arXiv:2610.18000
Development and experimental implementation of noise characterisation techniques on a superconducting quantum processor
Nelson M. M. Fernández, Francesco Grieco, Hetta Vappula, Stefano Mangini, Boris Sokolov, Elsi-Mari Borrelli, Matteo A. C. Rossi, Sergey N. Filippov, N. Walter Talarico
Abstract
Noise characterisation not only enables error mitigation but also provides valuable information about the intrinsic performance of a quantum device, identifying dominant error contributions and bounding noise stability well beyond the aggregated calibration data. Deploying such methods on a specific hardware topology must combine efficiency and accuracy, assess noise stability, and validate the reconstructed noise model. Here we develop and implement these methods on a superconducting processor with square-lattice topology. First, we revise the cycle-benchmarking protocol and reduce the bias arising from unlearnable degrees of freedom via a regularised cost function. Second, we extend a previously developed tensor-network tomography method to match the device layout and show how its accuracy can be improved by repeated layer execution. Both methods allow approximate sublayer noise models to be extracted from parent ones. Using noise characterisation as a tool, we map the noise structure, track drift, and validate the adequacy of the learnt noise models against experimentally observed noisy estimations in benchmark physical and chemistry circuits. These results inform further improvements in the underlying technology and provide estimates of the size of noise-mitigated circuits that the device can support.
AI Summary
The paper examines how efficiently learned layer-noise models predict circuit outcomes on a square-lattice superconducting processor, using experiments on four- to ten-qubit subsets of a 53-qubit device. It combines noise reconstruction, sublayer extraction, drift measurements, and validation against Loschmidt-echo and quantum-chemistry circuits. Two characterisation methods are developed. For cycle benchmarking, a nonnegative ridge fit replaces symmetry-based assumptions about individually unlearnable Pauli fidelities. Simulations show reduced coefficient bias for the tested rate distribution; regularisation selects a gauge-compatible model rather than resolving unlearnability. Tensor-network channel tomography is extended from chains to square-lattice locally purified networks, with detector tomography correcting measurement errors. Joint fitting across repeated layer applications improves simulated reconstruction: at $5\times10^4$ shots, normalised squared Frobenius error decreases from $7.7\times10^{-3}$ to $4.4\times10^{-4}$. Hardware validation gives mixed accuracy. Tensor-network predictions underestimate echo return probabilities by approximately $14$–$16\%$ on a ten-qubit chain and $26\%$ on a $2\times5$ lattice. For a ten-qubit chemistry circuit with 36 CZ gates, predicted and measured energies differ by $0.0155$ Ha, or $0.87$ measurement standard errors. Sublayer models are approximately extracted from parent layers. Four characterisations of a favourable four-qubit subset indicate $1.59\%$ relative noise variation over roughly four hours. Limitations include uncorrected preparation errors, approximate trace preservation, restricted model expressivity, contraction costs exponential in lattice width, and inconsistent hardware benefits from repeated-layer fitting. A projected mitigation accuracy below $13\%$ for approximately 400 CZ gates and $10^8$ shots assumes favourable small-register noise and stability scale to 50 qubits; it is not experimentally demonstrated.
Error found: Repeated-layer tomography improves the controlled simulation and the untwirled hardware example, but worsens the reported zero-angle prediction error on the twirled hardware acquisition. The demonstrated reduction of bias in unlearnable rates depends on a narrowly distributed simulation ensemble and ground-truth tuning, and does not establish general coefficient accuracy. Tensor-network echo predictions retain systematic discrepancies of approximately 14–26% on the principal chain and lattice examples, despite successful agreement for the chemistry energy. The stated TEM scaling gives a random-error contribution of approximately 62.8 at depth 30, contradicting the claim of below 10% error across depths 15–30. Quadratic regularisation is not generally equivalent to Shannon-entropy maximisation under the measured-product constraints, so the claimed maximum-entropy interpretation is incorrect.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.