Broadband Purcell Filter for Fast Superconducting Qubit Reset and Readout

AI summary

Overview

Research area: Superconducting quantum computing hardware — specifically microwave filter engineering and qubit control/readout architecture for scalable processors.

Technical level: Advanced. The paper assumes familiarity with circuit QED, Purcell decay, dispersive readout, transmon parameters (κ, χ, g), and distributed microwave circuit design.

Scope: The paper designs, fabricates and experimentally characterizes a distributed broadband Purcell filter in a flip-chip superconducting processor that simultaneously provides a 1 ms-scale Purcell-limited relaxation-time floor over 1.2 GHz, 500 ns readout without a Josephson parametric amplifier, and 100 ns reset at 99.6% efficiency.

What This Paper Is About

Fast qubit reset and fast readout both require strong coupling to a lossy environment, but that same coupling makes the qubit decay through the Purcell effect. The authors build a microwave filter that separates the reset path from the readout path so both can be optimized independently, while a protected frequency band keeps the qubit isolated. The goal is a hardware layout that removes the usual trade-off between speed of operations and qubit coherence, as needed for quantum error correction.

Key Contributions

1. A three-notch-band architecture in one distributed structure. A three-quarter-wavelength (3λ/4) coplanar waveguide backbone carries a readout node placed λ_int/4 from the open end (an intrinsic notch) plus a parallel λ_bs/4 bandstop stub. Separating the two notches creates a broadband Purcell-protected window, and an additional reset stub inserted between them forms a half-wavelength (λ/2) reset mode below the qubit frequencies whose frequency is set by the stub length alone.

2. Decoupled reset channel using dedicated per-qubit reset resonators. Rather than coupling qubits directly to a shared filter bus (which risks filter-mediated stray interactions between qubits), each qubit gets its own reset resonator, designed with g_q,rst > κ_rst/4 and κ_rst ≫ 2χ_rst.

3. A deliberate κ–χ mismatch that suppresses photon-shot-noise dephasing. Because the reset resonator is not optimized for dispersive readout, residual photons in it dephase the qubit roughly 70 times less than readout-resonator photons do, so the system need not wait for the reset resonator to empty before continuing.

4. A compensation protocol for the readout mode. Extending the reset stub perturbs the readout mode; the authors derive an analytic correction to the electrical length of the grounded section of the backbone (δθ_s) that holds the readout-mode frequency fixed while the reset mode is tuned from 3.8 to 4.5 GHz.

Main Findings

• Broadband Purcell protection: The extracted Purcell-limited relaxation time T_1,P stays at a floor of at least 1 ms across approximately 1.2 GHz (4.7–5.9 GHz), in agreement with SPICE circuit simulation. The measured (total) T_1 of qubit Q1 in the protected band is approximately 35 μs, with a median T_1 of 34.2 μs over 4.4–5.6 GHz.

• Two transmission zeros define the protected band: Measured S_21 zeros sit at ω_int/2π = 5.6 GHz (intrinsic notch) and ω_bs/2π = 4.9 GHz (bandstop notch), with the readout passband near 7.3 GHz and the reset band near 4.1 GHz.

• Extracted band linewidths: 300 MHz for the readout band and 100 MHz for the reset band; the readout bandwidth can be widened to as much as 800 MHz by adjusting the readout-node position and reset-stub length (Appendix B).

• Fast readout without a parametric amplifier: Single-shot histograms are taken with a 500 ns measurement pulse and no Josephson parametric amplifier.

• Fast reset at high efficiency: With a t_rst = 100 ns reset pulse (more than ten times the reset-resonator decay time), the excited-state population drops from P_e = 92.67% to P_e^rst = 0.37%, giving η_rst = 1 − P_e^rst/P_e = 99.6%.

• Underdamped iSWAP reset dynamics: For Q1, κ_rst/2π ≈ 18 MHz and g_q,rst/2π ≈ 32 MHz, so g_q,rst > κ_rst/4. The chevron pattern shows a decay envelope with time constant approximately 17.6 ns, and the excitation decays to baseline within 100 ns. Ending the pulse at the first population minimum allows a reset shortened to approximately 20 ns.

• 70× dephasing suppression: Injected thermal-photon noise gives a dephasing sensitivity of γ_rst ≈ 1.02 μs⁻¹ for the reset resonator versus γ_ro ≈ 70.4 μs⁻¹ for the readout resonator — a factor of about 70. These data come from a separate sample with κ_rst/2π ≈ 14.4 MHz and χ_rst/2π = 0.16 MHz.

• Thermal photons matter: Adding 5 GHz high-pass filters to the read-in and readout lines reduces the residual population after reset from 1.87% (without HPFs, separate cool-down) to 0.37%.

• Qubit Q1 parameters: ω_q,max/2π = 5.664 GHz, α/2π = −210 MHz, readout resonator ω_r/2π = 7.28 GHz with κ_r/2π = 7.58 MHz, reset resonator ω_rst/2π = 4.08 GHz with κ_rst/2π = 18.0 MHz.

• A measured limitation on intrinsic T_1: Both T_1 and Rabi drive strength oscillate with a common period of approximately 110 MHz, matching the free spectral range of a 90-cm coaxial drive line anchored at the mixing chamber, indicating the intrinsic T_1 of this sample is limited by drive-line-induced loss.

Methodology in Plain English

The team built the filter as patterned microwave wiring (coplanar waveguide) on a chip rather than as a lumped-element circuit. A long backbone section carries the readout signal; placing the output tap a quarter-wavelength from one end creates a notch that blocks qubit-frequency photons from escaping, and a second quarter-wavelength stub attached in parallel creates another notch. Two notches spaced apart leave a wide window where the qubit is protected while the readout frequency, placed above the qubits, still passes through. A third stub placed between the two notches forms a resonant mode at a lower frequency, and its length alone sets where that reset mode lands.

The processor is a flip-chip stack: qubits and their dedicated reset resonators on the top chip, filter and readout resonators on the bottom, with each reset resonator sitting directly above its readout resonator to save space. The device is cooled to 10 mK.

To measure protection, a direct T_1 measurement is not enough because it mixes many loss channels, so the authors instead drive the qubit through the readout port at varying frequency, measure the Rabi oscillation strength, and convert that into a Purcell decay rate — calibrating the drive power at a bias point where Purcell decay clearly dominates. This is cross-checked against a SPICE model of the full circuit with 50 Ω terminations that computes the real part of the admittance seen at the qubit port.

Reset is characterized by preparing the excited state, pulsing the qubit into resonance with its reset resonator for a variable time, and mapping the resulting chevron pattern; efficiency is extracted from single-shot histograms. Dephasing from stray photons is measured by deliberately injecting calibrated white noise alongside a coherent tone into each resonator and recording the spin-echo dephasing rate at the qubit sweet spot versus added photon number.

Why This Matters

The work shows that the three functions a superconducting processor needs from its microwave environment — readout, qubit protection, and reset — can be allocated to separate, independently tunable frequency bands in a single manufacturable distributed structure. This removes the requirement that the optimal readout resonator also serve as the reset channel, which previously forced a compromise between signal-to-noise and reset speed, and it does so without the stray qubit–qubit interactions that come from coupling qubits directly to a shared filter bus.

Real-world applications:

• Quantum error correction cycles: Syndrome extraction requires resetting ancilla qubits repeatedly; a 100 ns reset at 99.6% efficiency (with a route to 10-ns-scale reset) directly shortens cycle time.
• Multiplexed readout in large arrays: A 300 MHz readout band (extendable to 800 MHz) supports many readout resonators with uniform linewidths on a shared feedline.
• Cryogenic hardware reduction: 500 ns single-shot readout without a Josephson parametric amplifier removes a demanding cryogenic component from the readout chain.
• Frequency-allocation flexibility in processor design: Independently parameterizable reset, protection and readout bands let qubit frequencies be biased across a wide window with a stable 1 ms-scale Purcell floor.

Industry relevance: The architecture is built from standard coplanar-waveguide structures in a flip-chip process, using aluminum on one sapphire chip and tantalum on the other, and the compensation formula lets a foundry-style design flow retune the reset frequency without disturbing the readout channel — practical requirements for producing repeated, uniform multi-qubit processor units.

Future Directions

• Pushing reset toward 10 ns: The authors argue that suppressed dephasing from residual reset photons may allow subsequent operations to start before the reset resonator is fully depleted, particularly when the first-minimum excitation swap is used. Demonstrating this in a full algorithm sequence is left open.
• Scaling to multi-qubit processors: The paper characterizes a single example qubit (Q1); verifying that dedicated per-qubit reset resonators avoid crosstalk and maintain reset fidelity across many qubits is the next step.
• Reducing thermal occupation of low-frequency reset resonators: The reset band sits near 4 GHz, where Boltzmann statistics give higher thermal occupation; the improvement from 1.87% to 0.37% residual population with 5 GHz high-pass filters suggests further filtering and infrared shielding work is needed.
• Removing drive-line-limited coherence: The observed 110 MHz oscillation in both T_1 and Rabi strength points to standing waves on the 90-cm coaxial drive line as the limit on intrinsic T_1, so improved infrared filtering or line design is required before the filter's full protection can be seen in raw T_1.

Target Audience

Experimental superconducting-qubit researchers and cryogenic microwave engineers who design readout and reset chains for multi-qubit processors; quantum-computing hardware teams planning scalable architectures with multiplexed readout; and graduate students or advanced readers in circuit QED who want a concrete worked example of how distributed filter design trades off bandwidth, Purcell protection and reset speed. Readers without a background in microwave circuit parameters (κ, χ, g, electrical length) will find the design sections demanding, though the motivation and headline results are accessible.

Authors’ abstract

Rapid reset and readout of qubit states are essential for quantum error correction, yet accelerating these operations through stronger coupling to a dissipative environment inevitably increases qubit decay via the Purcell effect. Here we present a broadband Purcell filter that decouples the reset and readout paths, enabling both operations to be independently optimized without compromising qubit coherence. The filter employs two engineered notches - an intrinsic notch and a bandstop notch - to provide broadband Purcell protection, together with an additional reset stub that creates a reset mode below the protected band. To enable fast reset while suppressing filter-mediated interactions between qubits, we couple each qubit to a dedicated reset resonator. We experimentally demonstrate Purcell-limited relaxation times exceeding 1 ms across a 1.2 GHz bandwidth, simultaneously with 500 ns readout without a Josephson parametric amplifier and 100 ns reset with 99.6% efficiency. The reset resonator is designed with a deliberate kappa-chi mismatch, which suppresses photon-shot-noise-induced dephasing by a factor of 70 compared to the readout resonator. Our work provides a scalable hardware solution that resolves the traditional trade-off between fast qubit operations and qubit protection, advancing the prospects for fault-tolerant quantum computing.