Leveraging machine learning and digital twins to characterize and improve gate fidelities in quantum computers

Ana Gramajo¹, Alastair Marshall¹, Yousof Mardoukhi¹, Deepak Khurana¹, Satya Bade¹, Marc Bernot¹, Roman Razilov¹, Anurag Saha Roy¹, Shai Machnes^1*

¹Physics, Qruise GmbH, Saarbrucken, Germany

* Presenter:Shai Machnes, email:shai.machnes@qruise.com

Developing practical quantum computing (QC) architectures poses significant challenges, particularly in scaling high-fidelity entangling gates. In parallel, machine learning (ML) is taking an increasingly significant role in scientific work. In this talk we demonstrate how novel ML techniques can help accelerate development of QCs.

Specifically, in-depth characterization of quantum devices is crucial not just for the optimal operation of high fidelity gates but more importantly, to identify the true system parameters, which allow creation a digital twin of the QPU and its control stack. Such an accurate digital twin of the system is critical for gaining insight into the quantum device - what works, what doesn't, and what physical phenomena are limiting gate fidelities.

We use novel ML tools combining differentiable physics simulations with data from a broad set of characterization experiments of a superconducting QPU, to build a model with a high predictive power, that accurately predicts the outcome of experiments outside the training dataset. We show how this predictive model is then used to identify limiting factors of the QC's performance, and how we can addressed some of those factors via appropriate control techniques.

More generally, we believe this demonstrates how the use of ML for creation of accurate digital twins combining physics equations and experimental data, can significantly accelerate scientific and engineering endeavors.

Keywords: quantum computing, machine learning, differentiable simulations, digital twins, gate fidelities