New AI Method Boosts Quantum Computer Reliability

Quantum computers hold immense promise for solving problems beyond classical reach, but today's devices, known as Noisy Intermediate-Scale Quantum (NISQ) systems, face a critical bottleneck: errors accumulate rapidly as qubits lose coherence over time. Existing s for mapping logical circuits to physical qubits often ignore this dynamic decay, focusing instead on static hardware topology. A new algorithm called TRAM (Transverse Relaxation Time-Aware Qubit Mapping) directly addresses this gap by incorporating real-time decoherence data into the compilation process, offering a practical path to more reliable quantum computations on current hardware.

TRAM achieves this by integrating three key components that work together to optimize qubit mapping. First, the Community Detection-assisted Quantum Transverse Relaxation Partitioning (CQTP) algorithm analyzes calibration data—including two-qubit gate errors, readout errors, and coherence times (T2)—to identify clusters of qubits that are both well-connected and noise-resilient. This creates stable partitions that serve as the foundation for placing logical circuits, transforming raw hardware metrics into actionable zones for computation. The algorithm uses a reward function that balances modularity gain, temporal similarity of T2 times, and error variance, with tunable weights (ω1 and ω2) to adapt to different circuit demands, such as long-entanglement or gate-intensive workloads.

Second, the Time-Weighted Heatmap-Based Initial Mapping (THIM) algorithm assigns logical qubits to physical ones by considering the temporal structure of the circuit. It weights two-qubit interactions based on their execution order, giving higher priority to gates that occur later in the circuit, as they are more exposed to decoherence. This approach minimizes the expected noise cost over time, using a global cost function that aggregates weighted routing costs across all logical qubit pairs. The routing cost itself combines gate-error accumulation along hardware paths with a decoherence penalty derived from T2 times, ensuring that both immediate and cumulative errors are factored into the mapping decisions.

Third, the Time-Adaptive Dynamic SWAP (T-SWAP) algorithm manages routing during circuit execution by dynamically scheduling SWAP operations. It evaluates candidate SWAPs using a heuristic cost function that accounts for physical proximity and cumulative decoherence, penalizing overused or fragile qubits to avoid error hotspots. This adaptive scheduling reduces redundant SWAP gates and minimizes congestion on error-prone paths, which helps lower both circuit depth and error accumulation. The algorithm operates by iteratively checking executable gates and inserting SWAPs only when necessary, with decay factors that track qubit usage to balance routing efficiency with coherence preservation.

Evaluated on Qiskit-based simulators with realistic noise models, TRAM demonstrates significant improvements over the widely used SABRE algorithm. Across standard benchmarks, it increases overall fidelity by an average of 3.59%, with gains reaching up to 6.21% in specific cases like the fredkin_n3 circuit on the ibm-perth simulator. It also reduces the total number of two-qubit gates by 11.49% and shortens circuit depth by 12.28%, indicating more efficient compilation with lower overhead. For example, on the ibm-guadalupe simulator, TRAM cut gate counts by up to 22.63% and depth by up to 22.00% for certain benchmarks, while on ibm-brooklyn, it achieved reductions of up to 14.69% in gates and 24.60% in depth.

These have immediate for the practical use of NISQ devices, which are constrained by limited qubit counts and high error rates. By making decoherence mitigation a primary optimization objective, TRAM enables higher-fidelity execution of quantum algorithms without requiring additional hardware resources. This could accelerate applications in areas like quantum chemistry and optimization, where even small improvements in reliability can make experiments more feasible. The algorithm's scalability and reliance on standard calibration data mean it can be applied across various quantum platforms, from small 7-qubit systems like ibm-perth to larger 65-qubit architectures like ibm-brooklyn, enhancing hardware utilization and supporting parallel task execution.

However, TRAM does have limitations, as noted in the paper. Its performance depends on the accuracy and timeliness of calibration data, which can vary over time and between devices. The tuning of parameters like ω1 and ω2 requires careful adjustment based on circuit characteristics, such as entanglement dwell time and gate density, which may not always be straightforward for arbitrary workloads. Additionally, while TRAM reduces SWAP overhead, it does not eliminate the fundamental need for routing in limited-connectivity hardware, and its improvements are bounded by the intrinsic noise levels of the qubits themselves. Future work could explore integrating real-time noise monitoring or extending the framework to other quantum computing platforms beyond superconducting devices.