten+3 more|arXiv (Cornell University)|May 11, 2021

Quantum Computing Algorithms and Architecture参考文献 2被引用 25

一句话总结

本文开发基于RCS的算法，在任意非Clifford门的一层中高效估计总噪声（包括串扰），并在IBM量子硬件上进行验证，将交叉熵度量与Pauli噪声模型联系起来。

ABSTRACT

The increasing scale of near-term quantum hardware motivates the need for efficient noise characterization methods, since qubit and gate level techniques cannot capture crosstalk and correlated noise in many qubit systems. While scalable approaches, such as cycle benchmarking, are known for special classes of quantum circuits, the characterization of noise in general circuits with non-Clifford gates has been an unreachable task. We develop an algorithm that can sample-efficiently estimate the total amount of noise induced by a layer of arbitrary non-Clifford gates, including all crosstalks, and experimentally demonstrate the method on IBM Quantum hardware. Our algorithm is inspired by Google's quantum supremacy experiment and is based on random circuit sampling. In their paper, Google observed that their experimental linear cross entropy was consistent with a simple uncorrelated noise model, and claimed this coincidence indicated that the noise in their device was uncorrelated -- a key step in hardware development towards fault tolerance. As an application, we show that our result provides formal evidence to support such a conclusion.

研究动机与目标

• Motivate the need for scalable noise characterization beyond single-gate benchmarking due to crosstalk and correlated noise.
• Develop a sampling-efficient method to bound total noise in a layer of arbitrary non-Clifford gates.
• Demonstrate the method experimentally on IBM Quantum hardware and relate it to simple noise models.
• Provide formal evidence that observed cross-entropy metrics can indicate uncorrelated noise in practice.

提出的方法

• Model noise as a Pauli channel induced by a layer of arbitrary gates and show twirling reduces it to Pauli-diagonal form; aim to estimate the total error rate λ = sum of nonzero pα.
• Use random circuit sampling with Haar-random single-qubit gates interleaved with two-qubit gates in an alternating architecture to measure average circuit fidelity vs depth.
• Estimate fidelity at multiple circuit depths and fit to an exponential decay F ≈ e^(−λd) to extract λ (and SPAM factor A).
• Employ unbiased linear cross-entropy estimators (uXEB) as a sample-efficient fidelity proxy, showing consistency with true fidelity under various noise models.
• Discuss variance scaling of cross-entropy estimators with sample count M, supporting practical sampling requirements.
• Apply and validate the approach on IBM Quantum hardware to benchmark total noise per layer and diagnose crosstalk.

实验结果

研究问题

• RQ1 Can random circuit sampling provide a sample-efficient estimate of the total noise induced by a layer of arbitrary non-Clifford gates, including crosstalk?
• RQ2 Does the average fidelity of random circuits decay exponentially with depth under a Pauli-noise model, and can λ be reliably extracted from data?
• RQ3 Do linear cross-entropy metrics correlate with the true fidelity under realistic noise, including correlated and non-local noise?
• RQ4 Can RCS benchmarking distinguish non-local crosstalk from other noise sources, and what does this imply about uncorrelated noise in existing experiments?
• RQ5 Is this approach scalable to larger qubit counts, and what are the practical limitations and potential scaling pathways?

主要发现

• RCS benchmarking yields an exponential decay of average circuit fidelity with depth under a Pauli-noise model, enabling extraction of the effective noise rate λ.
• Unbiased linear cross-entropy estimators provide a sample-efficient fidelity estimate that aligns with true fidelity across i.i.d. and correlated noise models.
• Experiments on IBM Quantum hardware show exponential decay curves in both direct fidelity estimation and cross-entropy methods, with consistent λ estimates (within reported errors).
• Comparisons with simultaneous RB reveal RCS can diagnose crosstalk type effects; correlated noise that would inflate RB can be captured differently by cross-entropy, supporting conclusions about noise uncorrelation in practice.
• Simulations illustrate that correlated high-weight Pauli errors have limited impact on the estimated ENR λ, supporting the use of RCS as a global noise measure.
• The authors discuss practical scaling challenges and propose potential scaling strategies for 50+ qubits, including group-based simulations or Clifford-enhanced schemes