WorksPhysics and Astronomy
Photon-Counting Readout via Electro-Optic Microwave-to-Optical Transduction for Axion Haloscopes
Abstract
Axion dark matter searches require highly sensitive detection schemes with high scan rate to efficiently explore the parameter space. Building upon the conventional Sikivie haloscope, where axions convert into microwave photons inside a magnetic-field-immersed resonant cavity, we present a photon-counting scheme based on an electro-optic microwave-to-optical transducer. Through a comprehensive theoretical model under realistic experimental constraints, we find that the optimal operating regime employs an overcoupled microwave cavity, trading peak transduction efficiency for reduced transducer noise. Furthermore, combining parameters already demonstrated in existing devices, in a device with optimized extrinsic coupling rates, would yield a scan-rate enhancement of 5-10 relative to quantum-limited homodyne detection. Improvements in quality factors, coupling rates, and thermal management push this enhancement beyond 50. While our quantitative model is benchmarked at 9 GHz, the scheme adapts well to haloscopes across the 5-12 GHz range, where shrinking cavity volumes severely degrade the achievable scan rate. These results demonstrate the potential of electro-optical photon detection as a high-speed detection technique for axion dark matter searches.
Introduction
Dark matter is the non-luminous, non-absorbing matter that constitutes approximately 84.4% of the universe’s matter and 26.4% of its energy density [1]. Its gravitational effects are essential for understanding cosmic evolution and structure formation. The evidence for dark matter is overwhelming [2, 3, 4, 5, 6, 7], yet its fundamental nature remains unknown. Among the many proposed candidates, including weakly interacting massive particles (WIMPs), sterile neutrinos, axions, axion-like particles, and dark photons [1], axions stand out as particularly compelling. Beyond being a viable dark matter candidate, axions address the long-standing strong CP problem in quantum chromodynamics (QCD): why does the neutron have an electric dipole moment close to zero, despite nonzero values being allowed in the QCD Lagrangian? [8, 9, 10]
Axions could be detected through their coupling to electromagnetism. A broad class of axion search experiments is designed to convert axions into photons within a strong static magnetic field. If this conversion happens inside a resonant cavity, whose resonant frequency matches the axion mass through , the axion signal power can be resonantly enhanced known as a Sikivie haloscope [11]. The axion-photon coupling is parameterized by and is constrained by the solution to the strong CP problem. Because the axion mass is unknown, haloscopes must be tunable to search a wide range of frequencies. In current experiments, the signal power is expected to be – W. Detecting such a weak signal requires a strong magnetic field, a high- cavity in millikelvin temperatures, and low-noise quantum sensors. The scan rate is a crucial figure of merit and is related to various experimental parameters by
where is the external magnetic field, is the effective detection volume, is the loaded quality factor, and is the system noise temperature. The Axion Dark Matter Experiment (ADMX) uses a 136L cavity operating at the TM mode at 1 GHz to search for QCD axions with masses of a few at a scan rate of 543 MHz/year [12]. However, higher masses lead to smaller volume cavities (), more metallic losses (), and increased noise from the Standard Quantum Limit (SQL) (), causing the scan rate to scale as . The same ADMX haloscope scaled to 10 GHz would have a scan rate of Hz/year, which is untenably slow. For this reason, the QCD axion parameter space is largely unexplored above 1 GHz. Yet many theoretical predictions place the axion mass above 2 GHz, strongly motivating the exploration of this parameter space. It is imperative to develop detector and quantum sensing technologies to make this parameter space accessible 1.
Conventional axion haloscopes rely predominantly on microwave amplifiers and homodyne-based readout schemes [19, 20, 21, 22]. For a phase-preserving linear amplifier, the added quantum noise imposes a fundamental noise floor, corresponding to at least one input-referred noise photon [23, 24, 25, 26, 27]. Although single-mode squeezing has been developed to enhance the scan rate beyond the quantum limit, its practical performance is constrained by finite squeezing and degradation of the squeezed state during amplification and transmission [28, 29, 30]. Photon-counting detection provides a different route that can circumvent the added-noise constraint associated with amplification, since it measures photon number rather than field quadratures [31, 32]. In the microwave domain, however, photon-counting schemes are challenged by limited detection efficiency, bandwidth, and relatively high dark count rates [33, 34, 35]. Microwave-to-optical quantum transduction offers an alternative approach that could potentially overcome these limitations by transferring the microwave photons to the optical regime [36, 37, 38, 39, 40]. Once converted, the signal is read out using superconducting nanowire single-photon detectors (SNSPDs), which provide exceptional detection efficiency with an ultra-low dark count rate [18, 41].
In this work, we propose and analyze a readout scheme for axion haloscopes based on microwave-to-optical transduction followed by optical photon counting. We evaluate the scheme using an electro-optic transducer designed at 9 GHz as a benchmark, although it can be adapted to other haloscope frequencies by redesigning the transducer’s microwave resonator. Our theoretical model incorporates key experimental parameters of the transducer, including the single-photon coupling rate, the quality factors (intrinsic loss rates), microwave and optical extrinsic coupling rates, pump power, and repetition rate. Furthermore, the model accounts for noise contributions from the sensing cavity, the transducer microwave mode, and detector dark counts. We investigate the trade-off between the transduction efficiency and total noise, and optimize the extrinsic coupling coefficients to maximize the scan-rate enhancement with respect to homodyne-detection baseline. Under benchmark experimental conditions, we find that enhancements of are possible with parameters demonstrated in existing devices, operating with optimized extrinsic coupling rates. Furthermore, enhancements in excess of 50 are accessible when pushing experimental conditions to their limits. These results highlight the potential of low-loss, low-noise, and high-repetition-rate electro-optic transducers for rapid axion searches and provide a promising pathway toward high-speed axion experiments in high frequencies.
Axion signal model
In a haloscope setup, axions from the galactic dark matter halo are converted into microwave photons inside a resonant cavity permeated by a strong static magnetic field , via the inverse Primakoff effect. In natural units, the expected signal power is given by [26]:
where is the axion-photon coupling constant, is the local dark matter density, is the axion mass, is the effective cavity volume including the geometric form factor, is the loaded quality factor of the cavity and is the coupling coefficient of sensing cavity. Here, is the extrinsic coupling rate, is the intrinsic loss rate and is the Lorentzian lineshape function given by
where the dimensionless frequency detuning is defined as . Here, and denote the axion frequency and the cavity’s resonance frequency, respectively, while represents the cavity’s bandwidth. Due to the virialized velocity dispersion of dark matter in the galactic halo, the axion signal exhibits a characteristic Maxwell-Boltzmann-like lineshape centered at frequency , with a relative bandwidth of (corresponding to a quality factor of ) [42]. To incorporate this weak signal into our quantum input-output framework, we describe the axion as a coherent drive in sensing cavity with annihilation operator . The intracavity field operator evolves according to [23, 43]
where is the angular resonance frequency of the cavity, represents the total cavity decay rate, and is the axion-photon coupling rate. The operators and denote the input vacuum and noise field from the transmission line and the intrinsic bath noise, respectively. Solving for the steady state (see Appendix A), we obtain the output state from the sensing cavity as a Gaussian state with mean vector and covariance matrix given by
where is the photon number of axion, is the phase of the axion-induced microwave field and is the thermal photon number of sensing cavity.
Transducer-based axion sensing
In this section, we present a comprehensive theoretical analysis of axion readout scheme based on the microwave-to-optical transduction and benchmark its performance with microwave homodyne detection. We trace the signal and noise propagation across the conversion and photon detection, and evaluate the resulting scan rate and the scan-rate enhancement relative to homodyne detection.
Figure 1. Schematic of the transducer-based axion sensing scheme. A strong magnetic field in the sensing cavity generates axion-induced microwave photons, which are routed to a transducer cavity and converted into optical photons through an optical pump and a nonlinear crystal for detection with a superconducting nanowire single-photon detector (SNSPD).
Signal and Noise Transfer
Assuming a steady-state conversion process, the optical output field operator can be obtained from the transducer’s dynamical equations, together with the corresponding input-output relations [44, 45]:
Here, we adopt a systematic subscript convention: the primary index (o or m) denotes the subsystem (optical or microwave), whereas the secondary index (c or i) denotes the extrinsic coupling and intrinsic loss channels, as summarized in Table II of Appendix B 2. and are the optical and microwave detunings with respect to the output signals. The total decay rate of each mode is the sum of these two contributions, and , and is the pump-enhanced coupling rate, where is the intracavity pump photon number and is the single-photon electro-optic coupling rate. From the derivations presented in Appendix B 2, the microwave-to-optical conversion efficiency and the associated added noise are given by
| Symbol | Definition |
| Optical mode | |
| Output optical frequency | |
| Extrinsic coupling rate | |
| Intrinsic quality factor | |
| Intrinsic loss rate, | |
| Total damping rate, | |
| Detuning, | |
| Microwave mode | |
| Microwave mode frequency | |
| Output microwave mode frequency | |
| Extrinsic coupling rate | |
| Intrinsic quality factor | |
| Intrinsic loss rate, | |
| Total damping rate, | |
| Detuning, | |
| Electro-optic coupling | |
| Single-photon coupling rate | |
| Pump photon number | |
| Effective coupling strength, |
Table II. Definitions of the parameters used in the transducer model.
where is the cooperativity, and are extraction efficiencies of the optical and microwave modes, respectively, and denotes the effective thermal occupancy of the transducer’s microwave mode. The quantum state by quantum transduction is given by
where the transducer noise photon is given by
Scan-Rate Enhancement Relative to Homodyne Detection
The ultimate performance of a dark matter haloscope experiment is governed by its capability to scan a broad frequency range for axion masses. Having established the signal and noise transfer of the transducer-based readout chain, we now formulate the signal-to-noise ratio (SNR), the scan rate, and its performance enhancement relative to microwave homodyne detection.
A detection event occurs when one or more photons are registered on the SNSPD. For a quantum transducer operating at a photon-counting measurement rate , a total of individual photon detection cycles are performed over an integration time . For a pulsed pump, is the repetition rate, whereas for a continuous pump, . As presented in Appendix § 10 2, the SNR is given by
where is the efficiency of the single photon detector, is the dark count photon number per detection window. We next consider a frequency scan over a total range , discretized into frequency steps with a step size . Following the derivation detailed in Appendix § 10 2, the resulting scan rate required to achieve a target total signal-to-noise ratio is given by
where is the peak axion-induced photon number . For conventional microwave homodyne detection, the SNR and scan rate are derived in Appendix C 1 as:
where is the resolution bandwidth of the spectrum analyzer, and is the input-equivalent noise photon number of system given by Friis formula (See Eq. (66) of Appendix B 1)
where and are the peak and base noise photon numbers, respectively. Here, and denote the noise contributions from the Josephson parametric amplifier (JPA) and the high electron mobility transistor (HEMT) amplifier, respectively, with being the JPA gain. Combining Eq. (17) and Eq. (19), the enhancement in scan rate with the same total SNR is given by:
Performance Analysis
Evaluating the performance enhancement requires knowledge of both the transducer efficiency and noise, both of which depend on the cooperativity . In this section, we first express the cooperativity in terms of the intrinsic cooperativity and the tunable coupling coefficients . Then we summarize the transducer’s device and noise parameters at 9 GHz across a realistic range of values in Table I. After setting up the benchmark parameters, we maximize the enhancement by optimizing the tunable coupling coefficients and characterize the trade-off between the transducer efficiency and the noise governed by the microwave coupling coefficient . Furthermore, we explore the scan-rate enhancement of transducer-based photon counting over conventional homodyne detection across key operational parameters, including the single-photon coupling rate , the microwave quality factor , the pump power , and the thermal noise (). Our analysis considers two operating schemes of quantum transducer: continuous-wave pumping, which provides low added noise but is limited by the transduction efficiency, and pulsed pumping, which offers higher efficiency at the cost of increased noise and a lower repetition rate.
| Parameter | Symbol | Value |
| Axion Sensing Cavity | ||
| Sensing resonance frequency | 9.0 GHz | |
| Loaded quality factor | ||
| Axion quality factor | ||
| extrinsic coupling ratio | 2.0 | |
| Cavity temperature | 50 mK | |
| Electro-Optic Transducer | ||
| Pump frequency | 192.421 THz | |
| Optical resonance frequency | 192.43 THz | |
| Microwave resonance frequency | 9.0 GHz | |
| Optical quality factor | ||
| Microwave quality factor | – | |
| Single-photon coupling rate | 40–140 Hz | |
| Pump detuning | 10 MHz | |
| Continuous-wave operation | ||
| Pump power | 100–800 W | |
| Transducer temperature | 100–300 mK | |
| Photon counting measurement rate | 9.0 kHz | |
| Pulsed operation | ||
| Pump power | 50–300 mW | |
| Microwave noise photon | 0.01–0.1 | |
| Intrinsic cooperativity | 1–10 | |
| Repetition rate | 100 Hz | |
| Single Photon Detector | ||
| Single-photon detection efficiency | 0.98 | |
| Dark photon counting rate | 1 Hz | |
| Homodyne Detection | ||
| Frequency bin resolution | 3 kHz | |
| Total temperature | 100 mK |
Table I. Benchmark parameters for the transducer-based axion haloscope. Values are chosen based on experimental capabilities for searches near 9 GHz
Parametric Characterization of Cooperativity
The overall system performance is dictated by the conversion efficiency and noise . A critical engineering trade-off arises from the dual role of the extrinsic coupling rates . High extrinsic coupling rates are essential extracting the converted signals. On the other hand, increasing the extrinsic coupling also increases the total damping rate , which in turn degrades the cooperativity since . We parameterize the cooperativity using the tunable coupling coefficients and the intrinsic cooperativity , which depends on the intra-cavity pump photon number , single-photon coupling rate and intrinsic loss rates. Using standard input-output theory, is directly related to the input pump power .
For electro-optic transducers, the intrinsic loss rates are set by the intrinsic quality factors of the two resonators, and , where and are the operating angular frequencies and the intrinsic quality factor of the optical and microwave modes, respectively. These parameters are determined by material absorption, surface scattering, and fabrication quality, and are independent of the optical pump during normal operation, except in high-power regimes where pump-induced losses will arise. In contrast, the extrinsic coupling rates can be engineered and optimized, for instance, by designing the waveguide-cavity spacing or antenna geometry. We parameterize the tunable extrinsic coupling rates using the optical and microwave coupling coefficient defined by and . Consequently, the extraction efficiency is expressed as . We define the intrinsic cooperativity
The electro-optic cooperativity is expressed as:
Electro-optic transducers are commonly operated with pulsed optical pumps. The cooperativity is proportional to pump photon number , whereas heating and thus noise follows the average power. A low duty cycle therefore buys high cooperativity within fixed cooling budget, a strategy applied in bulk [38, 39, 46] and thin-film [47, 48] lithium niobate as well as in aluminum nitride [49]. For example, in the high-cooperativity regime demonstrated in Ref. [38], the intrinsic cooperativity reaches for , with and . Consequently, we consider a realistic range of intrinsic cooperativity for our performance analysis. Both in the steady-state regime (under continuous pumping) and the quasi-steady-state regime (where the pulse duration exceeds the cavity ring-up time), the intrinsic cooperativity can be further parameterized by the single-photon coupling rate and the pump power . The single-photon coupling rate is determined by the field overlap between the microwave mode and optical mode. Based on the steady-state input-output relation, for a given pump power , the intra-cavity pump photon number is expressed as [40]
where is the pump detuning and is the angular frequency of the pump field. Therefore, given the intrinsic loss rates , single-photon coupling rate , pump power , and their respective resonance frequencies, the resulting cooperativity depends on the tunable coupling coefficients to be optimized.
Enhancement Regime for Electro-Optic Transducers
To evaluate the practical performance of the transducer-based axion haloscope, we consider an axion search targeting a frequency around 9 GHz. Table I summarizes the benchmark parameters in this subsection.
The axion signal is evaluated at a benchmark frequency of GHz with a bandwidth of kHz. The effective temperature of the sensing cavity has been reported to be below 60 mK in Ref. [30]. We adopt mK, within this range and consistent with the temperatures routinely achieved in dilution refrigerators hosting haloscope cavities. The thermal photon number in the sensing cavity is then determined by the Bose-Einstein distribution:
For conventional homodyne detection, the effective noise photon number for standard microwave homodyne detection is defined by
The dominant contribution comes from the JPA. We evaluate (i.e., in Eq. (20)) assuming a total effective noise temperature of 100 mK for the microwave components. The frequency resolution is typically chosen to be 1/3 to 1/5 of the axion bandwidth kHz. We set kHz for homodyne detection.
Continuous-wave pump operation
For a continuous-wave (CW) electro-optic transducer operating in steady-state conversion, the total number of modes is given by the time-bandwidth product over an integration time . Thus the photon-counting
measurement rate . The average dark count per mode is given by for a dark count rate of Hz. A pump power in the order 100 W yields lower added noise (i.e., the effective temperature ) but also a lower transduction efficiency than pulsed operation [46, 47, 48]. Pump detuning is essential to mitigate photothermal instabilities and optical bistability caused by high-power pump absorption. Accordingly, we set the pump detuning to MHz as a representative experimental benchmark [40]. The optical quality factor is set to . The microwave quality factor is – for on-chip resonators and projected – for state-of-the-art 3D hybrid microwave cavities [40, 50, 51]. In steady-state conversion, the relation between and follows the Bose-Einstein distribution in Eq. (25). The steady-state input-output relation directly establishes the relation between the input pump power and the intra-cavity photon number, uniquely determining and . In contrast, pulsed operation outside the quasi-steady-state regime requires solving the time-dependent intra-cavity dynamics or utilizing an effective .
With mK, Hz, and W, optimal choices of emerge from the trade-off between the overall transduction efficiency and the noise . Through numerical simulations, we find a peak enhancement of at and , with a transducer efficiency of . With fixed at its optimum, we show the trade-off versus in Figure 2. Existing microwave-to-optical transducers were designed for quantum networking and qubit readout, where the goal is to maximize the conversion efficiency. They therefore operate near critical microwave coupling. The axion photon-counting scheme analyzed here instead favors a strongly overcoupled microwave port. We observe that a scan-rate enhancement is obtained at a relatively low transduction efficiency. This is because photon counting is limited by noise photon rather than by the standard quantum limit (), i.e., a weak signal can still be detected with a favorable signal-to-noise ratio.

Figure 2. Trade-off between transduction efficiency and noise in the continuous-wave regime. Transduction efficiency (red solid), transducer noise (blue dashed), and scan-rate enhancement (black solid) versus the microwave coupling ratio , with the optical coupling fixed at its optimal value . Each curve is normalized to its own maximum (, , ). Vertical dashed lines mark the maxima of () and of (); the shaded band indicates the displacement between them. Here, mK and , Hz, .
Beyond these benchmark values, we evaluate the enhancement , maximized over for the parameters listed in Table I. Figure 3(a) displays the enhancement over a range of single-photon coupling rate and microwave quality factor , while Figure 3(b) shows the enhancement over a range of transducer temperature and pump power , with optimized over at each point. The 3D microwave cavity transducer design in Ref. [40] features quality factors of – with a relatively small single-photon coupling rate of Hz, placing its performance in the bottom-right regime of Figure 3(a) with . Conversely, on-chip thin-film lithium niobate transducers experiments in Ref. [48] offer a complementary route, exhibiting a much larger single-photon coupling rate ( Hz) but lower quality factors (–). Consequently, such devices operate in the upper-left regime of Figure 3(a), yielding at . Although Figure 3(b) suggests that both a higher pump power and a lower transducer temperature increase the enhancement, these two parameters are not independent. A larger pump power raises the effective mode temperature through optical absorption and quasiparticle generation, with the precise dependence determined by the thermal anchoring of the device. In practice, current on-chip demonstrations are limited by fiber-to-chip losses and by the pump power they can sustain.
Figure 3. Scan-rate enhancement compared to standard homodyne detection under continuous-wave pumping. Dashed lines: the corresponding enhancement . (a) Contour map of the optimal enhancement as a function of the and , at a fixed temperature mK and pump power . (b) Optimal enhancement versus and , at a fixed and Hz.
Pulsed pump operation
While CW operation is constrained by steady-state heating, pulsed pumping allows high pump photon number for high cooperativity. Pulsed operation relaxes the thermal budget of the device: a peak pump power of mW, applied at a duty cycle of 1–5%, dissipates only mW of average optical power at the mixing chamber while maintaining a high intra-cavity pump photon number , sufficient for larger cooperativity and hence near unity internal efficiency . Pump-induced heating means the mode occupancy relaxes only partially within one cycle and does not reach its steady-state ( mK) value. In experiments, the thermal photon number is typically on the order of 0.1 at frequencies around 3–9 GHz [38, 39, 46, 47, 48, 49, 52, 53]. Specifically, evaluated at 9 GHz, the effective input noise photon number falls within the range –0.1. Since the bandwidth of transducer (–10 MHz) is much larger than the signal bandwidth , the transducer responds essentially instantaneously. The pulse duration and detection time window is , with dark photon number for 1 Hz dark photon counting rate. For the numerical evaluation for pulsed pump operation, we fix the dark photon number within this detection window and the repetition rate at Hz.
We now repeat the optimization for pulsed operation,
using the parameters of Table I. With intrinsic cooperativity and noise photon number , Figure 4(a) shows the enhancement as a function of . We find a peak enhancement of at and , for which the total transducer efficiency is . The optimum again lies in the overcoupled regime. The large pump power lifts the internal efficiency, leaving more efficiency to trade away; the optimal therefore moves deeper into the overcoupled regime, where is further suppressed. Figure 4(b) shows the enhancement obtained by maximizing over the coupling ratios at each point of the plane. The bulk-LiNbO transducer of [38] reaches cooperativities up to , corresponding to . For a repetition rate of 100 Hz, the enhancement ranges from for to for . Such an occupancy has been demonstrated on the same platform at lower pump power in Ref. [53].

Figure 4. (a) Trade-off between transduction efficiency and noise in the pulsed-pump regime. Transduction efficiency (red solid), transducer noise (blue dashed), and scan-rate enhancement (black solid) versus the microwave coupling ratio , with the optical coupling fixed at its optimal value . Each curve is normalized to its own maximum (, , ). Here and . (b) Contour map of the enhancement as a function of the intrinsic cooperativity and transducer noise , at a fixed repetition rate Hz.
Within the quasi-steady-state approximation, the cooperativity can be expressed in terms of the single-photon coupling rate , the pump power, and the intrinsic loss rates (i.e., the quality factors). Over the parameter ranges listed in Table I, we evaluate the scan-rate enhancement. Figure 5(a) maps the optimal enhancement over the plane, while Figure 5(b) shows it over the plane, with optimized independently at each point. The enhancement increases with a higher single-photon coupling rate and quality factor , exceeding in Figure 5(a) when Hz and . Similarly, it exceeds 50 in Figure 5(b) when and mW.
Figure 5. Scan-rate enhancement compared to standard homodyne detection under pulsed pumping. Dashed lines: the corresponding enhancement . (a) Contour map of the optimal enhancement as a function of the and , at a fixed noise photon and pump power mW. (b) Optimal enhancement versus and , at a fixed and Hz.
Discussion
We have shown that converting axion-induced microwave photons to the optical domain and counting them can substantially outperform conventional homodyne readout. The scan-rate enhancement originates from that photon counting scheme avoids the standard quantum limit of phase-preserving amplification. A central result of our analysis is that the optimal operating point is not the one that maximizes the transduction efficiency. Because the figure of merit is set by the balance between converted signal and transducer noise, the optimum lies in the overcoupled regime, at a microwave coupling ratio well above the value that maximizes . The efficiency is traded away in exchange for a stronger suppression of the transducer noise. Current transduction experiments have not been developed for this specific application and operate in the near critical coupling regime to maximize efficiency. Still, the key ingredients for photon counting applications have been demonstrated on various platforms, both on-chip and 3D designs. The main challenge is to combine high cooperativity with low microwave occupancy, and optimal optical coupling to sustain this regime over long pump pulses.
Looking ahead, the scheme is attractive in high frequency, where broadband quantum-limited phase-preserving amplification remains challenging above 10 GHz. The electro-optic interaction itself is not a limitation in this range. The quantum transducers can be engineered to address microwave modes ranging from a few to ten GHz, provided that the microwave resonance matches the free spectral range of the optical cavity. The transducer bandwidth of 1–10 MHz set by the microwave linewidth then fully covers the signal linewidth. And the central conversion frequency can be tuned over 100 MHz by tuning the microwave cavity itself, allowing the transducer to follow the haloscope cavity as it is stepped.
While our model is specific to electro-optic transduction, the framework may be extended to other platforms, such as piezo-optomechanical transducers, by including the corresponding noise and efficiency contributions of the intermediate mechanical mode. Our proposed scheme could be further enhanced using two-mode squeezing input state and joint detection receiver as proposed in [54, 55, 56, 57]. Combining the readout with multi-cavity arrays or higher-order-mode cavities would allow the enhancement demonstrated here to multiplicatively compound with these architectures, translating into a massive boost in the scan rate and providing a concrete pathway toward high-speed axion searches at the MHz/day level in the 5–12 GHz range.
Acknowledgments
This manuscript has been authored by the Fermi Forward Discovery Group, LLC under Contract No. 89243024CSC000002 with the U.S. Department of Energy, Office of Science, Office of High Energy Physics. The authors acknowledge support from the DOE Office of Science Early Career Research Program, Office of High Energy Physics (HEP), Detector R&D and QIS programs, and the Fermilab Laboratory Directed R&D (LDRD) program. R.C. acknowledges support from the U.S. Department of Energy, Office of Science, National Quantum Information Science Research Centers, Superconducting Quantum Materials and Systems Center (SQMS).
Data Availability
The results presented in this work can be reproduced using the model and equations provided in the main text and Appendices, together with the parameters listed in Table I. The code used to generate the figures is available from the authors upon reasonable request.
Axion state in sensing cavity
The axion-induced field amplitude is expressed as
where is the eigenfrequency of the axion field. The coupling between the axion field and the electromagnetic field is described by the axion-photon interaction term in the Lagrangian
where is the axion-photon coupling constant, is the electromagnetic field tensor, and is its dual. Under the background magnetic field approximation where , keeping the first-order part of expression, the interaction Hamiltonian is given by
The cavity mode expansion and the standard second quantization in natural units can be written as
where is the effective volume of the mode and . Applying the rotating-wave approximation by neglecting the rapidly oscillating counter-rotating terms and , the interaction Hamiltonian can be simplified to
where the effective coupling strength is given by
Here, quantifies the effective spatial overlap governing the axion-photon conversion. In natural units, taking typical values of and a background magnetic field of (corresponding to 10 T), the coupling rate evaluates to (approximately ). From , the coupling between the axion field and the cavity mode is given by . Now, we consider a single-port (reflective) sensing cavity. The dynamics of the cavity mode is governed by the Heisenberg-Langevin equation in the time domain:
where is the resonant frequency of the cavity, and represent the input field and intrinsic noise operators, respectively, and , with and denoting the intrinsic and extrinsic decay rates. To derive the steady-state transfer function, we transform the equations into the frequency domain using the Fourier convention . Hereafter, we work exclusively in the frequency domain unless stated otherwise. The resulting equation of motion and the corresponding input-output relation are given by:
where . From the above equations, the output mode in frequency domain is given by
where the transfer coefficients are given by
Treating the axion field as a classical driving source within the cavity, the resulting steady-state photon field forms a coherent state satisfying . The input noise modes are characterized by the vacuum state for and the thermal state for with the correlation function:
where the mean thermal photon number is given by the Bose-Einstein distribution
In the absence of a thermal background, the power of axion signal can be derived from . From ,
For stationary process , where is the amplitude of coherent state. The power
Here, we treat the axion signal as a narrow-band process centered around with an effective bandwidth and define
where represents the average spectral density. In experiments, the relevant parameters are typically defined as
where is the coupling coefficient, is the loaded quality factor, and represents the normalized detuning. Substituting these definitions into Eq. (44), we obtain the detected power:
Comparing Eq. (49) with the axion haloscope power formula,
we establish the relation for the signal amplitude:
The real quadrature operators are defined as and , satisfying the commutation relation . For a coherent state , the corresponding displacement vector (mean values) and the corresponding covariance matrix can be obtained from above quadrature definitions:
where , denotes the anti-commutator, and is the identity matrix. From Eq. (37), we can derive the transformation of the mean vector and covariance matrix of the output state. Since , , and , then
Since is narrow-band process compared with the sensing cavity, . By Eq. (44),
where . Furthermore, since , the output covariance matrix becomes
where is the Lorentzian shape function.
State transfer
Phase-Insensitive Amplifier and Noise Performance
The input-output relation for a general linear amplifier and a Josephson parametric amplifier (JPA) can be expressed as [23, 24]
where and denotes the environmental annihilation operator. The thermal state of the environment is characterized by its mean thermal photon number,
For a phase-insensitive broadband JPA, the covariance matrix of the environment state is given by . Consequently, the input-output mapping constitutes a Gaussian channel, yielding the output covariance matrix and displacement vector:
Defining the input-referred noise via the relation , we obtain
To account for the subsequent amplification stages, we consider a concatenated high-electron-mobility transistor (HEMT) amplifier. The resulting output covariance matrix and displacement vector become
In the limit where , the input-referred noise simplifies to
By induction, the Friis formula for cascaded linear amplifiers, including the JPA, HEMT, and subsequent mixer, yields the comprehensive input-referred noise:
The total gain . Following this cascaded amplification chain, the final amplified signal state is characterized by the mean displacement vector and covariance matrix:
Finally, we note that the quantum limit arises from two contributions in Eq. (61): the input quantum fluctuations and the amplifier fluctuations . By employing a squeezed input and a corresponding phase-sensitive amplifier, one can achieve but . This approach is known as squeezing-enhanced sensing.
Quantum transduction
Assuming a steady-state conversion process, the input-output relations for electro-optical transduction in the frequency domain are given by:
where the variables and coefficients characterizing the transducer model are comprehensively summarized in Table II, categorized by the optical mode, microwave mode, and electro-optic coupling. Solving the quantum Langevin equations yields the optical output mode as
where , and . The conversion from to is determined by
To maximize the conversion efficiency, the detunings are set to , yielding an efficiency of
where , and . Setting and respectively, the conversion bandwidths for the microwave and optical channels are given by
Since these conversion bandwidths are typically greater than —much larger than the typical signal bandwidth—the conversion efficiency can be treated as constant for the signal.
At optical frequencies and room temperature, the thermal occupation of the optical input and intrinsic noise modes is negligible (), whereas for the microwave noise mode, the correlation is given by
where represents the thermal photon number. The add noise photon from is given by
Given that the input microwave mode has the mean and covariance matrix specified in Eqs. (56) and (57), the mean and covariance matrix of the output quantum state follow directly from their definitions and are given by:
where the noise photon number is given by
Signal-to-noise ratio and scan rate
Homodyne detection
In homodyne detection, a single frame of the power spectrum is obtained via an analog-to-digital converter (ADC) and a fast Fourier transform (FFT). Each spectrum has a time duration of , where is the spectral bin resolution. To suppress statistical noise, independent frames are collected, yielding a total integration time of
According to the central limit theorem, the sample mean of the power spectrum, , approximately follows a normal distribution:
When axion signal presents, the output state is a Gaussian state characterized by the mean displacement vector and covariance matrix:
Assume that the phase is uniformly distributed over the averaging interval, the moments of the quadratures are given by
Since and , the signal-to-noise ratio (SNR) simplifies to
According to (67), the system’s input-referred noise exhibits Lorentzian frequency dependence due to . Therefore, we can write the noise as
where is the function of Lorentzian line shape and . Consider a scan process with a frequency step size of , where is the total number of scanning steps. The total SNR is given by
Here, we have converted the discrete summation into an integral approximation, which is valid for . And we assume a broad scan range such that . Given that the scanning rate is defined as , the total squared SNR can be rewritten as
Substituting and (91), we obtain
Photon detection
A superconducting nanowire single-photon detector is not photon-number resolving: it registers a click whenever one or more photons are absorbed. The natural quantity is therefore the complementary no-click probability. Since the signal absorption and the dark-count process are statistically independent, then
The signal is a Gaussian state with displacement and covariance matrix . A detector of efficiency is modeled as a pure loss channel, which maps and . Then
Dark counts form an independent Poisson process with
where is the dark counting rate and is the time window. A mode of bandwidth is resolved in a time window , so the number of dark counts per measurement window is and . Combining these two factors,
With and , the probability of click is given by
For a sufficiently long integration time , the number of independent modes is , where in the continuous-pump regime (set by the signal coherence bandwidth), and is the pulse repetition rate in the pulsed-pump regime. The random variable has mean and variance
The signal-to-noise ratio is given by
Consider a scan process with a frequency step size of , where is the total number of scanning steps. Similar to Eq. (92), the total SNR is given by
References
- [1]R. Workman, M. Moskovic, M. Carena, D. Robinson, T. DeGrand, J. Anderson, G. Venanzoni, K. Black, R. Mitchell, T. Sjöstrand, et al., Progress of Theoretical and Experimental Physics 2022, 083C01 (2022), https://academic.oup.com/ptp/article-pdf/2022/8/083C01/49175539/ptac097.pdf.
- [2]V. C. Rubin, W. K. Ford, Jr., N. Thonnard, and D. Burstein, Astrophys. J. 261, 439 (1982).
- [3]K. G. Begeman, A. H. Broeils, and R. H. Sanders, Monthly Notices of the Royal Astronomical Society 249, 523 (1991), https://academic.oup.com/mnras/article-pdf/249/3/523/18160929/mnras249-0523.pdf.
- [4]A. N. Taylor, S. Dye, T. J. Broadhurst, N. Benitez, and E. van Kampen, The Astrophysical Journal 501, 539 (1998).
- [5]P. Natarajan, U. Chadayammuri, M. Jauzac, J. Richard, J.-P. Kneib, H. Ebeling, F. Jiang, F. van den Bosch, M. Limousin, E. Jullo, H. Atek, A. Pillepich, C. Popa, F. Marinacci, L. Hernquist, M. Meneghetti, and M. Vogelsberger, Monthly Notices of the Royal Astronomical Society 468, 1962 (2017), https://academic.oup.com/mnras/article-pdf/468/2/1962/11210742/stw3385.pdf.
- [6]M. Markevitch, A. H. Gonzalez, D. Clowe, A. Vikhlinin, W. Forman, C. Jones, S. Murray, and W. Tucker, The Astrophysical Journal 606, 819 (2004).
- [7]N. Aghanim, Y. Akrami, M. Ashdown, J. Aumont, C. Baccigalupi, M. Ballardini, A. J. Banday, R. B. Barreiro, N. Bartolo, and et al., Astronomy & Astrophysics 641, A6 (2020).
- [8]R. D. Peccei and H. R. Quinn, Phys. Rev. Lett. 38, 1440 (1977).
- [9]S. Weinberg, Phys. Rev. Lett. 40, 223 (1978).
- [10]F. Wilczek, Phys. Rev. Lett. 40, 279 (1978).
- [11]P. Sikivie, Phys. Rev. Lett. 51, 1415 (1983).
- [12]N. Du, N. Force, R. Khatiwada, E. Lentz, R. Ottens, L. J. Rosenberg, G. Rybka, G. Carosi, N. Woollett, D. Bowring, A. S. Chou, A. Sonnenschein, W. Wester, C. Boutan, N. S. Oblath, R. Bradley, E. J. Daw, A. V. Dixit, J. Clarke, S. R. O’Kelley, N. Crisosto, J. R. Gleason, S. Jois, P. Sikivie, I. Stern, N. S. Sullivan, D. B. Tanner, and G. C. Hilton (ADMX Collaboration), Phys. Rev. Lett. 120, 151301 (2018).
- [13]B. Ary dos Santos Garcia, D. Bergermann, A. Caldwell, V. Dabhi, C. Diaconu, J. Diehl, G. Dvali, J. Egge, E. Garutti, S. Heyminck, F. Hubaut, A. Ivanov, J. Jochum, S. Knirck, M. Kramer, D. Kreikemeyer-Lorenzo, C. Krieger, C. Lee, D. Leppla-Weber, X. Li, A. Lindner, B. Majorovits, J. P. A. Maldonado, A. Martini, A. Miyazaki, E. Öz, P. Pralavorio, G. Raffelt, J. Redondo, A. Ringwald, J. Schaffran, A. Schmidt, F. Steffen, C. Strandhagen, I. Usherov, H. Wang, and G. Wieching (MADMAX Collaboration), Phys. Rev. Lett. 135, 041001 (2025).
- [14]R. Cervantes, G. Carosi, S. Kimes, C. Hanretty, B. H. LaRocque, G. Leum, P. Mohapatra, N. Oblath, R. Ottens, Y. Park, et al., Physical Review D 106, 102002 (2022).
- [15]R. Cervantes, G. Carosi, C. Hanretty, and et al., Phys. Rev. Lett. 129, 201301 (2022).
- [16]A. J. Millar, S. M. Anl͟age, R. Balafendiev, P. Belov, K. van Bibber, J. Conrad, M. Demarteau, A. Droster, K. Dunne, A. G. Rosso, J. E. Gudmundsson, H. Jackson, G. Kaur, T. Klaesson, N. Kowitt, M. Lawson, A. Leder, A. Miyazaki, S. Morampudi, H. V. Peiris, H. S. Røising, G. Singh, D. Sun, J. H. Thomas, F. Wilczek, S. Withington, M. Wooten, J. Dilling, M. Febbraro, S. Knirck, and C. Marvinney (Endorsers), Phys. Rev. D 107, 055013 (2023).
- [17]T. A. Dyson, C. L. Bartram, A. Davidson, J. B. Ezekiel, L. M. Futuraura, T. Liu, and C.-L. Kuo, Phys. Rev. Appl. 21, L041002 (2024).
- [18]J. Chiles, I. Charaev, R. Lasenby, M. Baryakhtar, J. Huang, A. Roshko, G. Burton, M. Colangelo, K. Van Tilburg, A. Arvanitaki, S. W. Nam, and K. K. Berggren, Phys. Rev. Lett. 128, 231802 (2022).
- [19]P. Sikivie, D. B. Tanner, and Y. Wang, Phys. Rev. D 50, 4744 (1994).
- [20]C. Hagmann, P. Sikivie, N. S. Sullivan, and D. B. Tanner, Phys. Rev. D 42, 1297(R) (1990).
- [21]C. Hagmann, D. Kinion, W. Stoeffl, K. van Bibber, E. Daw, H. Peng, L. J. Rosenberg, J. LaVigne, P. Sikivie, N. S. Sullivan, D. B. Tanner, F. Nezrick, M. S. Turner, D. M. Moltz, J. Powell, and N. A. Golubev, Phys. Rev. Lett. 80, 2043 (1998).
- [22]L. Zhong, S. Al Kenany, K. M. Backes, B. M. Brubaker, S. B. Cahn, G. Carosi, Y. V. Gurevich, W. F. Kindel, S. K. Lamoreaux, K. W. Lehnert, S. M. Lewis, M. Malnou, R. H. Maruyama, D. A. Palken, N. M. Rapidis, J. R. Root, M. Simanovskaia, T. M. Shokair, D. H. Speller, I. Urdinaran, and K. A. van Bibber, Phys. Rev. D 97, 092001 (2018).
- [23]A. A. Clerk, M. H. Devoret, S. M. Girvin, F. Marquardt, and R. J. Schoelkopf, Rev. Mod. Phys. 82, 1155 (2010).
- [24]C. M. Caves, Phys. Rev. D 26, 1817 (1982).
- [25]M. Hatridge, R. Vijav, D. H. Shictter, J. Clarke, and I. Siddiqi, Phys. Rev. B 83, 134501 (2011).
- [26]S. J. Astalos, G. Carosi, C. Hagmann, D. Kinion, K. van Bibber, M. Hotz, L. J. Rosenberg, G. Rybka, J. Hoskins, J. Hwang, P. Sikivie, D. B. Tanner, R. Bradley, and J. Clarke, Phys. Rev. Lett. 104, 041301 (2010).
- [27]M. M. Brubaker, L. Zhong, Y. V. Gurevich, S. B. Cahn, S. K. Lamoreaux, M. Simanovskaia, J. R. Root, S. M. Lewis, S. Al Kenany, K. M. Backes, I. Urdinaran, N. M. Rapidis, T. M. Shokair, K. A. van Bibber, D. A. Palken, M. Malnou, W. F. Kindel, M. A. Anil, K. W. Lehnert, and G. Carosi, Phys. Rev. Lett. 118, 061302 (2017).
- [28]M. Malnou, D. A. Palken, B. M. Brubaker, L. R. Vale, G. C. Hilton, and K. W. Lehnert, Phys. Rev. X 9, 021023 (2019).
- [29]K. M. Backes, D. A. Palken, S. A. Kenany, B. M. Brubaker, S. B. Cahn, A. Droster, G. C. Hilton, S. Ghosh, H. Jackson, S. K. Lamoreaux, A. F. Leder, K. W. Lehnert, S. M. Lewis, M. Malnou, R. H. Maruyama, N. M. Rapidis, M. Simanovskaia, S. Singh, D. H. Speller, I. Urdinaran, L. R. Vale, E. C. van Assendelft, K. van Bibber, and H. Wang, Nature 590, 238 (2021).
- [30]X. Bai, M. J. Jewell, J. Echevers, K. van Bibber, A. Droster, M. H. Esmat, S. Ghosh, E. Graham, H. Jackson, C. Laffan, S. K. Lamoreaux, A. F. Leder, K. W. Lehnert, S. M. Lewis, R. H. Maruyama, R. D. Nath, N. M. Rapidis, E. P. Ruddy, M. Silva-Feaver, M. Simanovskaia, S. Singh, D. H. Speller, S. Zacarias, and Y. Zhu (HAYSTAC Collaboration), Phys. Rev. Lett. 134, 151006 (2025).
- [31]C. Braggio, L. Balembois, R. Di Vora, Z. Wang, J. Traverso, P. Pallegioix, G. Carugno, A. Ortolan, G. Ruoso, U. Gambardella, D. D’Agostino, P. Bertet, and E. Flurin, Phys. Rev. X 15, 021031 (2025).
- [32]A. V. Dvïtx, S. Chakram, K. He, A. Agrawal, R. K. Naik, D. I. Schuster, and A. Chou, Phys. Rev. Lett. 126, 141302 (2021).
- [33]K. Inomata, Z. Lin, K. Koshino, W. D. Oliver, J.-S. Tsai, T. Yamamoto, and Y. Nakamura, Nature Communications 7, 12303 (2016).
- [34]A. L. Pankratov, L. S. Revin, A. V. Gordeeva, A. A. Yablokov, L. S. Kuzmin, and E. Il’ichev, npj Quantum Information 8, 61 (2022).
- [35]A. L. Pankratov, A. V. Gordeeva, A. V. Chiginev, L. S. Revin, A. V. Blagodatskin, N. Crescini, and L. S. Kuzmin, Nature Communications 16, 3457 (2025).
- [36]N. Lauk, N. Sinclair, S. Barzanjeh, J. P. Covey, M. Saffman, M. Spiropulu, and C. Simon, Quantum Science and Technology 5, 020501 (2020).
- [37]J. Holzgrafe, N. Sinclair, D. Zhu, A. Shams-Ansari, M. Colangelo, Y. Hu, M. Zhang, K. K. Berggren, and M. Lončar, Optica 7, 1714 (2020).
- [38]R. Sahu, W. Hease, A. Rueda, G. Arnold, L. Qiu, and J. M. Fink, Nature Communications 13, 1276 (2022).
- [39]L. Qiu, R. Sahu, W. Hease, G. Arnold, and J. M. Fink, Nature Communications 14, 3784 (2023).
- [40]C. Wang, I. Gonin, A. Grassellino, S. Kazakov, A. Romanenko, V. P. Yakovlev, and S. Zorzetti, npj Quantum Information 8, 149 (2022).
- [41]C. M. Natarajan, M. G. Tanner, and R. H. Hadfield, Superconductor science and technology 25, 063001 (2012).
- [42]P. Sikivie, Phys. Rev. D 32, 2988 (1985).
- [43]C. W. Gardiner and M. J. Collett, Phys. Rev. A 31, 3761 (1985).
- [44]M. Tsang, Phys. Rev. A 84, 043845 (2011).
- [45]M. Aspelmeyer, T. J. Kippenberg, and F. Marquardt, Rev. Mod. Phys. 86, 1391 (2014).
- [46]W. Hease, A. Rueda, R. Sahu, M. Wulf, G. Arnold, H. G. Schwefel, and J. M. Fink, PRX Quantum 1, 020315 (2020).
- [47]Y. Xu, A. A. Sayem, L. Fan, C.-L. Zou, S. Wang, R. Cheng, W. Fu, L. Yang, M. Xu, and H. X. Tang, Nature Communications 12, 4453 (2021).
- [48]H. K. Warner, J. Holzgrafe, B. Yankelevich, D. Barton, S. Poletto, C. J. Xin, N. Sinclair, D. Zhu, E. Sete, B. Langley, E. Batson, M. Colangelo, A. Shams-Ansari, G. Joe, K. K. Berggren, L. Jiang, M. J. Reagor, and M. Lončar, Nature Physics 21, 831 (2025).
- [49]W. Fu, M. Xu, X. Liu, C.-L. Zou, C. Zhong, X. Han, M. Shen, Y. Xu, R. Cheng, S. Wang, L. Jiang, and H. X. Tang, Phys. Rev. A 103, 053504 (2021).
- [50]S. Zorzetti, C. Wang, I. Gonin, S. Kazakov, T. Khabiboulline, A. Romanenko, V. P. Yakovlev, and A. Grassellino, Physical Review B 107, L220302 (2023).
- [51]A. Reineri, F. Crisa, A. Murthy, M. Shinde, D. Bafia, C. Wang, T. Roy, A. Romanenko, J. Zasadzinski, A. Grassellino, et al., EPJ Quantum Technology 13, 104 (2026).
- [52]A. Sekine, B. Murakami, and Y. Doi, npj Nanophotonics 3, 42 (2026).
- [53]T. Werner, E. Riyazi, S. Hawaldar, R. Sahu, G. Arnold, P. Falthansl-Scheinecker, J. A. S. Naranjo, D. Loi, L. N. Kapoor, M. Zemlicka, et al., arXiv preprint arXiv:2602.00928 (2026).arxiv.org/abs/2602.00928
- [54]H. Shi and Q. Zhuang, npj Quantum Information 9, 27 (2023).
- [55]M. Rosati and A. Solana, Optica Quantum 2, 390 (2024).
- [56]H. Shi, A. J. Brady, W. Górecki, L. Maccone, R. Di Candia, and Q. Zhuang, npj Quantum Information 11, 48 (2025).
- [57]C. Wang, D. Hu, S. Zorzetti, A. Grassellino, A. Romanenko, and Z. Zhang, Physical Review Letters 136, 250801 (2026).


