RIKEN Holds 99.9% Fidelity Driving Five Silicon Qubits From One Line

RIKEN researchers have driven five silicon spin qubits at once from a single shared microwave control line and held single-qubit gate fidelity at approximately 99.9%, according to a study published in Nature Physics. The catch is in the comparison: the same five qubits, tested one at a time, each scored well above 99.99%, so running them together carries an error rate at least ten times higher, and that gap is the real story. For developers and researchers tracking when silicon quantum processors might grow past laboratory-sized devices, it is a measured answer to whether parallel control on shared wiring is viable.

This article covers what the team demonstrated, how they corrected the error source they identified, what the results do and do not show, and what remains unanswered.

What Happened

The work comes from the RIKEN Center for Emergent Matter Science and collaborators. Yi-Hsien Wu is first author and carried out the experiments as part of his PhD research. Leon Camenzind, a RIKEN research scientist and co-first author, supervised the experiments. Seigo Tarucha is project manager of a silicon quantum computing project under JST's Moonshot Goal 6.

The team drove five silicon spin qubits (quantum bits that store information in the spin of a single electron confined in silicon) from a single shared microwave control line. With all five operating together, accuracy stayed at approximately 99.9%. Each of the five qubits, tested individually, showed single-qubit gate fidelity well above 99.99%, and the best individual qubit approached 99.999%. With up to three qubits running simultaneously, fidelity held above 99.99%.

Those tiers are distinct and worth keeping apart: above 99.99% for single qubits and for up to three at once, roughly 99.9% for all five at once. Fidelity here is the fraction of gate operations carried out correctly, so a drop from 99.99% to 99.9% means errors go from about one in ten thousand operations to about one in a thousand. Camenzind put the significance this way: "This shows that high-fidelity control can be maintained even when several qubits are operated at the same time and represents an important step toward scaling up silicon-based quantum processors."

The Technical Breakthrough

The hidden obstacle is a side effect of the control signal itself. The researchers traced much of the accuracy loss to the AC Stark shift (a change in a qubit's effective energy levels caused by the microwave field used to control it), which produces unwanted phase shifts in the qubit's quantum state. A phase shift is a drift in the timing relationship that encodes a qubit's information, so left uncorrected it shows up as gate errors when several qubits are driven together from one line.

The fix is a calibration that compensates for those phase shifts using measurements between pairs of qubits. Wu explained the advantage: "Rather than requiring an entirely separate calibration procedure for every possible combination of qubits, our method uses calibrations between pairs of qubits to compensate for the unwanted phase shifts." The number of qubit pairs grows far more slowly than the number of possible qubit combinations, so the approach should scale better than exhaustive calibration, though the source does not test that beyond five qubits.

The result combines three ingredients: the pairwise calibration, an improved device design, and carefully adjusted microwave control pulses.

The calibration does not remove every error. The remaining loss is linked to drive-induced decoherence (the loss of a qubit's quantum coherence caused by the microwave drive itself). The phase shifts are corrected; this decoherence is not. Camenzind acknowledged the limit: "This is a limitation, but further improvements to device design could help mitigate this source of error."

One caveat applies to every number above. All reported figures are single-qubit gate fidelities. The source reports no two-qubit gate results, which means the entangling operations that give a quantum computer its power are not covered by this demonstration.

Why It Matters for Industry

Giving every qubit its own independent control system would make large devices hard to manage. Shared control lines reduce the wiring and hardware needed per qubit, which is among the practical bottlenecks for building bigger processors. The RIKEN result suggests that sharing a line does not have to cost as much accuracy as parallel driving normally does: five qubits on one line still reached about 99.9%.

Silicon adds a second industrial argument. Spin qubits can be built on a platform compatible with established semiconductor manufacturing, and the researchers lean on that point. Camenzind tied the two points together: "Our study supports the use of shared control lines as a practical approach to scaling up quantum processors. Combined with silicon's compatibility with established semiconductor manufacturing, we believe that this advance could help pave the way toward larger quantum processors capable of performing increasingly complex computations."

Note the wording: "could help pave the way." The source gives no cost, yield, or timeline figures, so the commercial payoff remains a stated expectation rather than a measured one.

Competitive Landscape

The RIKEN paper, as relayed in the release, names no competitors and offers no benchmark comparisons against other quantum hardware, and the verified material for this story contains no peer data. This article therefore does not cite other companies' qubit results. What the numbers do allow is a comparison inside the experiment itself:

One qubit at a time: well above 99.99%, with the best qubit approaching 99.999%.
Up to three qubits at once: above 99.99%.
All five qubits at once: approximately 99.9%.

Anyone setting this result beside other platforms should check three things first: whether the other figure is a single-qubit or two-qubit gate, how many qubits were driven simultaneously, and whether they shared one control line. The five-qubit result is a scaling demonstration for single-qubit gates, not a head-to-head ranking.

The Bigger Picture

The RIKEN team states its aim plainly: "Our long-term goal is to build a fault-tolerant quantum computer based on silicon spin qubits, and reliable parallel control is an essential step toward that goal." A fault-tolerant machine is one that can detect and correct its own errors while computing. The project sits under JST's Moonshot Goal 6, per Tarucha's title.

The gap between that goal and this result is substantial. Five qubits with single-qubit gates is an early step on a long path. A full processor needs two-qubit gates, error correction, and control at far larger scale, and none of those appear in this paper's reported results. What the paper does supply is evidence on one specific question: whether the control hardware can be shared without giving up most of the accuracy. For five qubits, the answer is that it costs roughly one decimal place of fidelity.

What's Next

Three open items follow directly from the source. First, device-design changes to reduce drive-induced decoherence, which Camenzind says could mitigate the main remaining error.

Third, and most consequential, is two-qubit gate performance under shared control.

The wiring problem looked like an engineering footnote and now looks like a central constraint. RIKEN has shown that the phase-shift half of that constraint can be calibrated away, and named the decoherence half as the remaining obstacle. The next test is not a sixth qubit but the first shared-line entangling gate.

For a researcher designing a multi-qubit silicon device, the takeaway is concrete: calibrating pairs instead of every combination keeps characterization effort from exploding, while drive-induced decoherence is the number to watch as you add qubits to a line, because that is where the reported accuracy slipped from above 99.99% toward 99.9%.

-- Aria Lin, Enterprise Technology Analyst

Sources: SEC · arXiv · ScienceDaily