Measurement-based decoupling boosts qubit fidelity on IBM’s Eagle processor

Introduction

A team led by Jeongwoo Jae of Samsung SDS has introduced measurement‑based dynamical decoupling (MDD), a protocol that extracts partial measurements from idle qubits, reconstructs their reduced density matrices, and applies tailored local unitaries to keep the qubits aligned with their ground state. Tested on IBM’s 127‑qubit Eagle processor, MDD produced a 450‑fold increase in the success probability of a 14‑qubit quantum Fourier transform (QFT) and improved ground‑state energy estimates for nitrogen‑dimer simulations on 56 qubits. The key advantage is a measurement overhead that scales linearly with the number of subsystems, making the approach practical for large‑scale devices.

How MDD Works

1. Partial Pauli‑basis measurements are performed on each idle qubit during a circuit’s idle periods.
2. From these results the single‑qubit reduced density matrix is reconstructed, giving expectation values \(\langle X\rangle, \langle Y\rangle, \langle Z\rangle\).
3. A local unitary \(Ud = Ry(-\thetad)Rz(-\phi_d)\) is computed so that the qubit’s state is rotated as close as possible to the ground state \(|0\rangle\).
4. The unitary is applied immediately, effectively cancelling the dominant amplitude‑damping and dephasing channels identified by the measured Kraus operators.

Unlike conventional dynamical decoupling, which relies on pre‑programmed pulse sequences (e.g., XY‑4, UDD), MDD adapts in real time to the actual noise observed on each qubit. This adaptive nature allows it to achieve the maximal entanglement fidelity attainable by any bang‑bang scheme, at least to first order in evolution time.

Demonstrated Performance on Eagle

• Relaxation / dephasing times: 250 µs (T₁) and 170 µs (T₂) – typical for Eagle’s transmons.
• QFT benchmark: The 14‑qubit QFT success probability rose from ~0.2 % (standard DD) to >90 % when MDD was applied, a 450‑fold improvement.
• N₂ diagonalization: In a 56‑qubit sample‑based quantum diagonalization, MDD reduced the error in ground‑state energy estimation by a factor of three compared with the standard XX‑pulse DD.
• Scalability: Measurement overhead grows as \(O(N)\) rather than \(O(N^2)\) seen in graph‑embedding or learning‑based DD methods, keeping classical post‑processing tractable even on fully connected architectures.

Why Linear Overhead Matters

As quantum processors scale, idle‑time decoherence becomes a dominant error source. Traditional error‑mitigation techniques often require quadratic resources to model cross‑talk and connectivity, quickly becoming infeasible. MDD’s linear scaling means that adding more qubits only modestly increases the number of required measurements, preserving the protocol’s practicality for future devices with hundreds or thousands of qubits.

Practical Takeaways

• Implementation simplicity: MDD primarily uses the \(R_y\) rotation, which can be realized virtually via phase updates, reducing hardware demands.
• Broad applicability: The protocol improves not only algorithmic benchmarks like QFT but also scientific simulations (e.g., molecular energy calculations) where high fidelity is essential.
• Compatibility: MDD can be layered on top of existing transpilation pipelines, targeting idle periods identified during circuit optimization.

Closing Thoughts

Measurement‑based dynamical decoupling offers a scalable, noise‑aware route to preserving quantum information on today’s noisy intermediate‑scale quantum (NISQ) hardware. By turning partial measurements into actionable control signals, MDD bridges the gap between classical error‑characterization and real‑time quantum error mitigation. Its successful validation on IBM’s Eagle processor suggests that similar gains are within reach for other superconducting platforms, paving the way for more reliable large‑scale quantum computations.