Anchor: Reducing Temporal and Spatial Output Performance Variability on Quantum Computers

TL;DR

Quantum cloud services on NISQ devices suffer from strong temporal and spatial variability in output fidelity due to hardware noise. Anchor introduces a three-component framework—Circuit Maps Generator, TVD Predictor (Random Forest), and Linear Programmer—that distributes circuit shots across multiple maps and computers to minimize variability while controlling mean fidelity. The approach yields substantial reductions in variability (average ~73% across experiments) and competitive mean TVD, validated through simulations and real hardware, with open-source code and data. This variability-aware mapping strategy enhances reliability and predictability of quantum computations in cloud environments, enabling more robust practical applications.

Abstract

Quantum computing, which has the power to accelerate many computing applications, is currently a technology under development. As a result, the existing noisy intermediate-scale quantum (NISQ) computers suffer from different hardware noise effects, which cause errors in the output of quantum programs. These errors cause a high degree of variability in the performance (i.e., output fidelity) of quantum programs, which varies from one computer to another and from one day to another. Consequently, users are unable to get consistent results even when running the same program multiple times. Current solutions, while focusing on reducing the errors faced by quantum programs, do not address the variability challenge. To address this challenge, we propose Anchor, a first-of-its-kind technique that leverages linear programming to reduce the performance variability by 73% on average over the state-of-the-art implementation focused on error reduction.

Figures (16)

• Figure 1: (a) An example four-qubit quantum circuit. L0-L3 are the four "logical" qubits to which one-qubit U3 and two-qubit CX gates are applied from left to right. For CX gates, the red dot indicates the control qubit, and the yellow dot indicates the target qubit. At the end of the computation, measurement gates are applied, which then generate one output state (e.g., $\ket{1010}$). An output probability distribution over all the states is generated when the circuit is run and measured multiple times (e.g., 1024 times or "shots"). This circuit's ideal output distribution is shown when run on a noise-free quantum computer. (b) The layout of an example quantum computer with seven "physical" qubits with different error rates due to hardware noise. (c)-(d) Impact of the hardware noise effects when the circuit is run on two different circuit maps: Map A with lower error rates causes less output error (TVD), and Map B with higher error rates causes more TVD.
• Figure 2: When a user submits a job to the quantum computing cloud, the job is first added to a queue and dispatched to one of the computers in the cloud. Each computer has its own queue of jobs, from which jobs are scheduled and mapped to a region within the computer. After running the desired number of shots, the output probability distribution of the job is returned to the user.
• Figure 3: (a) Temporal and (b) spatial variability in the TVD of a quantum circuit when the 4-qubit VAR algorithm is run on the IBM quantum computers. We use the IBM QASM simulator to generate the ideal and noisy results. We use the same routing algorithms (routing determines how the circuit is laid out on the selected qubit map) for both map selection techniques. Refer to Sec. \ref{['sec:methodology']} for the complete methodology.
• Figure 4: Overview of the design of Anchor's quantum circuit mapper for reducing performance variability.
• Figure 5: Anchor's linear programmer divides shots across maps in a way that equalizes TVD across computers.