The rare but large failure events in the Google Willow processor

by Erik Aurell (KTH)

Thursday 19 Dec 2024, 13:00 → 15:00 Europe/Stockholm

AlbaNova C4:3059 - Café Planck (AlbaNova Main Building)

Google Quantum AI has recently demonstrated quantum error correction with a new 105-qubit chip called "Willow" [1]. This demonstration is based on a robust observation that logical qubits have lower error rates the more physical qubits they are built on, and has been made possible by improved properties of these same physical qubits.

In this talk I will draw attention to a curious collective error mode described by Google, which appears about once per hour in the runs, and which sets a new, at least for now, error floor of 10^-10 for logical qubits, of any size. The (publicly) available data can be seen in Fig 3b and Fig S16 (in Supplementary Information) of the paper; and is mentioned in the second to last sentence in the paper abstract. But it does not appear to have been much high-lighted elsewhere. For instance, it is not mentioned in the post on Scott Aaronson's blog devoted to the paper [2] or in comments by Sabine Hossenfelder [3]. Possibly the new error mode observed by Google could be a so far not understood side-effects of chip fabrication, possibly something more fundamental, and therefore more challenging to address.

[1] Google Quantum AI and Collaborators. Quantum error correction below the surface code threshold. Nature (2024). https://doi.org/10.1038/s41586-024-08449-y

[2] https://scottaaronson.blog/?p=8525 "The Google Willow thing"

[3] https://www.youtube.com/shorts/Q1gQHIHPq5Q