Sketch of the phase diagram. The region labeled “weak noise” in green represents where quantum correlations extend to the full device, enabling beyond classical experiments. In the “strong noise” regime, the system may be approximately represented by the product of multiple uncorrelated subsystems, and the experiment might be spoofable by classical algorithms. The anti-concentration phase transition separates the regimes of a concentrated distribution of output bitstrings and a broad (or anti-concentrated) distribution. In the experiment they identified this transition for noisy RCS circuits. Credit: Google Quantum AI

Google Quantum AI has made a significant advancement by doubling the quantum circuit volume in its Random Circuit Sampling (RCS) benchmark. Utilizing the Sycamore processor, this achievement elevates the complexity of quantum computations while maintaining high fidelity compared to previous benchmarks established in 2019. The updated RCS benchmark continues to be a crucial tool for evaluating quantum computers’ performance in tasks that challenge classical supercomputers.

In a recent experiment, the Sycamore processor executed circuits with 67 qubits and 32 cycles, involving 880 two-qubit gates. This result signifies an increase in circuit volume while preserving fidelity at 1.5 × 10⁻³, demonstrating Google’s quantum systems’ ability to handle more intricate circuits despite noise interference.

The study also highlights two distinct quantum phase transitions as noise levels rise with the number of qubits, revealing the operational behavior of quantum systems under varying noise conditions. This emphasizes the importance of operating within a low-noise regime to achieve quantum advantage, where quantum systems can outperform classical systems.

To validate the RCS results, the research team employed cross-entropy benchmarking (XEB). By comparing the outputs of a noisy quantum processor to an ideal, noise-free system, researchers estimated the overall fidelity, which remained consistent despite the circuit volume increase, reinforcing the reliability of RCS as a benchmarking approach.

The study further investigated classical “spoofing” algorithms’ effectiveness in mimicking quantum performance in RCS scenarios. Results indicated the classical methods were unable to replicate Google’s quantum processor outcomes, underscoring RCS’s credibility in evaluating quantum hardware capabilities as systems scale in size and complexity, highlighting quantum computing’s advantages over classical approaches.

For more information, visit Google Quantum AI’s blog here or read the full paper in Nature here.

October 12, 2024