Revisiting Noise-adaptive Transpilation in Quantum Computing: How Much Impact Does it Have?

TL;DR

The paper re-evaluates the prevailing belief that noise-aware transpilation must be performed daily to maintain high fidelity on modern superconducting quantum hardware. Through a large-scale, month-long empirical study on six IBM 127-qubit devices using 16 algorithms, it shows that noise-aware mapping often overuses a small subset of qubits, increasing fidelity variability, while simple random mapping achieves comparable mean fidelity with reduced fluctuations. Lower optimization levels (L1/L2) deliver similar fidelity to full optimization (L3) but with dramatically reduced compilation time, especially for larger circuits. Temporal analysis reveals minimal dependence of fidelity on time since calibration, and circuits compiled once maintain fidelity across multiple calibration cycles, enabling transpilation reuse. The work advocates lightweight, diversity-aware transpilation strategies to reduce classical overhead while preserving reliability, with implications for scalable, iterative quantum workloads such as variational algorithms.

Abstract

Transpilation, particularly noise-aware optimization, is widely regarded as essential for maximizing the performance of quantum circuits on superconducting quantum computers. The common wisdom is that each circuit should be transpiled using up-to-date noise calibration data to optimize fidelity. In this work, we revisit the necessity of frequent noise-adaptive transpilation, conducting an in-depth empirical study across five IBM 127-qubit quantum computers and 16 diverse quantum algorithms. Our findings reveal novel and interesting insights: (1) noise-aware transpilation leads to a heavy concentration of workloads on a small subset of qubits, which increases output error variability; (2) using random mapping can mitigate this effect while maintaining comparable average fidelity; and (3) circuits compiled once with calibration data can be reliably reused across multiple calibration cycles and time periods without significant loss in fidelity. These results suggest that the classical overhead associated with daily, per-circuit noise-aware transpilation may not be justified. We propose lightweight alternatives that reduce this overhead without sacrificing fidelity -- offering a path to more efficient and scalable quantum workflows.

Figures (8)

• Figure 1: Example 127-qubit IBM Nazca and Brisbane quantum computers. The circles represent the qubits (color-coded according to their respective T1 times), and the lines represent the qubit connections (color-coded as per ECR gate errors). Darker color indicates better values.
• Figure 2: We run three groups of experiments (all three groups are run daily), where each group is assembled into one job with $16\times{}3=48$ circuits. Here, 16 is the number of algorithms run, and we run each cluster of algorithms thrice within each group. All groups are submitted in parallel on each available computer daily. If a group's job does not finish within the calibration cycle for which it is optimized, the job is canceled or disregarded. Groups may end up running in any order. Each color in the visualization represents a specific transpiled circuit variation according to the mapping methodology or optimization level.
• Figure 3: Our analysis indicates that across all quantum computers, a small set of qubits gets used more frequently.
• Figure 4: Relationship between qubit usage and different KLD metrics (exponentiated) demonstrates the strong correlation.
• Figure 5: The average KLD difference for each algorithm when run with the optimal mapping vs. random mapping is low.