HomeFull speed ahead. Keep track of advancements in every frontier

TheTechno Optimist.

Willow and error-correction roadmapGoogle Quantum AI—
Willow and error-correction roadmap
Courtesy of Google · Press kit, editorial use (opens blog.google)

Willow and error-correction roadmap

by Google Quantum AIUS

ProtoStage 2 of 5

Working on Milestone 3: a long-lived logical qubit with about 1 error per million operations.

Updated 22 Oct 2025Checked 25 Sep0 updates this week

Milestones

Next · Milestone 3: long-lived logical qubit (~1 error in a million)
  1. Milestone 1: beyond-classical computation (Sycamore)23 Oct 2019Complete.
  2. Milestone 2: error-correction prototype22 Feb 2023Complete.
  3. Willow: below-threshold surface code9 Dec 2024Complete.
  4. Quantum Echoes verifiable advantage22 Oct 2025Complete.
  5. Milestone 3: long-lived logical qubit (~1 error in a million)NowCurrent milestone.
  6. Milestone 4: logical gateNot yet reached.
  7. Milestone 5: engineering scale-up (100 logical qubits)Not yet reached.
  8. Milestone 6: large error-corrected quantum computerTarget 2029Not yet reached.

Most important updates

  • 22 Oct 2025
  • May 2025
  • 9 Dec 2024
  • 22 Feb 2023
  • 23 Oct 2019

Upcoming

  1. 2029Milestone 6: large error-corrected quantum computer (next)

Current obstacles

  • Rare error burstsLong runs showed rare error bursts that put a floor on logical errors. Milestone 3 needs them suppressed.
  • Wiring and cryogenic scaleGoing from about 100 to about a million qubits needs new control electronics, wiring and cooling.

Physics limits

  • Below threshold, but only justError correction helps only when physical errors are below about 1%. Willow cut logical errors about 2× per code size step; reaching one error in a million needs roughly a thousand physical qubits per logical qubit.
  • Cosmic rays hit many qubits at onceHigh-energy particles striking the chip release bursts of vibrations that disrupt many qubits together. Error correction assumes errors are independent, so such bursts must be blocked by shielding or chip design.
  • A million qubits can't each have cablesEach qubit today needs several coaxial lines from room temperature. A million-qubit machine can't fit millions of cables in a fridge, so control electronics must move into the cold, where cooling power is tiny.

How it works

4 parts
Sycamore, Willow's 2019 predecessor, in the Deutsches Museum: a grid of superconducting qubits on a chip bonded to a wiring layer
Sycamore, Willow's 2019 predecessor, in the Deutsches Museum: a grid of superconducting qubits on a chip bonded to a wiring layerPhoto: Coldupnorth · CC BY-SA 4.0 (opens commons.wikimedia.org)
Chip

Superconducting grid

105 qubits sit on a square grid in a fridge at about 10 millikelvin, with couplers between neighbours.

Checks

Repeated parity checks

About every microsecond, helper qubits check groups of 4 data qubits. Flips in these checks show where errors happened.

Decode

Real-time decoding

Normal computers, including machine-learning models, read the check results and track errors as the program runs.

Threshold

Bigger is better below threshold

Once qubits are good enough, each bigger code multiplies protection. Willow showed this for code sizes 3, 5 and 7.

Spec sheet

vs Sycamore
SpecWillow and error-correction roadmapSycamoreChange
Physical qubits105R (reported)53-72 (Sycamore)—no comparable change
Mean T1 coherence~68 µsR (reported)~20 µs▲ 48 µs
Two-qubit gate error (mean)0.33%R (reported)——no comparable change
Single-qubit gate error0.035%R (reported)——no comparable change
Measurement error0.77%R (reported)——no comparable change
Largest surface codeDistance 7R (reported)Distance 5▲ 2
Logical error per cycle (d=7)0.143%R (reported)——no comparable change
Error suppression per distance step2.14xR (reported)——no comparable change

R reported by the company

Papers & demos

4 items
  1. Oct 2025paper
    Quantum Echoes: out-of-time-order correlator measurement on Willow (opens nature.com)A verifiable beyond-classical computation with a proposed route to molecular structure measurements.
  2. Dec 2024paper
    Quantum error correction below the surface code threshold (opens nature.com)Willow's largest logical qubit cut errors by over 2x at each step up in code size.
  3. Feb 2023paper
    Suppressing quantum errors by scaling a surface code logical qubit (opens nature.com)A bigger code slightly beat a smaller one, the first sign that errors drop with scale.
  4. Oct 2019paper
    Quantum supremacy using a programmable superconducting processor (opens nature.com)53-qubit Sycamore sampled random circuits in 200 seconds, a task estimated to take classical supercomputers far longer.

Update log

5 updates

Wed 22 Oct 2025

  • Major: PaperResearch

May 2025

  • Minor: PaperResearch

Mon 9 Dec 2024

  • Major: BlogHardware

Wed 22 Feb 2023

  • Major: PaperResearch

Wed 23 Oct 2019

  • Major: PaperResearch

About Google Quantum AI

The team behind Willow and error-correction roadmap

Google Quantum AI

Willow chip, surface-code error correction

Alphabet's quantum lab in Santa Barbara, California. It showed beyond-classical computing in 2019. Its Willow chip (2024) made the first logical qubit whose errors fall as the code grows. Michel Devoret and ex-lead John Martinis shared the 2025 Physics Nobel.

  • Willow (105 qubits, Dec 2024): logical errors halved with each step up in code size.
  • Quantum Echoes (Oct 2025) ran a physics measurement about 13,000x faster than the Frontier supercomputer.
  • On a 6-step roadmap. Steps 1 and 2 are done. Now working on a long-lived logical qubit.

quantumai.google (opens quantumai.google)Company profile →

Founded
201313 yrs
Headquarters
United States
Status
Subsidiaryof Alphabet
Valuation
Alphabet-owned
Works in
QuantumSuperconducting
Coverage
1 program · 6 updateslatest 22 Oct 2025checked 25 Sep
Partners
NASA Ames (opens nasa.gov)QuEraUC Santa Barbara (opens ucsb.edu)
People
Hartmut Neven (opens en.wikipedia.org)Founder and leadMichel Devoret (opens en.wikipedia.org)Chief Scientist, hardware; 2025 Nobel laureate
Primer