ADAPT: Mitigating Idling Errors in Qubits via Adaptive Dynamical Decoupling

Poulami Das, Poulami Das 0005, Swamit S. Tannu, Siddharth Dangwal, Moinuddin K. QureshiMicro 2021 · DOI

TL;DR

This paper introduces ADAPT, a novel software framework that intelligently applies Dynamical Decoupling (DD) to mitigate idling errors in quantum computers. By adaptively identifying which qubits benefit most from DD, ADAPT significantly improves application fidelity, outperforming both no-DD and DD-on-all strategies.

What Problem Does This Paper Address?

Near-term quantum computers, often referred to as Noisy Intermediate-Scale Quantum (NISQ) devices, are plagued by hardware errors that limit the fidelity of computations. Beyond errors occurring during gate operations, qubits are also susceptible to "idling errors" that accumulate when a qubit is not actively performing any task. These errors can arise from various sources, including decoherence, dephasing, and crosstalk from nearby active qubits. While Dynamical Decoupling (DD) is a known technique to combat these idling errors by applying specific pulse sequences, prior work has primarily focused on its application at the device level or for individual qubits. The effectiveness of DD at the application level, especially when applied indiscriminately to all idle qubits, is not well understood and can sometimes even degrade performance due to the overhead of the DD pulses themselves. This paper addresses the challenge of robustly and judiciously applying DD at the application level to maximize quantum program fidelity.

Methods

The core innovation presented is the Adaptive Dynamical Decoupling (ADAPT) framework. ADAPT operates as a compiler pass that integrates into existing quantum compilation workflows. Its key steps are:

  • Decoy Circuit Generation: ADAPT constructs a "Decoy Circuit" that mirrors the structure of the original quantum program, particularly its CNOT gate layout, which is a major source of idling errors. To make these decoy circuits efficiently simulatable on classical computers, ADAPT uses a Clifford Decoy Circuit (CDC) approach, replacing non-Clifford gates with their closest Clifford equivalents. To further improve the correlation between the decoy and original circuits, a Seeded Decoy Circuit (SDC) is proposed, which uses a limited number of non-Clifford gates in the initial layers.
  • DD Sequence Search: ADAPT uses the decoy circuit to estimate the effectiveness of different DD sequences. Instead of exhaustively searching all possible combinations (which scales exponentially), ADAPT employs a localized search algorithm that evaluates DD combinations within small neighborhoods of qubits (e.g., 4 qubits at a time). This reduces the search complexity to be linear with the number of qubits.
  • Optimal DD Application: The DD sequence that maximizes the fidelity of the decoy circuit is then applied to the original quantum program. ADAPT supports two common DD protocols: XY-4 and IBMQ-DD.

Datasets

The paper does not describe datasets used for training. Instead, it focuses on experimental evaluation using existing quantum benchmarks.

  • Benchmarks: The paper utilizes several standard quantum benchmarks, including Bernstein-Vazirani (BV), Quantum Fourier Transform (QFT), and Approximate Optimization Algorithm (QAOA), with varying numbers of qubits (from 5 to 10). These benchmarks are chosen to represent different structures and complexities relevant to NISQ applications.
  • Hardware Platforms: Experiments were conducted on three IBM quantum computers: IBMQ-Guadalupe (16 qubits), IBMQ-Paris (27 qubits), and IBMQ-Toronto (27 qubits). These platforms were selected to test ADAPT's performance across different hardware characteristics and calibration cycles.

Key Insights and Ideas

  • Indiscriminate DD is Sub-optimal: Applying DD to all idle qubits does not always improve fidelity and can sometimes degrade it. The overhead of DD pulses can outweigh the benefits of noise suppression.
  • Program-Specific DD: The effectiveness of DD is highly dependent on the specific quantum program, its structure, and the qubits involved. There is no one-size-fits-all DD strategy.
  • CNOTs Drive Idling Errors: Crosstalk from CNOT operations is a significant contributor to idling errors, making programs with many CNOTs particularly vulnerable.
  • Decoy Circuits as Proxies: Structurally similar decoy circuits, especially those preserving CNOT patterns, can effectively predict the impact of DD on the original program's fidelity, enabling classical simulation-based optimization.
  • Localized Search is Scalable: A localized search strategy for DD sequences is crucial for making the optimization process tractable for larger quantum systems.

Key Results

ADAPT demonstrated significant improvements in application fidelity across various benchmarks and IBM quantum hardware:

  • Average Improvement: ADAPT improved application fidelity by an average of 1.86x compared to running programs without any DD.
  • Maximum Improvement: In the best cases, ADAPT achieved up to a 5.73x improvement in fidelity compared to the no-DD baseline.
  • Comparison to All-DD: ADAPT also outperformed the "DD on all qubits" strategy, showing an average improvement of 1.2x.
  • Protocol Agnosticism: ADAPT's effectiveness was consistent across both the XY-4 and IBMQ-DD protocols, indicating its generalizability.
  • Correlation: The paper showed a strong correlation (Spearman's Correlation Coefficient of 0.78) between the fidelity of a program and its corresponding Clifford Decoy Circuit, validating the decoy circuit approach.

Limitations and Future Work

  • Decoy Circuit Limitations: While effective, decoy circuits may not perfectly capture all nuances of the original program's fidelity trends, especially in cases of high output variance. This can lead to ADAPT's performance being slightly below the "runtime best" policy in some scenarios.
  • Search Space Exploration: ADAPT's localized search, while efficient, explores a limited subset of the total DD sequence space. For larger systems, a more comprehensive search might yield further improvements, though at a higher computational cost.
  • Future Work: The authors suggest that ADAPT can be integrated with other compiler optimizations and error mitigation techniques. Further exploration of different DD pulse types and more advanced decoy circuit designs could also enhance performance. The framework's applicability to other quantum hardware platforms and different types of quantum algorithms is also a promising avenue for future research.

Summary updated

BibTeX
@inproceedings{das2021adapt,
  author = {Poulami Das and Poulami Das and Swamit S. Tannu and Siddharth Dangwal and Moinuddin K. Qureshi},
  title = {{ADAPT: Mitigating Idling Errors in Qubits via Adaptive Dynamical Decoupling}},
  booktitle = {Micro},
  year = {2021},
  doi = {10.1145/3466752.3480059},
}
Figure