High-throughput electro-optic upconversion and downconversion with few-photon added noise

TL;DR

This work tackles the challenge of designing a microwave–optical transducer with both high throughput and low noise to enable quantum networking among superconducting processors. The authors demonstrate a membrane-based opto‑electromechanical transducer achieving a throughput $\Theta = \eta B D$ exceeding 7 kHz, with input-referred noise near the quantum limit ($N_{\text{add}}^{\text{up}} \approx 2.6$, $N_{\text{add}}^{\text{down}} \approx 3$ photons) and $\eta$ up to ~0.4 over a ~20 kHz band, enabling quantum-enabled operation in both directions. They analyze noise contributions via $N_{\text{add}}^{\text{up/down}}$ decomposed into motion, electromagnetic, and interference terms, and show that downconversion can achieve very low added noise by exploiting higher $\Gamma_{\text{o}}$, while upconversion remains limited by optical noise and sideband dynamics. A theoretical integration of the quantum capacity over the transducer bandwidth yields $\mathcal{C}_{\text{ub}}(N_{\text{add}}, \Theta) \approx \frac{\pi\Theta}{\ln(2)}\left(1 - N_{\text{add}} + N_{\text{add}}\ln N_{\text{add}}\right)$ for $N_{\text{add}} < 1$, highlighting the balance between throughput and noise and guiding design toward quantum-enabled rates at the few‑kHz scale. The results point to practical routes to quantum networks with reasonable averaging times by reducing stiff-mode noise, increasing electromechanical coupling, and maintaining low $N_{\text{add}}$, bringing transducer-assisted entanglement distribution within experimentally accessible memory lifetimes.

Abstract

A microwave-optical transducer of sufficiently low noise and high signal transfer rate would allow entanglement to be distributed between superconducting quantum processors at a rate faster than the lifetimes of the quantum memories being linked. Here we present measurements of a membrane-based opto-electromechanical transducer with high signal throughput, as quantified by an efficiency-bandwidth-duty-cycle product of 7 kHz, approaching quantum-enabled operation in upconversion as well as downconversion, with input-referred added noise of 3 photons. In downconversion, throughput of this magnitude at the few-photon noise level is unprecedented. Using the quantum channel capacity, we also find an expression for the maximum rate at which quantum information can be transduced, providing insight into the importance of improving both a transducer's throughput and noise performance. With feasible improvements, the high throughput achieved with this device positions membrane-based transducers as a strategic choice for demonstrations of a quantum network with reasonable averaging times.

Figures (4)

• Figure 1: Upconversion transducer performance with $B=22$ kHz. The apparent efficiency $\eta$ quantifies the fraction of the incident microwave signal (blue circles) that is transduced to optical frequencies (red circles), rather than lost (empty dashed circles). The upconversion input-referred added noise $N_\text{add}^\text{up}$ (gray dashed circles) compares the noise rate measured at the output $N_\text{out}^\text{up}$ (gray circles) to a potential signal at the input. (a) Apparent transduction efficiency $\eta$ vs. signal detuning from pump $\omega/2\pi$. (b) Output-referred upconversion noise spectrum. (c) Input-referred added noise spectrum in upconversion. The grayed-out regions indicate frequency bands contaminated by noise from the high thermal occupation of optomechanically stiff modes. Excluding these regions and integrating the noise across the bandwidth of the transducer yields $\bar{N}^\text{up}_\text{add}=2.7$ (horizontal dashed line).
• Figure 2: Downconversion transducer performance with $B=19$ kHz. (a) Apparent transduction efficiency $\eta$ vs. signal detuning from pump $\omega/2\pi$. (b) Downconversion output-referred noise spectrum. The broadband noise contribution extending beyond the damped linewidth of the transduction mode is due to unwanted additional occupation of the LC circuit resulting from the microwave pump. Destructive interference between this microwave noise and the mechanical motion it drives results in noise squashing (Eq. \ref{['eq:Nadddown']}). (c) Input-referred added noise spectrum in downconversion. Broadband circuit noise divided by the frequency-dependent $\eta$ results in $N_\text{add}^\text{down}$ increasing with detuning from the center of the transduction bandwidth. Additionally, noise squashing near resonance suppresses the noise level near $\omega_\text{m}/2\pi=1.27$ MHz.
• Figure 3: Transducer performance. Device performance in upconversion (a) and downconversion (b). Maroon diamonds are measurements of our transducer taken while sweeping $\Gamma_\text{e}$ with fixed $\Gamma_\text{o} = 2\pi\cdot11$ kHz in (a), and $\Gamma_\text{o}/2\pi$ roughly fixed in (b), in the range 9-12 kHz. Inconsistency in $\Gamma_\text{o}$ in (b) is due to drifts in the mode matching between the optical pump and the transducer optical cavity over the longer averaging times required for downconversion spectrum measurements. Error bars indicate one standard deviation, and are dominated by statistical uncertainty in the fit parameters used to infer the measurement efficiency. Error bars in the $y$-direction are smaller than the data points. The black line is performance modeled using independently measured parameters. Transducers have nonzero quantum capacity for $N_\text{add}<1$ (light blue region). The quantum capacity of transducers in upconversion (c) and downconversion (d). Measurements presented in this work (maroon diamonds) are compared with other reported results (circles, labeled by first author)mirhosseini2020superconductingmeesala2024nonsahu2022quantumkumar2023quantumxie2025scalableweaver2024integratedbrubaker2022optomechanicaljiang2023opticallyhigginbotham2018harnessingzhao2024quantum . The high throughput measured in downconversion in this work is enabled by our platform's robustness to optical circulating power. Contours indicate $\mathcal{C}_\text{ub}$ as a function of a transducers throughput and $N_\text{add}$, assuming $\eta\ll 1$ and frequency-independent $N_\text{add}(\omega)$.
• Figure 4: Microwave effective mode occupancy. (a) $\bar{n}_\text{e}$ vs. $\Gamma_\text{e}$ as measured electrically (red squares) and optomechanically (blue circles), compared with electrical measurement of a prior device (yellow triangles) brubaker2022optomechanical. The dotted lines are linear fits to the data. The $\Gamma_\text{e}$-dependent noise is reduced by approximately two orders of magnitude relative to the prior device. (b) We represent the electrically measured data as a function of microwave intracavity photon number $n_\text{circ}$ to account for differences in electromechanical coupling $g_\text{e}$ and microwave loss $\kappa_\text{e}$. We attribute the improvement shown in this representation to a reduction of noise photons generated in the annealed Si$_3$N$_4$ film.