Jiuzhang 4.0: 3,050 Photons, a 10^54 Gap Over Supercomputing
On May 13, a team from the University of Science and Technology of China led by Pan Jianwei, Lu Chaoyang, Zhang Qiang, and Liu Nailai, together with multiple partner institutions, published the "Jiuzhang 4.0" photonic quantum computing prototype in *Nature*. For the first time, it manipulates and detects the quantum states of 3,050 photons. On Gaussian boson sampling, it outpaces the world's fastest supercomputer by a factor of 10^54—generating one sample takes just 25 microseconds, where a supercomputer would need more than 10^42 years.
This is an order-of-magnitude leap for the photonic quantum computing route, and it cements China's world-leading position in optical quantum computing.
What 3,050 Photons Actually Means
Think of Jiuzhang 4.0 as a three-dimensional maze with 8,176 exits, where photons perform extremely complex interference walks.
- 1,024 quantum squeezed states: the "high-energy fuel" of the running system, the foundation for building complex quantum entanglement.
- 8,176 modes: the total number of paths/dimensions photons can traverse.
- 3,050 photons: roughly equivalent to 3,000+ qubits. The previous Jiuzhang 3.0 manipulated only 255 photons—this is a more than 10x improvement, exponentially expanding the computable state space.
- A new benchmark for quantum advantage. Gaussian boson sampling is the standard problem for demonstrating quantum computational advantage. Jiuzhang 4.0's speedup over El Capitan (the world's fastest supercomputer) reaches the 10^54 order—one of the strongest advantage claims internationally to date.
- The unique position of the photonic route. Superconducting circuits, ion traps, neutral atoms, and photonics are the four main approaches. Photonics requires no cryogenics and naturally suits specific sampling problems. The Jiuzhang series (2020 → Jiuzhang 2.0 in 2022 → Jiuzhang 3.0 in 2023 → Jiuzhang 4.0 in 2026) has shipped four generations in six years—a steady iteration cadence.
- A stepping stone toward fault-tolerant hardware. Jiuzhang 4.0 can do more than sampling: it can generate bosonic error-correcting codes and large-scale entangled cluster states needed for fault-tolerant quantum computing. The team's stated next goals: a trillion-quantum-mode 3D cluster state, and eventually fault-tolerant photonic quantum computing hardware.
- Practical applications. In the short term, Gaussian boson sampling is useful for image recognition and graph-theoretic computation; in the long term, it is a component of universal fault-tolerant quantum machines.
- Jiuzhang 4.0 is a special-purpose analog machine, not a universal computer. A universal quantum computer needs millions of error-corrected qubits—still a long road. But the dual lead in scale and low loss leaves room for the fault-tolerant path.
- Compare with the superconducting route (IBM, Google): photonics pursues "sampling advantage + bosonic codes," while superconducting pursues "logical qubits + surface codes." Both are approaching fault tolerance, but with entirely different engineering constraints.
- For China's quantum industry: the Jiuzhang series keeps setting records, and combined with earlier results like 420 km cold-atom entanglement and QuantumCTek progress, photonics is delivering from prototype machines up through the network layer.
What was the bottleneck? The biggest obstacle to scaling photonic quantum computing is photon loss: the larger and more complex the optical network, the more easily photons get lost, crippling computing power.
Jiuzhang 4.0's breakthrough is a "programmable spatiotemporal hybrid encoding" architecture: photons interfere simultaneously in both time and space, raising network connectivity while keeping the physical hardware scale contained. The team also built a high-efficiency optical parametric oscillator light source, bringing total system efficiency to 51%. The result: sampling in a 102,461-dimensional Hilbert space.