aiqbench analysis
This study evaluates Digital Annealer strategies for optimizing CNOT gate overhead in NISQ circuit transpilation, demonstrating significant reductions in equivalent gate counts compared to standard Qiskit and t|ket> baselines. While the hybrid approach of combining DA-based global mapping with heuristic routing yielded consistent improvements, full DA routing faced feasibility limitations on specific benchmarks, highlighting current constraints in short-horizon classical optimization workflows.
Digital Annealer-Assisted CNOT Oriented Quantum Circuit Transpilation with Integrated QUBO Mapping and Routing
arXiv:2605.11500v2 Announce Type: replace Abstract: Limited qubit counts and two-qubit gate errors motivate reducing CNOT overhead in noisy intermediate-scale quantum (NISQ) transpilation. In circuit-partitioning workflows, repeated hardware sampling can motivate additional classical search, although this study does not evaluate end-to-end runtime or hardware fidelity. We evaluate two Digital Annealer (DA) strategies on seven fixed standalone benchmark instances and a 64-node 8 x 8 square-grid. Hybrid combines DA-based global initial mapping with Qiskit routing, whereas Full DA applies DA to both initial mapping and iterative routing. Hybrid achieved lower mean equivalent CNOT counts than both configured baselines on four benchmarks and identical means on the remaining three. The unweighted macro-average reductions were 6.14% relative to Qiskit and 17.9% relative to $t|ket\rangle$. Among the six benchmarks with feasible Full DA outputs, Full DA achieved lower means than ISAAQ on five, with a success-conditioned macro-average reduction of 48.9%; however, it produced no feasible BV output and did not outperform Hybrid on any benchmark. These results identify DA-generated global placement followed by heuristic routing as the primary positive result and characterize the limitations of the evaluated short-horizon DA routing workflow. Equivalent CNOT count is a structural proxy and does not directly measure execution fidelity.
Abstract:Limited qubit counts and two-qubit gate errors motivate reducing CNOT overhead in noisy intermediate-scale quantum (NISQ) transpilation. In circuit-partitioning workflows, repeated hardware sampling can motivate additional classical search, although this study does not evaluate end-to-end runtime or hardware fidelity. We evaluate two Digital Annealer (DA) strategies on seven fixed standalone benchmark instances and a 64-node 8 x 8 square-grid. Hybrid combines DA-based global initial mapping with Qiskit routing, whereas Full DA applies DA to both initial mapping and iterative routing. Hybrid achieved lower mean equivalent CNOT counts than both configured baselines on four benchmarks and identical means on the remaining three. The unweighted macro-average reductions were 6.14% relative to Qiskit and 17.9% relative to $t|ket\rangle$. Among the six benchmarks with feasible Full DA outputs, Full DA achieved lower means than ISAAQ on five, with a success-conditioned macro-average reduction of 48.9%; however, it produced no feasible BV output and did not outperform Hybrid on any benchmark. These results identify DA-generated global placement followed by heuristic routing as the primary positive result and characterize the limitations of the evaluated short-horizon DA routing workflow. Equivalent CNOT count is a structural proxy and does not directly measure execution fidelity.
7 pages, 4 figures. Accepted to IEEE Quantum Week (QCE) 2026
arXiv:2605.11500 [quant-ph]
(or arXiv:2605.11500v2 [quant-ph] for this version)
https://doi.org/10.48550/arXiv.2605.11500 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 295 words
