Better Qubit Reading by Using Two Paths

In the paper's words

Readout error dominates the overhead of quantum error correction in superconducting processors, limiting logical error rates and scaling prospects. Existing dispersive readout chains suffer from common‑mode disturbances such as amplifier drift, thermal fluctuations, and stray coupling, which add directly to the measured quadrature and reduce single‑shot fidelity. A gap remains in applying the well‑established differential amplification principle of electrocardiography to quantum measurement.

The abstract, verbatim; every line below is quoted from the text.

What it claims

that differential readout yields a measurable SNR improvement—would be falsified if an experiment implementing the balanced architecture observed no reduction in the variance of the difference channel relative to the single‑ended channel, despite calibrated common‑mode injection. Specifically, measuring σ_rm diff²>2σ_rm ind² would contradict the analytic prediction.

Numbers it reports

1. Incorporating the two‑mode‑squeezing results of Ref. [6], which report a power‑SNR ratio of $1.66$ at an amplifier gain of $G=8\,$dB and system noise $N_{\rm sys}=50$ quanta, we use this benchmark for comparison, confirming that the differential scheme can surpass conventional phase‑preserving amplification.

   from Abstract

2. A general SNR gain expression $G_{\rm SNR}= \sqrt{2(1+r^{2})}$, where $r$ is the ratio of common‑mode to independent noise, yields a numerical gain of $1.581$ for $r=0.5$.

   from Abstract

3. Projected single‑shot fidelity improves from $98.5\%$ to $99.4\%$ when common‑mode noise dominates, and from $96.3\%$ to $99.8\%$ for $r=0.5$, reducing logical error rates by a factor of about six in a distance‑three surface code and by a factor of $\sim250$ in a $[[3,1,2]]$ repetition code.

   from Abstract

4. Current single‑ended dispersive readout typically achieves single‑shot fidelities of $98\!-\!99\%$, This overhead translates into prohibitive hardware costs and power consumption, and it stalls the deployment of fault‑tolerant quantum computers.

   from 1. Introduction

5. Numerical SNR improvement factor $\sqrt{2}\approx1.414$ for the case of equal independent and common‑mode noise variances (Section 4).

   from 1. Introduction

6. Integration of the two‑mode‑squeezing SNR ratio $R_{\rm diff}/R_{\rm single}=1.66$ reported in Ref. [6] as a benchmark (Section 4).

   from 1. Introduction

How it is done

1. Starting from a linearized input‑output description, we derive closed‑form expressions for signal‑to‑noise ratio (SNR) and single‑shot fidelity for both single‑ended and differential configurations.

   from Abstract

2. The common‑mode rejection concept has been explored in several domains that inform our approach.

   from 2. Background and Related Work

3. We consider a transmon qubit coupled dispersively to two nominally identical resonators (indexed $a$ and $b$) with coupling rates $g_{a}=g_{b}=g$ and resonator frequencies $\omega_{r}$.

   from 3.1 System Model

4. Substituting $\sigma_{\rm cm}=r\,\sigma_{\rm ind}$ into the expressions above yields $$ \mathrm{SNR}_{\rm SE}= \frac{S}{\sigma_{\rm ind}\sqrt{1+r^{2}}},\qquad \mathrm{SNR}_{\rm diff}= \sqrt{2}\,\frac{S}{\sigma_{\rm ind}}. $$ The amplitude‑SNR gain is therefore $$ G_{\rm SNR}= \frac{\mathrm{SNR}_{\rm diff}}{\mathrm{SNR}_{\rm SE}} = \sqrt{2(1+r^{2})}. $$ Two limiting cases: $r=0$ (no common‑mode noise): $G_{\rm SNR}= \sqrt{2}\approx1.414$. $r\to\infty$ (common‑mode dominated): $G_{\rm SNR}\to\sqrt{2}\,r$, growing without bound.

   from 4.3 General SNR Gain Formula

What would falsify it

1. The central claim—that differential readout yields a measurable SNR improvement—would be falsified if an experiment implementing the balanced architecture observed no reduction in the variance of the difference channel relative to the single‑ended channel, despite calibrated common‑mode injection.

   from 6.2 Failure Modes and Falsifiability

2. Specifically, measuring $\sigma_{\rm diff}^{2}\gt 2\sigma_{\rm ind}^{2}$ would contradict the analytic prediction.

   from 6.2 Failure Modes and Falsifiability

3. Observation of a variance reduction greater than $1.0$ (i.e., any measurable cancellation) would support the theory; a reduction equal to $1.0$ would falsify it.

   from (c) Minimal Next Experiment

Where the claim stops

1. Assumption Sensitivity Equal independent noise variance in both channels ($\sigma_{\rm ind}^{2}$) Moderate: deviations alter the $\sqrt{2}$ factor but the cancellation of $n^{\rm cm}$ remains.

   from 6.1 Limitations

2. Substituting the model yields $$ V_{\rm diff}= (S\,\sigma_{z}+n_{a}^{\rm ind}+n^{\rm cm})-(S\,\sigma_{z}+n_{b}^{\rm ind}+n^{\rm cm}) = n_{a}^{\rm ind}-n_{b}^{\rm ind}. $$ Thus the common‑mode term cancels exactly, and the noise variance becomes $$ \sigma_{\rm diff}^{2}= \langle (n_{a}^{\rm ind}-n_{b}^{\rm ind})^{2}\rangle = 2\,\sigma_{\rm ind}^{2}, $$ assuming the independent noises are uncorrelated and have equal variance.

   from 3.3 Differential (Einthoven) Readout

3. For binary discrimination with Gaussian noise, the single‑shot fidelity $F$ can be approximated by $$ F \approx \frac{1}{2}\,\operatorname{erfc}\!\left(-\frac{\mathrm{SNR}}{\sqrt{2}}\right), $$ where $\operatorname{erfc}$ is the complementary error function.

   from 3.4 Fidelity Approximation

Where to read

1. Abstract270 words · 1 min
2. 1. Introduction363 words · 2 min
3. 2. Background and Related Work359 words · 2 min
4. 3. Methods326 words · 1 min
5. 4. Analysis531 words · 2 min
6. 5. Results146 words · 1 min
7. 6. Discussion274 words · 1 min
8. 7. Practical Implications and Engineering Pathways383 words · 2 min
9. 8. Conclusion238 words · 1 min
10. Provenance and AI Involvement225 words · 1 min