Calibration-Aware Transpilation for Variational Quantum Optimization

TL;DR

This paper introduces a calibration-aware transpilation method for variational quantum algorithms on NISQ devices, optimizing for changing error rates and reducing computational overhead during execution.

Contribution

It proposes a three-step transpilation process tailored for variational algorithms that accounts for dynamic calibration data, improving efficiency and robustness.

Findings

Low latency transpilation demonstrated on IBM quantum hardware

Robustness of results under changing error rates

Efficient incremental transpilation process for variational algorithms

Abstract

Today's Noisy Intermediate-Scale Quantum (NISQ) computers support only limited sets of available quantum gates and restricted connectivity. Therefore, quantum algorithms must be transpiled in order to become executable on a given NISQ computer; transpilation is a complex and computationally heavy process. Moreover, NISQ computers are affected by noise that changes over time, and periodic calibration provides relevant error rates that should be considered during transpilation. Variational algorithms, which form one main class of computations on NISQ platforms, produce a number of similar yet not identical quantum ``ansatz'' circuits. In this work, we present a transpilation methodology optimized for variational algorithms under potentially changing error rates. We divide transpilation into three steps: (1) noise-unaware and computationally heavy pre-transpilation; (2) fast noise-aware matching; and (3) fast decomposition followed by heuristic optimization. For a complete run of a variational algorithm under constant error rates, only step (3) needs to be executed for each new ansatz circuit. Step (2) is required only if the error rates reported by calibration have changed significantly since the beginning of the computation. The most expensive Step (1) is executed only once for the whole run. This distribution is helpful for incremental, calibration-aware transpilation when the variational algorithm adapts its own execution to changing error rates. Experimental results on IBM's quantum computer show the low latency and robust results obtained by calibration-aware transpilation.