Silicon spin qubits have long been the promising-but-perpetually-behind platform in quantum computing: excellent coherence, CMOS-compatible fabrication, but lacking the fault-tolerant logical operations that superconducting circuits and trapped ions already demonstrated years ago.
image from Gemini Imagen 4
Researchers at the Shenzhen International Quantum Academy demonstrated the first universal logical gate set in silicon donor qubits using five phosphorus nuclear spins encoding two logical qubits via the [[4,2,2]] error-detecting code, achieving 87-88% single-qubit and 88.6% CNOT logical gate fidelities. The system's strongly biased noise profile—dominated by phase-flip rather than bit-flip errors—plays to the code's strengths since the [[4,2,2]] code cannot correct arbitrary single-qubit errors but only detects them. They also reported preparing logical magic states above the distillation threshold for fault-tolerant T gates, a first for silicon-based systems.
Silicon spin qubits have long been the promising-but-perpetually-behind platform in quantum computing: excellent coherence, CMOS-compatible fabrication, but lacking the fault-tolerant logical operations that superconducting circuits and trapped ions already demonstrated years ago. A paper published March 23 in Nature Nanotechnology closes that gap, at least partially. A team at the Shenzhen International Quantum Academy (SZIQA) implemented a universal logical gate set using five phosphorus nuclear spins in isotopically enriched silicon-28, encoding two logical qubits via the [[4,2,2]] quantum error-detecting code — the first demonstration of its kind in silicon donor qubits, according to Nature Nanotechnology.
The work, led by Academician Dapeng Yu and researcher Yu He, uses scanning tunneling microscopy (STM) lithography to place phosphorus atoms with atomic precision on a silicon surface, then overgrows an epitaxial capping layer to embed the device. The approach yields physical qubit coherence times around 523 microseconds. When those same spins encode logical qubits, coherence drops to roughly 208 microseconds — a reduction of about 60 percent, because entangled logical states are inherently more fragile than any individual physical qubit. That's the fundamental trade-off error correction always imposes.
The gate results are mixed. Single-qubit logical gates hit 87 to 88 percent fidelity via quantum process tomography. The CNOT logical gate came in at 88.6 percent. But the simultaneous Hadamard on both logical qubits — a two-qubit operation that matters for practical circuits — landed at just 75.6 percent. Readout fidelity sits around 81 percent. None of these numbers would be acceptable in a classical computing context, but for a five-qubit logical processor built on nuclear spins, they're in the range the field has learned to treat as a floor rather than a ceiling.
What makes the architecture interesting isn't the absolute performance. It's the noise profile. The system exhibits strongly biased noise: phase-flip errors dominate massively over bit-flip errors. This matters because the [[4,2,2]] code can't correct arbitrary single-qubit errors — it's an error-detecting code, not a fully-correcting one. The biased noise profile plays to the code's strengths: the errors it can't catch are the ones that barely happen anyway. Theoretical work suggests this kind of noise asymmetry can actually raise the effective fault-tolerance threshold, potentially reducing the qubit overhead needed for useful computation.
The team also prepared logical magic states — the resource states required for fault-tolerant T gates — and claims their fidelity crosses the distillation threshold. As SZIQA reported, this is the first time logical magic states have been prepared in a silicon-based system above the fidelity required for distillation. If true, it removes one of the theoretical obstacles to building practical fault-tolerant circuits in silicon. The paper itself provides the quantitative basis: state preparation fidelity after parity projection reaches 96.5 percent for the logical zero state and 95.5 percent for a logical Bell state, with acceptance rates around 83 percent.
The proof-of-concept algorithm is where the enthusiasm deserves calibration. The team ran a variational quantum eigensolver (VQE) on two logical qubits to calculate the ground-state energy of a water molecule. The result came in at 22.7 millihartree above the theoretical value. Chemical accuracy — the threshold at which quantum chemistry calculations become practically useful — sits around 1.6 millihartree. The gap is roughly 14x. The press release called this "an error of only 20 mHa" and claimed "the potential to meet chemical accuracy requirements in the future." That's technically true in the same sense that a car traveling 10 mph has "the potential" to reach highway speeds.
The VQE result does demonstrate that logical encoding doesn't catastrophically destroy the ability to run algorithms — the logical qubits retain enough coherence to produce a plausible answer for a simple molecule. For a five-qubit demonstration of fault-tolerant architecture, that's not nothing. But it should not be read as a step toward practical quantum chemistry on silicon.
The paper follows a January 2026 result from the same group in Nature Electronics, which demonstrated quantum error detection in the same silicon donor platform. Two Nature family papers in three months is a credible pace for the SZIQA group, and the trajectory is real. The [[4,2,2]] code is increasingly appearing across multiple qubit modalities — neutral atoms and nitrogen-vacancy centers have also demonstrated it — which suggests the quantum computing community is converging on it as a building block for larger concatenated schemes.
What this paper does not do is demonstrate a prototype fault-tolerant computer. Five qubits running one molecule is not a scalable architecture. The authors themselves note that higher-weight errors from cross-talk currently exceed the [[4,2,2]] code's error detection capability — the code can't catch the dominant error source in the current device. Scaling to the thousands of physical qubits required for useful fault-tolerant computation will demand concatenated codes layered on top of this architecture, each level adding overhead.
The honest summary: silicon spin qubits have now crossed the threshold from "physical qubits only" to "logical operations possible" in a single paper. The noise bias result and the magic state threshold crossing are the genuinely interesting developments for people building quantum error correction systems. The VQE result is a demonstration of concept, not a hint of commercial relevance. Whether STM lithography can deliver the fabrication precision required at scale — millions of donors with atomic precision across a full wafer — remains the central engineering question this approach has not answered.
The paper demonstrates that the physics works. The engineering question remains open.
Story entered the newsroom
Research completed — 7 sources registered. First universal logical gate set in silicon donor qubits by SZIQA/Dapeng Yu group. Five phosphorus nuclear spins in 28Si with [[4,2,2]] error-detectin
Draft (565 words)
Approved for publication
Headline selected: Silicon Qubits Just Did Something Superconducting Did Years Ago
Published (900 words)
Get the best frontier systems analysis delivered weekly. No spam, no fluff.
Quantum Computing · 19h 0m ago · 2 min read
Quantum Computing · 19h 57m ago · 4 min read
