An international research team including RIKEN announced on October 15, 2026 that it had operated five silicon spin qubits simultaneously through a single shared high-frequency control line while maintaining gate fidelity of about 99.9% (announcement). Gate fidelity indicates how closely an operation on a qubit matches the ideal operation. When three qubits were driven at once, each achieved fidelity above 99.99%. The team identified the main cause of the accuracy loss when multiple qubits are driven simultaneously, and developed a method that corrects errors measured for each pair of qubits. The result could reduce the wiring and calibration burden of integrating many qubits. However, the team also found that driving five qubits at once introduces another type of error that grows with drive power, revealing challenges for further scaling.
The research involved Yi-Sheng Wu, Leon Camenzind, and Seigo Tarucha of RIKEN, among others. The paper, "Simultaneous High-Fidelity Single-Qubit Gates in a Spin Qubit Array," was published in Nature Physics on the same day. A preprint was first made public in July 2025, so this announcement marks the peer-reviewed publication.
How One Control Line Operates Five Qubits
The team used a semiconductor device in which five quantum dots are arranged in a layered structure of isotopically purified silicon and silicon-germanium. A quantum dot confines an electron to an extremely small region, and the electron's spin (a quantum-mechanical property loosely analogous to rotation) is used to store and process quantum information. Silicon spin qubits are highly compatible with existing semiconductor manufacturing and are considered a leading candidate for building large-scale quantum processors. But if control wiring is added in step with the number of qubits, layout and connections become complicated, which hinders integration.
To address this, the team adopted a configuration in which all five qubits are operated through one shared high-frequency control line. The device incorporates micromagnets and is designed so that the magnetic field strength varies with each qubit's position. As a result, neighboring qubits have resonance frequencies that differ by roughly 150 MHz, allowing each qubit to be addressed selectively with microwaves at a different frequency. When the microwave electric field moves an electron slightly, the magnetic field the electron experiences also changes, which makes it possible to control the spin direction. By superimposing signals at the frequency corresponding to each qubit, multiple qubits can be driven at the same time through a common electrode.
Note that what is shared here is the high-frequency line used for spin control. The electrodes that form the quantum dots and the wiring used for readout were not all consolidated into a single line. The specific device configuration is shown in Figure 1 of the paper.
Although separating resonance frequencies allows selective control of individual qubits, the influence of microwaves aimed at other qubits cannot be completely eliminated. The problem is the "AC Stark effect," in which electromagnetic waves off resonance still shift the energy-level difference of a qubit. When the energy difference changes, the relative phase of the superposition state shifts, introducing errors into subsequent quantum operations. In other words, even if a non-target qubit does not undergo a large rotation, its computational accuracy may degrade because of the microwaves' influence.
On the other hand, operating qubits one at a time to avoid such interference is not necessarily a solution. In research on a silicon quantum processor published by Mills et al. in 2022, errors that arise from having one qubit wait while another is operated were already an issue. Because qubits are exposed to environmental noise even while idle, large-scale quantum computation needs to operate multiple qubits in parallel in order to reduce waiting time.
Pairwise Phase Correction Reduces the Calibration Burden
To suppress errors from simultaneous driving, the team combined control-pulse shaping with phase correction. First, it shaped the microwave pulse waveform using a mathematical window function known as a Kaiser window, suppressing direct interference from unintended frequency components. Then it measured how much phase shift accumulates on another qubit when a given qubit is driven, and developed a method that adjusts the phase of subsequent control signals to cancel the error. Because simultaneous driving also introduces errors in the spin rotation angle, the amplitude of the microwave pulses was adjusted as well.
The effect of this correction was confirmed in an experiment that drove two qubits simultaneously. Before correction, the gate fidelity of one qubit was only about 99.9%, but after correcting phase and amplitude, both qubits exceeded 99.99%. The team also measured separately the phase shifts that each of two drive signals imparted on another qubit, and confirmed that the sum of these values agreed well with the measured value when both signals were applied at once. Figures 3 and 4 of the paper show the effect of the correction and the validity of summing the phase shifts.
Importantly, even when three or more qubits are driven simultaneously, it is not necessary to measure the errors for every combination. By measuring the phase shift between each pair of qubits individually and adding the values, the phase error when multiple signals are applied at once can be corrected. This mitigates the problem of calibration work becoming rapidly more complex as the number of qubits grows.
The difference becomes more pronounced as the number of qubits increases. According to the paper's conclusion, the number of combinations in which two or more of N qubits are chosen and driven simultaneously is 2^N−N−1. By contrast, when pairwise phase shifts are measured while distinguishing the influencing qubit from the influenced one, the number of calibration sets required is N×(N−1). Comparing values obtained from the paper's formulas gives the following:
| Number of qubits | Combinations driven simultaneously (2 or more) | Pairwise phase calibrations |
|---|---|---|
| 5 | 26 | 20 |
| 10 | 1,013 | 90 |
For five qubits, there are 26 simultaneous-drive combinations versus 20 phase calibrations, not a large difference. At ten qubits, however, simultaneous-drive combinations rise to 1,013, while pairwise phase calibration needs only 90 sets. The number of simultaneous-drive combinations grows exponentially with the number of qubits, whereas the number of pairwise calibrations grows only quadratically. This difference matters for extending the control scheme to large quantum processors.
The table compares the counts of different kinds of measurement targets, however, and does not mean calibration time shrinks by the same ratio. In practice, calibrations beyond phase correction are also needed, such as control-pulse amplitude adjustment and qubit-state readout. The method is effective under conditions where the exchange coupling between qubits is effectively turned off and microwave-induced phase shifts can be treated as a sum of individual contributions.
Over 99.99% With Three Qubits Driven Together, About 99.9% With Five
The gate fidelities measured in this study represent the accuracy of basic single-qubit gates that rotate each qubit by 90 degrees about the X or Y axis. They do not indicate the fidelity of two-qubit gates that entangle qubits, nor the success rate of an entire computation using multiple qubits. The team evaluated accuracy using randomized benchmarking, which estimates gate accuracy from the decay in the probability of obtaining the expected state when sequences of randomly chosen quantum gates are run repeatedly.
In the experiment, when the five qubits were operated individually, all exceeded 99.99% fidelity, and some approached 99.999%. Next, when the central three qubits were driven simultaneously with a gate time of 250 nanoseconds (ns), all three stayed above 99.99%. The team also succeeded in driving all five at once, with four exceeding 99.9% fidelity. The remaining qubit, Q5, was reported at about 99.9%.
The main measurement results are summarized below.
| Operating condition | Basic gate fidelity | Measurement conditions / notes |
|---|---|---|
| Each qubit operated individually | All 5 above 99.99%, some close to 99.999% | Each qubit evaluated individually |
| 3 qubits driven simultaneously | All 3 targets above 99.99% | Central 3 qubits, gate time 250 ns |
| 5 qubits driven simultaneously | 4 above 99.9%, remaining Q5 about 99.9% | Gate times of 250 ns and 500 ns evaluated |
These summarize results shown in Figures 2, 4, and 5 of the paper, and all are fidelities of basic gates on individual qubits. The result for simultaneous operation of five qubits does not mean all five strictly exceeded 99.9%. The Figure 5 caption likewise distinguishes four above 99.9% from Q5 at about 99.9%. Nor can the individual gate fidelities be interpreted directly as the success probability of simultaneous operation of all five qubits as a whole.
The measurement method also includes devices for evaluating high fidelities. For each sequence length, the team prepared 35 random sequences and performed 350 single-shot measurements for each, then averaged the results. From the decay rate of the measured signal, it estimated the fidelity of operations composed of multiple basic gates, and then converted this to values for individual basic gates. For five-qubit readout, only three bits of information can be obtained in one measurement cycle, so the same sequence was run twice with different measurement settings to reconstruct each qubit's state.
That said, obtaining fidelities close to 99.999% does not mean the final digits could be pinned down precisely. The paper's Methods section explains that accurately measuring error rates as small as about one in 100,000 would require longer sequences and additional measurement points. The result shows extremely high control accuracy, but the statistical uncertainty of the measurements must also be taken into account.
Power and Noise Challenges Emerge in Five-Qubit Operation
Fidelity when driving five qubits simultaneously was lower than for three. The team cited as the main cause a reduction in quantum-state coherence due to the increased microwave power required for simultaneous driving. Coherence refers to the preservation of the phase relationships needed for superposition states. Operating five qubits in a short 250 ns requires strong microwaves, which may increase heating and microwave-induced noise.
Extending the gate time to 500 ns allows a smaller microwave amplitude and thus reduces the effects of strong driving. But the longer the operation, the more susceptible it is to phase disturbance from low-frequency noise, so fidelity does not necessarily improve. In fact, no large difference in five-qubit simultaneous-operation fidelity was seen between the 250 ns and 500 ns conditions. There are competing constraints: fast operation increases errors from drive power, while slow operation lengthens exposure to noise.
The phase correction developed here cancels reproducible phase shifts due to the AC Stark effect; it does not restore coherence lost to noise. As a future improvement, the team therefore proposes revising the micromagnet design to produce a stronger magnetic field gradient. A stronger gradient could allow electron spins to be driven efficiently with less microwave power. The paper suggests that a design with a higher field gradient could reduce power-related errors and further improve gate fidelity, but this is a prediction based on design and simulation.
The team estimates that even with the current device structure, ten qubits could potentially be driven simultaneously with a 500 ns gate time. This is based on a comparison in which the total microwave power for driving ten qubits at 500 ns would be comparable to that for the five-qubit, 250 ns condition demonstrated here. However, simultaneous operation of ten qubits was not actually performed, and whether about 99.9% fidelity could be maintained under those conditions remains to be verified.
Furthermore, real quantum computation requires not only operating individual qubits but also two-qubit gates that couple qubits together. The pairwise phase correction here was validated under conditions where exchange coupling between qubits was effectively disabled. When exchange coupling is enabled to run a two-qubit gate, however, phase changes that depend on the partner's quantum state may arise, and the paper notes that additional calibration would be needed. The correction method alone therefore cannot handle every operation involving multiple qubits.
A gate fidelity of 99.9% is considered one important target level for practical fault-tolerant quantum computing. However, the accuracy required for quantum error correction depends on the scheme adopted and the nature of the noise, and this result does not mean fault-tolerant quantum computing itself has been demonstrated. In an actual quantum processor, in addition to single-qubit gates, two-qubit gates, state preparation and readout, and errors during idle periods also affect the reliability of the computation as a whole.
The significance of this research is that it showed high-accuracy parallel operation is possible while keeping down the control wiring and calibration complexity that challenge large-scale integration of silicon qubits. Going forward, it will be important to reduce drive power and suppress noise, and to see whether high accuracy can be maintained in circuits that include inter-qubit coupling operations and readout. If these challenges are overcome, it would bring closer a silicon quantum processor that runs many qubits in parallel while avoiding the burden of adding an independent high-frequency control line for each qubit.