Native CCZ Gate with Fluxonium Qubits and a Microwave-Driven Coupler

AI summary

Overview

Research area: Superconducting quantum computing — native multi-qubit gate implementations in fluxonium–transmon hardware.

Technical level: Intermediate. The paper is experimentally oriented and its central results are readable with a working knowledge of qubit gates and benchmarking, though the Supplementary Information is heavy on circuit quantization and calibration theory.

Scope: The paper reports an experimental demonstration of a native three-qubit controlled-controlled-phase (CCZ) gate in a three-fluxonium processor with a microwave-driven transmon coupler, benchmarked at 99.39(5)% fidelity in 65 ns.

What This Paper Is About

Most superconducting quantum processors are built around single- and two-qubit gates because those are easy to calibrate and scale. Algorithmically important operations such as the Toffoli gate, however, must be broken down into many two-qubit gates (eight CZ gates for nearest-neighbour connectivity), which multiplies execution time and accumulated error. The authors ask whether a three-qubit entangling operation can instead be built directly into the hardware while keeping the simplicity, low parasitic interaction, and scalability that make two-qubit architectures work.

Key Contributions

1. A native, single-pulse CCZ operation. The gate is activated by one microwave 2π pulse applied to the coupler flux line, with the coupler returning to its ground state and imparting a π phase to the target computational state. No multi-drive sequence or multi-step circuit is needed.

2. A measured fidelity of 99.39(5)% at 65 ns. Benchmarking was done with cross-entropy benchmarking (XEB) using single-qubit references, and the gate operates in the coherence-limited regime at long pulse durations.

3. A simple two-step calibration protocol. For a fixed duration and pulse shape, only a Rabi-like amplitude measurement (minimizing residual coupler population after a full 2π rotation) and a Ramsey-type frequency measurement (monitoring conditional phase) are required; the calibration cycle was repeated three times.

4. Multiplexed phase control on |000⟩ and |111⟩. Two individually calibrated control-phase drives combined into one pulse produce simultaneous, independent phase accumulation, benchmarked at 98.87(12)% versus 97.7(2)% for the sequential implementation (both at 80 ns, limited by arbitrary waveform generator output power).

Main Findings

• Gate speed and fidelity: The best CCZ̄ fidelity of 99.39(5)% occurs at a 65 ns gate duration. Fidelity versus duration shows a rapid drop at short pulses (off-resonant excitation of nearby transmon transitions) and an approximately linear decrease at long durations consistent with decoherence.

• Equivalent decomposed performance: Reaching the same fidelity through a conventional decomposition, estimated under the superconducting-inspired SI1000 circuit-level noise model, would require CZ fidelities of approximately 99.94% — at the level of the best reported superconducting two-qubit gates.

• Low parasitic interactions: Conditional Ramsey oscillations give longitudinal ZZ coupling below 10 kHz; the individually reported static values are ZZ₁₂/2π = 9.8 ± 1.6 kHz, ZZ₁₃/2π = 9.2 ± 3.8 kHz, and ZZ₂₃/2π = 8.7 ± 1.7 kHz.

• Strong state-dependent coupler response: Detunings of the coupler transition from 4.901 GHz across the eight computational states range from 0 MHz (|000⟩) to 388 MHz (|111⟩), with intermediate values of 130, 132, 238, 228, 310, 310 MHz for |001⟩, |010⟩, |011⟩, |100⟩, |101⟩, |110⟩ respectively.

• Spectral selectivity sets coherent errors: The nearest unwanted transitions are separated by 130 MHz for the CCZ̄ implementation and 78 MHz for the CCZ implementation, playing a role analogous to qubit anharmonicity in single-qubit gates.

• Coherence, not mechanism, is the limit: Numerical simulations predict errors below 10⁻⁴ for durations around 100 ns in the absence of decoherence, and the measured trend approaches the coherence limit estimated from independent coherence-time measurements.

• Device coherence and frequencies: Fluxonium ω₀₁/2π = 0.730, 0.956, 0.984 GHz with T₁ = 92, 90, 75 μs and T₂^echo = 17, 10, 12 μs; the transmon coupler has ω₀₁/2π = 4.901 GHz with T₁ = 6 μs and T₂^echo = 6 μs.

• Two gate variants available: The |000⟩-conditioned protocol, denoted CCZ̄ (equivalent to CCZ up to single-qubit X rotations), was chosen for the main experiments because its coupler transition has better spectral isolation in this device.

• Robustness to simultaneous driving: Independent phase accumulation on |000⟩ and |111⟩ produced no observable additional coherent errors when individually calibrated tones were combined.

Methodology in Plain English

The processor has three fluxonium qubits, each capacitively connected to a shared transmon coupler. Because the fluxoniums are strongly coupled to the coupler but have large anharmonicity, the coupler's transition frequency shifts by a different amount depending on which computational state the three qubits are in — effectively labeling each of the eight basis states with its own coupler frequency. To implement the gate, the researchers drive the coupler at exactly the frequency corresponding to the target state with a pulse that completes one full Rabi cycle. The coupler starts and ends in its ground state, so no population is left behind, but the target state picks up a π phase. In effect, the three-qubit gate is a single-qubit rotation of the coupler, which is why calibration is as simple as calibrating a single-qubit gate.

Fidelity was extracted from cross-entropy benchmarking circuits that include randomly sampled single-qubit Clifford gates as references. The authors also built a model Hamiltonian from the circuit's capacitance matrix, extracted device energies from two-tone spectroscopy and conditional coupler transition measurements, and simulated the full system keeping five levels per fluxonium mode and three for the transmon. Measurements were performed in a dilution refrigerator at roughly 20 mK, with an impedance-matched parametric amplifier for readout, 2.0 GSa/s arbitrary waveform generators for control, and 2.5 GSa/s digitizers.

Why This Matters

Impact on research: The work shows that native multi-qubit gates need not trade away the properties that made two-qubit architectures successful — simple control, low parasitic coupling, and a clear scaling path. It puts a three-qubit gate on comparable footing with the best two-qubit gates without requiring an order-of-magnitude improvement in coherence times, since the demonstrated fidelity corresponds to decomposed implementations on qubits with coherence times roughly ten times longer.

Potential applications (as motivated in the paper):

• Grover search and Shor factoring, where the Toffoli operation is a recurring primitive.
• The Harrow–Hassidim–Lloyd linear-systems algorithm.
• Quantum chemistry simulations.
• Error-correction protocols that use measurement-free syndrome processing.

Industry relevance: Reducing gate-count overhead directly lowers the time and error budget of algorithm execution on superconducting hardware. The architecture described — a coupling element shared among qubits, driven by a single tone — is compatible with the fabrication and control infrastructure already used for microwave-activated CZ gates, which matters for manufacturers pursuing hardware-efficient scaling.

Future Directions

• Two-dimensional scaling. In 2D lattices, multiple transmons would connect to the same fluxonium island and couple capacitively; the authors propose auxiliary elements or complex couplers such as the double-transmon, and note that a three-qubit coupling element lets each fluxonium island connect to at most one transmon even in 2D.

• Improved coherence as a direct route to higher fidelity. Since the gate is coherence-limited, the authors state that better coherence times should translate directly into higher three-qubit gate fidelity.

• Better spectral isolation for the |000⟩-conditioned gate. The 78 MHz separation for CCZ versus 130 MHz for CCZ̄ currently limits the fidelity of the simultaneous implementation to that of the weaker constituent gate, so improving isolation on that transition is an open engineering target.

• Fully parallel native three-qubit operations. The multiplexed result is presented as evidence that adjacent coupling elements driving different computational states can operate simultaneously; confirming this at larger scale remains to be shown.

Target Audience

Superconducting-qubit experimentalists and device engineers, quantum-architecture researchers evaluating native multi-qubit primitives against decomposed circuits, and graduate students or technically literate readers who want a concrete, benchmarked case study in hardware-efficient gate design. Readers looking for algorithmic or error-correction protocol details will find only brief motivation here.

Authors’ abstract

Native multi-qubit gates could reduce the overhead associated with decompositions into single- and two-qubit operations, but whether they can simultaneously provide high fidelity, simple control and robustness against parasitic interactions in scalable architectures remains unclear. Here we experimentally realize a 65-ns native controlled-controlled-phase operation, locally equivalent to the Toffoli gate, with a fidelity of 99.39(5)% in a three-qubit processor unit based on fluxonium qubits coupled via a microwave-driven transmon coupler. The implemented operation would require CZ fidelities of approximately 99.94% if realized through a conventional decomposition. The gate is implemented with a single control pulse, that relies on a simple calibration procedure yielding coherence-limited performance. This processor unit naturally extends to scalable two-dimensional layouts with low parasitic interactions. Altogether, these results establish native multi-qubit gates as a viable hardware-efficient primitive for scalable superconducting quantum processors.