Generalized figure of merit for qubit readout

TL;DR

This paper introduces a generalized figure of merit based on Chernoff information for qubit readout, capturing the full information in analog outcomes and improving the assessment of readout performance in quantum computing.

Contribution

It proposes a new figure of merit using Chernoff information that accounts for analog readout outcomes and non-Gaussian noise, providing a unified framework for optimizing qubit readout.

Findings

Chernoff information fully characterizes asymptotic error rates.

Effective Gaussian noise models can describe non-Gaussian noise with the same Chernoff information.

The framework applies to small numbers of repetitions and non-QND imperfections.

Abstract

Many promising approaches to fault-tolerant quantum computation require repeated quantum nondemolition (QND) readout of binary observables such as quantum bits (qubits). A commonly used figure of merit for readout performance is the error rate for binary assignment in a single repetition. However, it is known that this figure of merit is insufficient. Indeed, real-world readout outcomes are typically analog instead of binary. Binary assignment therefore discards important information on the level of confidence in the analog outcomes. Here, a generalized figure of merit that fully captures the information contained in the analog readout outcomes is suggested. This figure of merit is the Chernoff information associated with the statistics of the analog readout outcomes in one repetition. Unlike the single-repetition error rate, the Chernoff information uniquely determines the asymptotic cumulative error rate for arbitrary readout noise. As a result, non-Gaussian readout noise common in experiments can be described by effective Gaussian noise with the same Chernoff information. Importantly, it is shown that such a universal description persists for the small number of repetitions and non-QND imperfections relevant to real experiments. Finally, the Chernoff information is used to rigorously quantify the amount of information discarded by analog-to-binary conversion. These results provide a unified framework for qubit readout and should facilitate optimization and engineering of near-term quantum devices across all platforms.