Five silicon spin qubits driven at once through one microwave line at ~99.9%

A RIKEN-led team has driven five silicon spin qubits at the same time through a single shared microwave line. It kept single-qubit gate fidelity at roughly the 99.9% level that is widely used as a fault-tolerance target. In their Nature Physics paper, published open access on 15 October 2026, the researchers say the main culprit behind fidelity loss in parallel operation is the AC Stark shift. They correct it with a calibration routine whose cost grows quadratically with qubit number, not exponentially. The result concerns single-qubit gates in a five-qubit array. It is not a full processor, but it removes one specific obstacle to sharing control hardware among many qubits.

01The problem: good qubits, bad neighbours

Silicon spin qubits store information in the spin of individual electrons confined in quantum dots. Their appeal is manufacturing. They are compatible with industrial semiconductor manufacturing. Single-qubit fidelities above 99% are routine, but the authors note in their preprint that extending high-fidelity control to simultaneous multi-qubit operation remains a major challenge.

Parallel operation matters because error correction needs it. According to the Quantum Computing Report's summary of the work, running error correction on a silicon processor requires operations on many qubits at the same time, and qubits that sit idle waiting their turn quickly dephase. Serial control is therefore not a way out. Giving each qubit its own dedicated control line is also unattractive. The RIKEN release points out that separate control systems for every qubit would make larger devices hard to manage, and shared lines are a more practical route.

The cost of sharing is that one microwave line carries tones for several qubits at once. Qubits in the array have different resonance frequencies, so each tone should address only its target. In practice, off-resonant drive power still perturbs the others.

02What the AC Stark shift does

The team's diagnosis is that microwave-drive-induced AC Stark shifts are the dominant source of fidelity degradation in this five-qubit silicon array. A microwave field changes a qubit's effective energy levels. This is not a population error. As RIKEN describes it, the field shifts the phase of the qubit's quantum state. When other qubits on the line are being driven, each qubit picks up phase offsets it was never meant to receive. These offsets rotate its state away from the intended gate.

The shift depends on which other qubits are driven and how hard. A naive fix would characterise every combination of active qubits, and the number of combinations grows exponentially with array size. The paper says its approach avoids the exponential overhead of full crosstalk characterization.

03Pairwise calibration

The protocol measures phase shifts between pairs of qubits and uses those measurements to compensate when more qubits are active. First author Yi-Hsien Wu says it avoids a separate calibration for every qubit combination. The preprint states the scaling directly: the calibration scales quadratically with qubit number and avoids exponential overhead. The Quantum Computing Report writes the pair count as N(N-1). For five qubits that is 20 ordered pairs.

The pairwise approach implies that the shifts combine approximately additively, so pair data can predict multi-qubit behaviour. That is my reading of the method, not a statement from the sources. The five-qubit result is evidence that it holds at this scale. Whether it holds for dozens of qubits sharing a line is untested.

Calibration was one of three ingredients. The release says the team also used an optimised device design and carefully tailored microwave pulses. The Quantum Computing Report lists Kaiser-window pulse shaping and a 125 ns single-qubit drive. The device is an isotopically purified ²⁸Si/SiGe quantum-dot array, and the collaboration includes National Taiwan University, ETH Zurich and TU Delft alongside RIKEN.

04The numbers

The five-qubit figure is described slightly differently across the sources. RIKEN's summary says accuracy stayed at approximately 99.9%. The Nature abstract says fidelities remain above the widely targeted fault-tolerant threshold of 99.9%. The preprint puts it at the threshold. I would treat 99.9% as the working number. The quoted values are primitive X and Y gate fidelities, so they are not whole-circuit performance. The authors say as much: overall fidelity also depends on two-qubit gates, state preparation and measurement, and idling. Two-qubit gates in silicon have generally lagged single-qubit gates, and the paper does not address them.

The drop from above 99.99% at three qubits to about 99.9% at five is the interesting part. The calibration removes much of the phase error but not all losses. The team attributes the remainder to drive-induced decoherence, a different mechanism. It comes from the higher microwave power needed to run several qubits at once and degrades coherence itself, rather than adding a phase offset. A phase offset can be measured and cancelled. Lost coherence cannot be calibrated back. Camenzind says better device design could mitigate it. The sources give no quantitative breakdown of how much each mechanism contributes at five qubits, nor a projection for larger arrays.

05Why it matters, and what is still open

The practical claim is about wiring and control complexity. Fewer impedance-controlled microwave lines per qubit mean less cabling and less hardware at cryogenic temperatures. That is a major concern for any chip-scale processor. The paper concludes that shared control lines can support the development of large-scale semiconductor quantum processors. RIKEN's Seigo Tarucha, who leads a silicon quantum computing project under Japan's JST Moonshot Goal 6, calls reliable parallel control an essential step toward a fault-tolerant machine.

Several things remain open:

• Scale. Five qubits is a small array. Pairwise calibration is polynomial, but the drive power, frequency crowding and decoherence penalty all rise with qubit count.
• Two-qubit and readout operations. These were not part of the demonstration. Whole-circuit fidelity will be set by the weakest operation.
• Drive-induced decoherence. This is now the acknowledged limit, and it must be reduced through device design.
• Stability. Calibration is only useful if the phase shifts stay put between recalibrations. The sources do not report drift.

06A related approach on a different platform

A separate result from the University of Osaka, published in npj Quantum Information, addresses speed against noise in GaAs spin qubits. The researchers added a third, electrically tunable quantum dot to a flopping-mode qubit. They reached drive frequencies above 100 MHz, and 271 ± 20 MHz under one tuning condition. With machine-learning feedback on slow frequency drift, they raised the estimated dephasing time from about 52 ns to 580 ns. They measured a 4 ns gate fidelity of 99.56%, rising to 99.72% with pulse optimisation. That is a different material system and a different bottleneck, and its fidelity is well below the RIKEN numbers. The common thread is post-fabrication tunability, which the Osaka group says could help correct device-to-device variation. Both papers treat control engineering, not just qubit quality, as the main barrier to scaling semiconductor qubits.

The RIKEN result shows that, in a five-qubit silicon array, much of the error from parallel driving was a calibratable phase effect rather than a fundamental limit. The remaining error comes from drive-induced decoherence.