In August 2026, a team at the University of Science and Technology of China (USTC) demonstrated below-threshold quantum error correction on the superconducting quantum processor Zuchongzhi 3.2, using a distance-7 surface code: logical error rates decreased significantly as code distance increased. To understand why this matters, one must start with the counterintuitive phenomenon known in quantum computing as 'the more you correct, the worse it gets.'
1. What 'More Correction, More Errors' Means
Quantum computing is like an extremely powerful but delicate race car: its potential is enormous, yet the slightest temperature fluctuation or electromagnetic interference can throw qubits off course. Quantum error correction (QEC) is an auto-stabilizing system for this car — but the correction apparatus itself has weight and can vibrate. If the underlying hardware is not stable enough, more frequent correction leads to more loss of control.
QEC has a breakeven line: correction only truly works when the corrected logical qubit is more reliable and persists longer than the raw physical qubits. Once past this threshold, increasing code distance and resources actually lowers logical error rates — 'the more you correct, the more correct it becomes.' This is the prerequisite for scaling quantum computing from theoretical feasibility to engineering reality.
Few teams worldwide have achieved below-threshold operation. Google's 2024 Willow processor provided the first hardware-level demonstration with surface-code logical qubits at distances 3, 5, and 7, where error rates fell exponentially with distance. What Zuchongzhi 3.2 achieved is independently crossing the same threshold on a homegrown processor platform, following an all-microwave route with greater scaling potential.
2. The Two Key Techniques on Zuchongzhi 3.2
The core of turning 'more correction, more errors' into 'more correction, more correctness' is making correction outpace noise propagation. The researchers introduced two techniques:
Technique 1: Leakage recovery. Ordinary qubits occupy only the '0' and '1' levels, but real devices have extra levels, and qubits can accidentally escape into states outside the computational subspace. Once leaked, conventional correction can neither see nor cleanly fix the problem, and the leakage can drag neighboring qubits down in subsequent operations. The leakage recovery design uses specific microwave drives to guide leaked qubits back into the computational space, removing excess energy and preventing leakage from accumulating over multiple correction rounds.
Technique 2: Fast reset. Ancilla qubits responsible for error detection undergo a fast, unconditional reset at the end of every round, ensuring they start each round 'clean' and do not carry residual errors forward.
These two techniques, combined with the independently developed all-microwave quantum state leakage suppression architecture, form the foundation for stable operation. The elegance lies in the suppression process being seamlessly embedded in the normal QEC workflow, with no additional hardware or time overhead.
As Professor Xue Peng (Beijing Computational Science Research Center) put it: quantum error correction used to be like 'repairing while leaking' — the more you fixed, the messier it got. This work effectively plugged the most stubborn leaks with recovery plus reset, producing the effect of 'the larger the scale, the more reliable.'
3. What an Independent Route Means
China had already established leadership in superconducting quantum computational advantage with Zuchongzhi 3.0 (March 2025 cover paper in *Physical Review Letters*, 105-qubit random circuit sampling 15 orders of magnitude faster than classical supercomputers). Zuchongzhi 3.2 now pushes that position further: error-correction capability competing on par with the strongest international efforts.
Notably, the routes are independent:
- Google's Willow uses traditional 'frequency tuning + multiple control lines' coupling.
- The Chinese approach uses 'all-microwave quantum state leakage suppression' — all control is done with microwave pulses.
- Cryogenics: cooling power demand doubles roughly every 100 qubits added
- Control cabling: microwave crosstalk causes spurious excitations; calibration, once 2–3 days of manual work, is now compressed to 4 hours with AI agents
- Packaging yield: currently around 80%
- Algorithm–hardware co-design: the 'valuable problem set' for the early fault-tolerant era is undefined
- Superconducting: error-correction threshold crossed; independent all-microwave route established
- Photonic: on-chip MBQC engineering (4-photon 16-qubit GHZ state + Grover at 0.987) + boson sampling scale breakthrough (Jiuzhang 4.0: 3,050 photons + 8,176 optical modes)
- Quantum networks: 420-km quantum memory entanglement + Xinghan-2 14.5-km quantum repeater
- Trapped ions: Origin's Wukong cloud platform — 520 million visits from 192 countries, 1 million tasks
- Quantum computing bids farewell to 'more correction, more errors' — how long until universal compute? - Toutiao
- Science watch: What makes the Zuchongzhi 3.2 superconducting processor special? - Toutiao
- Zuchongzhi 3.0 selected among China's Top 10 Science News of 2025 - quantum.ustc.edu.cn
- Major breakthrough in quantum error correction - qikan.com.cn
- When computers outrun time itself - dy.163.com
- On the eve of the breakthrough: from 'quantum' to 'mass production' - xh.xhby.net
Both routes have scaling potential. Microwave control's advantages: fewer cables, easier arraying, and no complex DC bias circuitry — wiring being one of the key bottlenecks for engineering million-qubit superconducting processors.
4. Several Key Judgments
Judgment 1: Universal quantum computers within 5–10 years. Academician Guo Guangcan (CAS): 'Within 5 to 10 years, universal quantum computers will emerge, and their size will be far smaller than today's supercomputers,' with reason to believe it could be sooner.
Judgment 2: Physical qubit count ≠ usable compute. Physical qubits are rough bricks; logical qubits are the standard building material. At equal correction distance, one logical qubit may require up to 100,000 physical qubits under the surface code. The community (including Caltech's Preskill group, recently proposing the 'mitten code') is working to compress this ratio, but approaches remain at the theoretical validation stage.
Judgment 3: Fault-tolerant quantum computing requires four lines running simultaneously: compute × cryogenics × calibration × ecosystem. Even with error correction solved, engineering hurdles remain:
All this suggests: now is not the moment to pick a single winner. Superconducting, photonic, trapped-ion, and neutral-atom platforms each have their strengths — the real race has just begun.
5. Concrete Coordinates for China's Quantum Industry
Viewed together — Zuchongzhi 3.2, Origin Quantum's PSE-CZ, Sizen's MBQC, 420-km quantum memory entanglement, and Jiuzhang 4.0's manipulation of 3,050 photonic states — these mark China's quantum industry coordinates as of August 2026:
'Full domestic stack + multiple parallel routes + engineering deployment' is the true posture of China's quantum technology today.
---
Sources