Operating two exchange-only qubits in parallel
Article Date: 26 November 2025
Article URL: https://www.nature.com/articles/s41586-025-09767-5
Article Image: 
Summary
Intel researchers demonstrate that two exchange-only qubits encoded across six quantum dots each can be driven in parallel with carefully calibrated exchange pulses. Using a 12-dot Tunnel Falls device fabricated on a 300-mm process, they implement charge-locking PSB readout, next-nearest-neighbour barrier–barrier compensation (virtualization) to reduce crosstalk, and pulse pre-distortions to counter signal-path filtering. Single-qubit Clifford fidelities reach ≈99.84% and 99.41% individually, dropping only ~0.05–0.07% when driven simultaneously. Two-qubit Clifford fidelity is 96.25% sequentially and 95.80% when maximally parallelised, while parallelisation reduces two-qubit gate time by ~37–40%. The team also reports an exchange-only iSWAP implementation and cross-entropy benchmarking (XEB) showing comparable per-cycle performance for sequential and parallel modes (~96.8% vs ~97.0%). Main remaining limits are calibration precision, charge noise and residual crosstalk; quadratic compensation and improved materials are suggested next steps.
Key Points
- First experimental demonstration of high-quality simultaneous exchange pulsing on next-nearest electrodes in a silicon 12-dot device.
- Introduced charge-locking PSB readout to avoid information loss from sequential readout and enable full two-qubit state extraction.
- Developed next-nearest-neighbour barrier–barrier compensation within a virtualization framework to suppress crosstalk during parallel pulses.
- Single-qubit Clifford fidelities: 99.84% (Q1) and 99.41% (Q2); simultaneous operation reduces fidelity by only 0.05–0.07%.
- Two-qubit Clifford fidelities: 96.25% sequential, 95.80% parallel — a small fidelity cost for ~40% shorter gate duration.
- Implemented an exchange-only iSWAP gate and used randomized benchmarking (RB) and XEB to cross-check performance.
- Monte Carlo analysis attributes errors to magnetic (nuclear-spin) noise, charge noise, calibration and non-Markovian control effects.
- Device uniformity from industrial 300-mm fabrication reduces calibration overhead and aids matrix regularity for virtualization.
Content Summary
The authors operate two exchange-only qubits (each encoded in three spins across three quantum dots) embedded in the central six dots of a 12-dot Intel Tunnel Falls array. They use linear virtualization extended to include next-nearest barrier–barrier compensation to correct capacitive and barrier crosstalk that appears during simultaneous pulsing. Pulse pre-distortion and buffer-time spectroscopy are applied to mitigate transmission-line filtering and pulse-overlap non-Markovian errors.
Performance characterisation used blind single-qubit RB, two-qubit RB, interleaved RB for CNOT/iSWAP/SWAP and cross-entropy benchmarking. Parallel single-qubit driving shows negligible fidelity loss. Parallelising commuting exchange pulses in two-qubit gates reduces time (up to ~40%) and nuclear-spin-related error, but residual calibration and non-linear crosstalk limit net fidelity improvement. XEB per-cycle fidelities for sequential and parallel modes are matched within uncertainties.
Context and Relevance
This experiment is practically important for scaling spin-based quantum processors. Parallel operations reduce idle time (and hence decoherence), are essential for running deeper near-term circuits within coherence windows, and are a prerequisite for effective quantum error correction. Exchange-only qubits avoid microwave control by using voltage pulsing alone, which simplifies control electronics and favours dense 2D layouts compatible with silicon foundry processes. Demonstrating parallelised exchange pulses with only modest fidelity loss is a clear step towards scalable, fault-tolerant architectures in silicon.
Why should I read this?
Short and sweet: if you care about scaling silicon spin qubits or about practical control schemes that dodge microwaves, this paper shows you can run two exchange-only qubits in parallel with nearly the same single-qubit quality and cut two-qubit gate times by almost half. It’s full of hands-on calibration tricks (barrier–barrier compensation, charge-locking readout, pulse pre-distortion) you can steal for your lab or design spec.