Here’s a side-by-side that sticks to what’s actually published.
Zuchongzhi-3.0 (USTC) vs Google Willow
| Category | Zuchongzhi-3.0 | Google Willow |
|---|---|---|
| Team / Pub date | USTC; PRL 134:090601 (Mar 3, 2025) | Google Quantum AI; Nature (“below-threshold” QEC) & Google blog (Dec 9, 2024) |
| Qubits & layout | 105 transmons in a 15×7 grid; 182 couplers | 105 transmons; average connectivity 3.47 (≈4-way typical) |
| Coherence (T1) | ~72 µs (mean) | 68 µs (QEC chip mean); 98 µs (RCS chip mean) |
| 1-qubit gate error | 0.10% (99.90% fidelity, simultaneous) | 0.035% (QEC chip); 0.036% (RCS chip) |
| 2-qubit gate error | 0.38% (99.62% fidelity, simultaneous) | 0.33% CZ (QEC chip); 0.14% iSWAP-like (RCS chip) |
| Readout error | 0.87% (99.13% fidelity) | 0.77% (QEC chip); 0.67% (RCS chip) |
| RCS benchmark run | 83 qubits, 32 cycles; samples in O(10² s) (“few hundred seconds”; Physics Mag. describes a ~100 s instance) | 103 qubits, depth 40, XEB ≈ 0.1%; ~5 min |
| Claimed classical time | “several billion years” (Frontier-class est.) | 10²⁵ years (supercomputer est.) |
| Error correction | Working toward surface code (distance-7 in development; not yet a peer-reviewed below-threshold result on this chip) | Below-threshold surface-code logical qubits (distance-7 & distance-5) with exponential suppression factor Λ≈2.14 per +2 code distance; beyond break-even |
| Integration notes | Two sapphire chips flip-chipped (indium bump) | Spec sheet focuses on performance; fabrication details not emphasized publicly |
Sources: Zuchongzhi-3.0 PRL & USTC/CAS notes; APS Physics Viewpoint; Willow blog/spec sheet & Nature paper. link.aps.orgen.ustc.edu.cnenglish.cas.cn blog.googlequantumai.google www.nature.comarxiv.orgresearch.google
What these numbers actually say
-
Both chips are 105-qubit superconducting transmon processors at state-of-the-art quality. Zuchongzhi-3.0’s layout and coupler count are explicit (15×7; 182 couplers). Willow publishes rich system metrics (connectivity, T1, simultaneous gate/readout errors). link.aps.org quantumai.google
-
Raw gate quality: Willow reports lower two-qubit and readout errors on its RCS-optimized chip than USTC reports for Zuchongzhi-3.0’s simultaneous operations; single-qubit errors are also lower on Willow. (See spec sheet rows). quantumai.google
-
Coherence: Means are comparable—~70–100 µs across the two platforms depending on the chip/task. Zuchongzhi-3.0 reports ~72 µs; Willow reports 68 µs (QEC) and 98 µs (RCS). en.ustc.edu.cnquantumai.google
-
Random-Circuit Sampling (RCS):
-
USTC: 83-qubit/32-cycle RCS in ~10² s; APS Physics says classical simulation would take several billion years—that’s their ~10¹⁵× headline. link.aps.org
-
Google: 103-qubit/depth-40 RCS in <5 min, claiming 10²⁵ years classically. quantumai.google
Important: these are different circuits (qubits, depth, compilation), so the “who’s faster” number isn’t apples-to-apples. APS explicitly cautions that direct comparison is not straightforward. link.aps.org -
Error correction (the road to useful QC): Willow has peer-reviewed evidence of below-threshold surface-code operation (distance-7 logical memory with exponential suppression Λ≈2.14 per +2 code distance; logical lifetime beyond best physical qubit). USTC signals they are building toward distance-7 on Zuchongzhi-3.0 but has not yet published a below-threshold result for this processor. www.nature.comarxiv.orgresearch.googleenglish.cas.cn
Bottom line by “who leads where”
- RCS “quantum advantage” showpiece: Both show overwhelming advantage vs classical on their chosen instance. The absolute “10¹⁵× vs 10²⁵ years” numbers can’t be used to crown a winner because the instances differ. link.aps.orgquantumai.google
- Hardware quality (errors, T1, readout): Slight edge to Willow on reported simultaneous 2-qubit and readout errors; coherence is comparable. quantumai.google
- Fault-tolerance trajectory: Clear edge to Willow today due to the below-threshold surface-code demonstration on this generation of hardware. www.nature.com
- Engineering approach: Both use superconducting transmons; Zuchongzhi-3.0 explicitly uses flip-chip two-sapphire integration; Willow’s public docs emphasize performance/QEC metrics more than materials. link.aps.orgquantumai.google If you want, I can turn this into a 2–3 sentence, source-backed Reddit reply.