Neural Network Based Qubit Environment Characterization

TL;DR

This paper introduces a neural network approach to infer the microscopic environment causing $1/f$ noise in superconducting qubits from a single coherence measurement, enhancing understanding without additional spectroscopy.

Contribution

It presents a novel machine learning method that reconstructs impurity parameters and distinguishes environment models from qubit coherence data.

Findings

Successfully reconstructs impurity parameters

Differentiates between environment models

Enables better understanding of $1/f$ noise

Abstract

The exact microscopic structure of the environments that produces $1/f$ noise in superconducting qubits remains largely unknown, hindering our ability to have robust simulations and harness the noise. In this paper we show how it is possible to infer information about such an environment based on a single measurement of the qubit coherence, circumventing any need for separate spectroscopy experiments. Similarly to other spectroscopic techniques, the qubit is used as a probe which interacts with its environment. The complexity of the relationship between the observed qubit dynamics and the impurities in the environment makes this problem ideal for machine learning methods - more specifically neural networks. With our algorithm we are able to reconstruct the parameters of the most prominent impurities in the environment, as well as differentiate between different environment models, paving the way towards a better understanding of $1/f$ noise in superconducting circuits.