⚛️ quantum physics

Optimizing Pump Conditions of Parametric Amplifiers for Fast Multiplexed Readout of Superconducting Qubits

This paper proposes and experimentally validates a pump calibration strategy for parametric amplifiers that optimizes the signal-to-noise ratio of the slowest qubit in a multiplexed readout chain, thereby reducing the total readout time by 320 ns without compromising the performance of any individual qubit.

Original authors: Jeongwon Kim, Wei Dai, Omrie Ovdat, Akiva Feintuch, Nir Alfasi, Yonuk Chong

Published 2026-06-23

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Jeongwon Kim, Wei Dai, Omrie Ovdat, Akiva Feintuch, Nir Alfasi, Yonuk Chong

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to listen to five different friends (superconducting qubits) who are all whispering their secrets to you at the same time. To hear them clearly, you need a very sensitive microphone (a parametric amplifier) that boosts their voices without adding static.

However, there's a catch: the microphone isn't perfect. Depending on how you tune its settings (the "pump condition"), it might make Friend A sound crystal clear while making Friend B sound muffled, or vice versa.

The Problem: The Slowest Runner Sets the Pace

In this experiment, the researchers had five "friends" (qubits) to read out simultaneously. To get a reliable answer from each, they needed to listen long enough to be sure of what was said.

The Bottleneck: Because they all start talking at the same time, the whole group has to wait until the slowest person finishes their sentence before the meeting can end.
The Limiting Qubit: In their setup, one specific qubit (let's call it Q4) was the slowest. It needed 1,280 nanoseconds (a billionth of a second) to be heard clearly. Even though the other four friends could be understood in just 560 or 640 nanoseconds, the whole system had to wait for Q4.

The Old Way: "Average" Tuning

Previously, engineers would tune the microphone to make the average sound of all five friends as good as possible.

The Flaw: This is like a teacher trying to teach a class by aiming for the "average" student. You might help the middle students, but you might accidentally make the slowest student even slower, or you might not help the slowest student enough to speed them up.
The Result: The total meeting time remained stuck at 1,280 nanoseconds because Q4 was still the bottleneck.

The New Strategy: "Bottleneck" Tuning

The authors propose a smarter strategy: Ignore the average; focus on the slowest one.

Identify the Bottleneck: First, figure out which qubit is taking the longest (Q4).
Set a Safety Net: Make sure the microphone settings don't make any of the other four friends too quiet (they must still be audible, above a certain "volume threshold").
Optimize for the Slowest: Tweak the microphone settings specifically to boost the signal for Q4, even if it means the other friends get slightly less boost than they had before.

The Analogy: The Relay Race

Think of this like a relay race where five runners must cross the finish line together.

Old Method: You train the team to have the best average speed. If Runner 4 is slow, the team still loses time waiting for them.
New Method: You look at Runner 4. You give them a special pair of running shoes (the optimized pump) that makes them run faster. You check to make sure the other runners still have shoes that work fine (they don't trip). Even if the other runners are now slightly slower than before, they are still faster than Runner 4 used to be.
The Outcome: Runner 4 finishes much faster. Since the whole team waits for the last person, the entire team finishes the race sooner.

The Results

By using this "bottleneck-focused" strategy:

They tuned the amplifier to specifically help the slowest qubit (Q4).
Q4's reading time dropped from 1,280 ns to 800 ns.
The other qubits took a little longer, but they were still fast enough that they didn't become the new bottleneck.
Total Time Saved: The entire multiplexed readout time dropped from 1,280 ns to 960 ns.
Efficiency: This is a 320 nanosecond reduction (about a 25% speedup) without needing to buy new hardware or change the physical setup. They just changed the "settings" to be smarter about who needed help the most.

Why It Matters

In quantum computing, time is precious. Every nanosecond a qubit spends waiting to be read is a nanosecond it could lose its delicate quantum state (decoherence). By shaving off 320 nanoseconds from the readout process, the researchers made the system faster and more efficient, simply by changing how they tuned the amplifier to prioritize the "slowest runner" in the pack.

Technical Summary: Optimizing Pump Conditions of Parametric Amplifiers for Fast Multiplexed Readout

Problem Statement
In superconducting quantum computing, fast and high-fidelity qubit readout is critical for operations such as quantum error correction (QEC) and measurement-based initialization. Low-noise parametric amplifiers, specifically traveling-wave parametric amplifiers (TWPAs) and impedance-matched Josephson parametric amplifiers (IMPAs), are essential first-stage components to boost the signal-to-noise ratio (SNR) within short readout times. However, in frequency-multiplexed readout architectures—where multiple qubits are read out simultaneously on a single feedline—the performance of these amplifiers is highly sensitive to pump conditions (frequency and power).

Due to environmental impedance mismatches and device nonidealities, TWPAs exhibit frequency-dependent gain profiles with ripples. Consequently, a pump condition that maximizes the average SNR improvement across a band may degrade the SNR for specific qubits. In a multiplexed readout scenario, the total operation time is dictated not by the average channel, but by the "limiting qubit"—the qubit requiring the longest duration to reach a target SNR. Conventional tuning procedures, which often target nominal gain specifications or average SNR improvements, fail to account for these ripples and do not specifically address the bottleneck imposed by the slowest readout channel.

Methodology
The authors propose a systematic strategy to optimize the TWPA pump condition specifically to minimize the total multiplexed readout time ( $T^{(MPX)}_{RO}$ ) without compromising the target SNR for any individual qubit. The methodology involves the following steps:

Identification of the Limiting Qubit: The team defines the total multiplexed readout time as the maximum of the individual readout times ( $T^{(i)}_{RO}$ ) required for each qubit to reach a target SNR (set to 3 in this experiment). The qubit with the longest $T^{(i)}_{RO}$ is identified as the limiting qubit.
Pump Calibration and Constraints: The TWPA pump frequency and power are swept to evaluate the gain and SNR improvement ( $\Delta$ SNR) at each qubit's readout frequency. A minimum $\Delta$ SNR threshold (6 dB in this study) is imposed for all qubits to ensure no channel falls below acceptable performance.
Optimization Strategies: Two operating points are compared:
- Average-Qubit-Optimized: The feasible pump condition that maximizes the average $\Delta$ SNR across all qubits.
- Limiting-Qubit-Optimized: The feasible pump condition that maximizes the $\Delta$ SNR specifically for the limiting qubit, while maintaining the minimum threshold for all other qubits.
Experimental Validation: The strategy was tested on a five-qubit multiplexed readout chain using a TWPA. Readout times were determined by finding the shortest pulse duration that achieves SNR = 3 for each qubit under different pump conditions.

Key Results
The experiment demonstrated that prioritizing the limiting qubit yields a faster overall readout cycle compared to optimizing for the average performance:

Baseline Performance: Under the "average-qubit-optimized" pump condition, Q4 was the limiting qubit, requiring a readout time of 1280 ns. Consequently, the total multiplexed readout time was 1280 ns.
Optimized Performance: By retuning the pump to maximize the $\Delta$ SNR for Q4 (the limiting qubit), its required readout time was reduced to 800 ns. Although this adjustment slightly increased the readout times for other qubits (e.g., Q1 and Q2 increased to 960 ns), these channels remained faster than the original bottleneck.
Net Gain: The new limiting qubits became Q1 and Q2, setting the total multiplexed readout time to 960 ns. This represents a 320 ns (25%) reduction in total readout time compared to the average-optimized approach.
Fidelity: The mean single-shot readout fidelity remained stable, decreasing only marginally from 95.45% to 95.32%, indicating that the reduction in time did not degrade the target SNR or introduce significant assignment errors.

Significance and Claims
The paper claims that this "limiting-qubit-optimized" strategy offers a practical, hardware-free method to mitigate experimental nonidealities like gain ripples in parametric amplifiers. By shifting the optimization focus from average performance to the specific bottleneck channel, the authors demonstrate a direct improvement in the system-level figure of merit: total multiplexed readout time.

The authors state that while the experiment utilized a TWPA, the strategy is directly applicable to IMPAs and other multiplexed readout architectures. They position this work as a step toward system-level optimization for scalable quantum processing units (QPUs), noting that future work will involve further accounting for gain compression and readout pulse shaping. The core contribution is a calibration protocol that integrates amplifier pump tuning with readout pulse optimization to minimize the idle time and decoherence accumulation of qubits during measurement cycles.