High fidelity control of superconducting qubits with optical transmitted signal

AI summary

Overview

Research area: Quantum computing hardware — cryogenic control and wiring for superconducting qubit processors, specifically replacing room-temperature-to-cryostat coaxial cables with optically transmitted microwave signals.

Technical level: Intermediate (assumes familiarity with transmons, gate fidelity benchmarking, and dilution refrigerator stages, but the central idea is physically intuitive).

Scope: An experimental demonstration that single-qubit and two-qubit gates on frequency-tunable transmon qubits can be driven entirely by microwave signals carried as laser intensity modulation over optical fiber and regenerated by photodiodes at the 4 K plate of a dilution refrigerator, with fidelities of 99.915% ± 0.005% and 99.676% ± 0.041%.

What This Paper Is About

Scaling superconducting quantum processors requires more and more microwave cables running from room-temperature electronics into a dilution refrigerator, and those cables consume both physical space and cooling power. The authors replace the coaxial cables carrying XY (drive) and Z (flux) control signals with optical fibers, modulating a laser at room temperature and converting the light back into microwave current on photodiodes mounted at the 4 K stage. The goal is to show that this "optical-assisted transmission line" can control qubits well enough to meet surface-code error-correction requirements.

Key Contributions

1. Full optical control of a tunable transmon processor. Both XY and Z control signals are transmitted optically and regenerated by InGaAs photodiodes at the 4 K stage, with no coaxial cables between room temperature and 4 K for those lines.

2. A thermal-load engineering comparison across refrigerator stages. Using the Bluefors XLD400 as a reference, the authors compare installing photodiodes at the 4 K, Still, cold plate (CP), and mixing chamber (MXC) stages, and identify the 4 K stage as the most feasible near-term choice for thousands of qubits.

3. Noise characterization of the optically transmitted Z line. The noise power spectral density is measured by qubit-based methods (Ramsey tomography oscilloscope, CPMG) and room-temperature instruments (multimeter, lock-in amplifier, spectrometer), spanning 10⁻³ to 10⁷ Hz, and is converted into estimated idle-gate error rates.

4. Gate benchmarking above surface-code thresholds. Cross-entropy benchmarking and Speckle Purity Benchmarking give single-qubit Pauli fidelities of 99.915% ± 0.005% and 99.854% ± 0.014%, and a two-qubit CZ Pauli fidelity of 99.676% ± 0.041%.

Main Findings

• Single-qubit gate fidelity: 99.915% ± 0.005% for qubit Q1 and 99.854% ± 0.014% for qubit Q2, using a 50 ns X/2 gate.

• Two-qubit CZ gate fidelity: Pauli fidelity 99.676% ± 0.041%, with a dressed fidelity of 99.445% ± 0.041%, for a 38 ns gate between Q1 and Q2.

• Control errors are not the limiting factor: Average control error is 0.004% for Q1 and 0.023% for Q2 on single-qubit gates, and 0.053% for the CZ gate. These are described as far below the errors introduced by qubit decoherence.

• Coherence with optical drive: T₁ = 58.58 ± 3.39 μs and T₂* = 2.40 ± 0.23 μs, measured 409 MHz away from the flux sweet spot; the fitted correlated-noise dephasing time T_φ2 is 2.47 ± 0.49 μs.

• Idle-gate error budget: Errors of a 50 ns idle gate from T₁ and T_φ2 are estimated at 0.028% and 0.013% respectively.

• Z-line noise is low enough: Integrating the measured PSD with Ramsey and spin-echo weighting functions gives an idle-gate error of about 0.04% at 10⁻³ Hz, despite several Lorentzian noise peaks in the spectrum.

• Photocurrent tradeoff: Lower photocurrent reduces noise but lengthens gates and increases decoherence; the chosen value for the single-qubit gate is 103 μA.

• Thermal headroom: Per-channel photodiode dissipation of 0.058 mW (XY only) at 4 K supports 25,816 channels against 1,500 mW of cooling power; with both XY and Z (0.752 mW per channel) the number is 1,996. At the Still stage the corresponding numbers are 688 and 53.2, and at the CP and MXC stages the supportable channel counts fall to 10.9/0.81 and 3.3/0.24 respectively.

• Passive heat load advantage: Conventional niobium-titanium and stainless-steel coaxial cables contribute roughly 1 mW of passive load at the 4 K layer (about 1,500 cables for 1.5 W of cooling), whereas the calculated passive load of an optical fiber at 4 K is about 5.6 μW — two orders of magnitude more links.

• Wavelength choice is constrained by cryogenic bandgap: InGaAs has a room-temperature bandgap of 0.735 eV (cutoff 1687 nm), rising to E_g(4 K) = 0.8208 eV (1511 nm), so the 1490 nm laser sits safely inside the cryogenic cutoff while staying compatible with C-band optics. One photon yields at most one free electron, which limits heat load.

• Modulator choice: A directly modulated distributed feedback laser is used rather than an external electro-optic modulator because it gives better linearity, suppresses higher harmonics, and avoids piezoelectric oscillation that distorts Z square pulses.

• Polarization sensitivity is a stability concern: Photodiode response depends slightly on input polarization, with polarization dependence loss D typically at the 10⁻² to 10⁻³ level, producing response fluctuation D/(1−D); the optimal X-gate drive power shifts with photocurrent as polarization drifts.

• Plug-and-play compatibility: Apart from power supply, the optical link has two microwave interfaces (input and output), so it behaves like a normal RF cable and requires no changes to the existing control electronics.

Methodology in Plain English

The team generated control signals with a homemade arbitrary waveform generator. Z pulses were square waves produced directly; XY pulses were gigahertz microwaves up-converted with an IQ mixer. Instead of sending these signals down coaxial cables, they used them to drive a directly modulated laser emitting at 1490 nm, and sent the light into the dilution refrigerator over optical fiber. At the 4 K plate, InGaAs PIN photodiodes with a 1 V reverse bias converted the light back into photocurrent, recreating the microwave drive. Impedance matching in the electronic wiring was designed carefully because even small analog distortion causes gate errors.

They tested this on two neighboring frequency-tunable transmon qubits sharing a readout cavity, with a tunable coupler setting their interaction. They measured T₁ and Ramsey oscillations, and they mapped the noise on the optically delivered Z line in two complementary ways: through the qubit itself using Ramsey tomography oscilloscope (for timescales above roughly 1 s) and CPMG sequences (500 ns to 25 μs), and at room temperature with a multimeter, lock-in amplifier, and spectrometer. The combined spectrum was fit to a model of white noise, 1/f noise, and three Lorentzian peaks. Finally, gates were calibrated in the standard way — frequency by Ramsey, amplitude and Drag Alpha by repeated X gates — and two-qubit CZ pulses were refined by stacking many CZ gates to amplify leakage and conditional-phase errors. Because impedance mismatch distorts square pulses, they applied multi-order correction voltages, which require a bias offset since photocurrent cannot be negative. Fidelity and purity were then extracted from cross-entropy benchmarking and Speckle Purity Benchmarking, with the difference between them taken as the control error.

Why This Matters

The paper shows that optical delivery of microwave control is not just a wiring convenience but is compatible with error-correction-grade gate fidelities on tunable transmons. It reframes the wiring bottleneck as something addressable with commercial photonics rather than requiring cryogenic CMOS or SFQ logic, and it quantifies exactly where in the refrigerator the conversion should happen.

Real-world applications:

• Scaling superconducting processors: Replacing most coaxial lines with fiber increases the number of control channels a given dilution refrigerator can host, which is a direct constraint on qubit count.

• Flexible lab architecture: With fiber links, the refrigerator can be physically separated from the control electronics to reduce noise and vibration, and hollow-core fiber could bring propagation close to the speed of light in vacuum, cutting feedback latency.

• Microwave signal distribution: The underlying photonic microwave techniques already underpin radio-over-fiber systems, photonic true-time-delay beamforming, and photonic analog-to-digital conversion, all of which share the same modulation-and-photodiode-detection building blocks.

• Cryogenic system design: The stage-by-stage heat-load tables give hardware designers a practical budget for where to place active cryogenic components.

Industry relevance: The link is described as "plug-and-play" with existing room-temperature control electronics, requiring no changes to the control system, which lowers the barrier to adoption by groups and companies already running commercial dilution refrigerators and rack electronics.

Future Directions

• Extending to thousands of channels in practice. The thermal analysis supports the 4 K stage for thousands of lines in principle, but the demonstration used two qubits; scaling behavior of crosstalk, per-channel calibration, and yield remains untested here.

• Distribution of the regenerated signal below 4 K. The authors note that delivering control signals from the 4 K or Still stage down to the qubits is a shared open challenge for both cryogenic electronics and optical schemes, and suggest low-thermal-conductivity superconducting cables such as NbTi as one avenue.

• Polarization and long-term drift control. Because photodiode response varies with input polarization at the 10⁻² to 10⁻³ level, stabilizing fiber polarization against temperature drift and stress is identified as necessary for long-term stability.

• Reducing active heat load through operation and design. The paper points to lowering duty cycle, improving signal-to-noise ratio, and reducing mutual inductance of control lines (for example, reaching T_φ1 above 1 ms with only 10 dB attenuation at 4 K and 0.4 pH mutual inductance) as ways to relieve cooling-power pressure at the coldest stages.

Target Audience

Quantum computing experimentalists and hardware engineers working on superconducting qubit scaling, cryogenic wiring, and control electronics; dilution refrigerator and cryogenic systems designers; and photonics researchers interested in microwave-over-fiber control of quantum devices. The paper is also useful for program managers assessing whether optical control is a viable path to larger processors, since it reports concrete fidelities and per-channel heat budgets rather than projections.

Authors’ abstract

Superconducting circuits exhibit remarkable potential for constructing large-scale quantum simulation and computation systems, featuring numerous qubits, extended coherence time, and precise control. Nevertheless, the growing number of signal cables poses a challenge in dilution refrigerators due to space and heat load constraints. To overcome this issue, we experimentally implemented an optically-assisted transmission line as an alternative to coaxial cables. By modulating microwave signals on laser intensities at room temperature and regenerating the signals at a cryogenic plate within the dilution refrigerator, we demonstrated full control of superconducting qubits using photocurrent. We demonstrate and benchmark both single-qubit and two-qubit gates on frequency tunable transmon qubits, achieving fidelities of 99.915% $\pm$ 0.005% and 99.676% $\pm$ 0.041%, respectively, which have reached the requirement of the surface code.