Generation of ultra-broadband frequency comb in strongly bistable nonlinear magnonic resonator

Abstract

Magnonic frequency combs (MFCs) offer a promising route to compact, energy-efficient platforms for on-chip coherent microwave signal generation and processing. Conventional on-chip comb generation typically relies on nonlinear resonators supporting equidistant, low-loss resonances driven by a monochromatic signal, resulting in fixed comb spacing. Here we introduce and experimentally demonstrate a distinct mechanism for ultra-broadband MFC generation using a highly nonlinear miniaturized magnonic resonator. The small resonator volume, combined with a slow-wave transducer, drives the system deep into the bistable regime where parametric excitation of propagating spin waves facilitates comb formation. Our approach yields over 350 comb lines spanning a 450 MHz bandwidth, with spacing continuously tunable via a two-tone external drive, representing an order-of-magnitude enhancement over prior reports at relatively low power. The platform is ultra-compact, scalable, and highly tunable, establishing a distinct frequency comb paradigm with transformative opportunities in microwave signal processing, neuromorphic computing, and precision sensing.

Researchers report a breakthrough in ultra-broadband magnonic frequency combs by driving a nonlinear YIG resonator far off-resonance, yielding over 350 comb lines across 450 MHz with continuously tunable spacing.

Introduction

Frequency combs—spectral structures of evenly spaced lines—are foundational tools across diverse scientific and technological domains^1–4. Initially developed in photonic systems^5–8, they have enabled breakthroughs in precision spectroscopy^9,10, optical clocks^5,11, high-speed communications^12–14 and quantum technology^15–17. More recently, the concept has been extended to magnonic systems, exploiting their exceptional tunability and compact form factors^18–22. This emerging approach, known as the magnonic frequency comb (MFC)^23–34, offers strong potential for complete on-chip integration without off-chip components, while providing widely reconfigurable operation bands and controllable comb-line spacing.

Conventional on-chip frequency comb generation relies on a nonlinear resonator supporting a series of equidistant, narrow-linewidth resonances^{7,8,12,35–37}. When driven by a high-power monochromatic signal, cascaded four-wave mixing provides gain to offset resonance losses, producing combs with fixed spacing equal to the resonator’s free spectral range (FSR). In magnonics, combs can also be generated by driving a single nonlinear magnonic resonance with multiple microwave tones²⁵, producing a well-defined comb without the need for multiple resonances and thereby offering significantly enhanced in-situ tunability and reconfigurability. Although this approach has attracted considerable interest, existing demonstrations are limited in bandwidth and comb-line count^23–34, primarily due to challenges in device design, inefficient magnon excitation, and insufficient nonlinear enhancement.

Conventional on-chip frequency comb generation relies on a nonlinear resonator supporting a series of equidistant, narrow-linewidth resonances^{7,8,12,35–37}. When driven by a high-power monochromatic signal, cascaded four-wave mixing provides gain to offset resonance losses, producing combs with fixed spacing equal to the resonator’s free spectral range (FSR). In magnonics, combs can also be generated by driving a single nonlinear magnonic resonance with multiple microwave tones²⁵, producing a well-defined comb without the need for multiple resonances and thereby offering significantly enhanced in-situ tunability and reconfigurability. Although this approach has attracted considerable interest, existing demonstrations are limited in bandwidth and comb-line count^23–34, primarily due to challenges in device design, inefficient magnon excitation, and insufficient nonlinear enhancement. It should be noted that previous studies of spin-wave parametric instabilities³⁸ have focused on the transition to chaos under resonant transverse pumping, a regime distinct from the deterministic, non-chaotic switching explored here.

To overcome these constraints, we introduce a miniaturized magnonic resonator based on a patterned 200-nm-thick yttrium iron garnet (YIG) film driven by a two-tone signal, and demonstrate ultra-broadband MFC generation through a mechanism distinct from prior approaches. Rather than relying on a series of equidistant resonances, our approach leverages strong, far-off-resonance detuning of the driving tones from the linear resonance to push the resonator deep into a large-amplitude bistable regime, where parametric excitation of propagating spin waves occurs at the pump frequencies. Cross-mode nonlinear interactions between these spin waves and the uniform mode induce rapid, deterministic oscillations of the uniform mode amplitude between two bistable states (Fig. 1a), giving rise to MFC formation. The small resonator volume, combined with a slow-wave microwave transducer, ensures high power density in the resonator—enhancing nonlinearity and widening the bistability loop—leading to low-threshold, high-density MFC generation. Our experiments demonstrate at least an order-of-magnitude improvement in comb bandwidth and line count over prior MFC demonstrations^23–34, along with continuously tunable spacing, and a threefold enhancement over a parallel study³⁹ that relies on fixed FSR-determined spacing. Our on-chip MFC platform provides a scalable and tunable solution for next-generation electronics, unlocking previously inaccessible regimes of spin-wave dynamics for advanced functionalities in computing, sensing, and quantum technologies.

Fig. 1. Concept and operating principles of the nonlinear magnonic microresonator.

a Schematic of the device architecture: a normally magnetized YIG microresonator flip-bonded onto a slow-wave microwave waveguide supporting spoof surface plasmon polaritons. The device is driven by a two-tone microwave pump consisting of two frequencies ω_s and ω_s + δ, which exhibits a time-varying envelope at the beat frequency δ. The device output is analyzed using a spectrum analyzer. b–d Magnon resonance spectra measured under small detuning (ω_s ≈ ω₀), large detuning (ω_s > ω₀), and extremely large detuning (ω_s − ω₀ > 100 Γ), respectively. Panel c is a subset of (d). Red dots indicate the accessible power levels. e Bistability and hysteresis in the magnon power at large detuning are observed when sweeping the input power. Transitions occur from the lower (upper) branch to the upper (lower) branch at $P_{\max }$ ($P_{\min }$) during upward (downward) sweeps. In the presence of parametric pumping of spin waves, $P_{\max }$ is reduced to P_par. f Magnon dispersion (purple curves) and parametric pumping at ω_s > ω₀. Off-resonant driving of uniform magnon mode at ω_s excites incoherent large-wavevector spin-wave magnon pairs (purple shaded regions) at the same frequency. g Temporal evolution of uniform magnon mode amplitude (blue) and parametric spin-wave magnons (red). N_par denotes the threshold for parametric excitation; N_lim marks the level where the uniform magnon mode shifts to ω_s.

Results

Large-amplitude nonlinear magnonic response

Our MFC device (Fig. 1a) comprises a perpendicularly magnetized yttrium iron garnet (YIG) thin-film microresonator driven by a two-tone, periodically varying microwave signal to generate the comb. Compared with conventional millimeter-scale YIG spheres, our thin-film microresonator (0.2 × 50 × 50 μm³) reduces the device volume by 4–6 orders of magnitude, substantially increasing power density for a given microwave drive and thereby enhancing comb generation. Integration with a spoof surface plasmon polariton (SSPP) waveguide further amplifies the driving field through the slow-wave effect⁴⁰, ensuring efficient power delivery to the uniform magnon mode of the YIG resonator. The intrinsic frequency ω₀ of this mode is tuned slightly below the cutoff frequency (around 10.1 GHz) of the SSPP mode, where field enhancement is strongest.

When driven by a detuned microwave signal at ω_s ≫ ω₀, the system supports a uniform magnon mode (k = 0) and spin waves with nonzero wavevectors (k ≠ 0), both exhibiting strong nonlinear dynamics. Their amplitudes—a for the uniform magnon mode and c_k for the spin-wave mode—evolve according to:

$$\frac{da}{dt}+\mathrm{\Gamma }a=-i{{\omega }^{~}}_{0}a-i{\omega }_{\mathrm{M}}\left({\sum }_k{c}_k{c}_{-k}\right)a^{*}+i{F}_{\mathrm{s}}\left(1+e^{-i\delta t}\right)e^{-i{\omega }_{\mathrm{s}}t},$$ 1

$$\frac{d{c}_k}{dt}+{\mathrm{\Gamma }}_k{c}_k=-i{{\omega }^{~}}_k{c}_k-i{\omega }_{\mathrm{M}}{a}^2{c}_{-k}^{*}.$$ 2

where the mode frequencies of the uniform magnon mode (${{\omega }^{~}}_0$) and the spin-wave modes (${{\omega }^{~}}_k$), nonlinearly shifted by self- and cross-mode interactions, are given by:

$${{\omega }^{~}}_0={\omega }_0+{\omega }_{\mathrm{M}}\mid a{\mid }^2+2{\omega }_{\mathrm{M}}{\sum }_k\mid c_k{\mid }^2,$$ 3

$${{\omega }^{~}}_k={\omega }_k+2{\omega }_{\mathrm{M}}\mid a{\mid }^2+2{\omega }_{\mathrm{M}}{\sum }_k\mid c_k{\mid }^2,$$ 4

where ω₀ (ω_k) is the intrinsic angular frequency of the uniform (spin wave) magnons, and Γ (Γ_k) denote the damping rate of the uniform (spin wave) magnon mode. For simplicity and without loss of generality, we assume Γ_k = Γ. Here, F_s and ω_s are the amplitude and angular frequency of the microwave drive, δ denotes the frequency difference between the two microwave components, which sets the comb line spacing of the resulting MFC. The coefficient ω_M represents both the nonlinear frequency-shift coefficient of the lowest (uniform) magnon mode in our perpendicularly magnetized YIG resonator (see Eq. (40) in Ref. ⁴¹) and its nonlinear coupling to the parametrically excited plane-wave magnons. For simplicity—and without loss of generality—we assume that the excited magnon modes are plane spin waves, allowing ω_M to be treated as a constant independent of wavenumber and spatial mode profiles. Incorporating realistic mode profiles would refine the quantitative values, but does not alter the underlying nonlinear interaction mechanism.

Equations (1, 2) describe two important effects. First, there is parametric interaction between the uniform resonator mode a and the pairs of plane spin wave modes c_±k, described by the term ${\omega }_{\mathrm{M}}\left({\sum }_k{c}_k{c}_{-k}\right)a^{*}$ in Eq. (1) and the term ${\omega }_{\mathrm{M}}{a}^2{c}_{-k}^{*}$ in Eq. (2). Second, these equations, along with Eqs. (3, 4) describe mutual nonlinear frequency shifts leading to nonlinear renormalization of the resonant frequency of the uniform resonator mode and spin waves.

Under the influence of the self-nonlinearity, the uniform resonator mode exhibits a hardening Duffing-type resonance, meaning its frequency shifts upward with increasing power, and becomes bistable at sufficiently high power (e.g., P₄ in Figs. 1b, c). Consequently, the achievable power of the uniform magnonic mode (P_SD) is limited when the microwave drive F_s is applied at small detuning Δω_s = ω_s − ω₀ ≈ 0 (Fig. 1b). For MFC generation, large magnon power is required, which can only be achieved with a driving at large detuning (Fig. 1c). Note that Fig. 1c illustrates a modest detuning Δω_s ≈ Γ for clarity. In practical devices, Δω_s can even exceed Δω_s > 100Γ, enabling much higher magnon power (Fig. 1d). Such power dependence is clearly demonstrated by the hysteresis loop in Fig. 1e, which shows that the system transitions to the upper branch at $P_{\max }$ during an up-sweep (see Supplementary Note 1). However, $P_{\max }$ grows rapidly at large detunings and quickly surpasses the power levels achievable in typical experimental settings, posing a major obstacle to reaching the high-amplitude regime required for broadband MFC generation.

The nonlinear interaction between the uniform magnon mode and parametrically excited spin waves provides a pathway to overcome this limitation. Driven by a strong off-resonance pump at ω_s, the uniform magnon mode excites pairs of spin waves with wavevectors ±k at the same frequency via the second-order Suhl process⁴² (Fig. 1f). As the population of these parametric magnons N_p = ∑_k∣c_k∣² grows, the uniform magnon mode experiences a nonlinear frequency shift ${{\omega }^{~}}_0={\omega }_0+{\omega }_{\mathrm{M}}{N}_0+2{\omega }_{\mathrm{M}}{N}_{\mathrm{p}}$. This shift reduces the effective pump detuning, increasing the uniform magnon mode population N₀ = ∣a∣², and, consequently, further amplifying the parametric excitation process. Through this positive feedback loop, the uniform magnon mode amplitude N₀ rises rapidly (Fig. 1g), accompanied by a rapid frequency shift until N₀ = N_lim, where ${{\omega }^{~}}_0$ exceeds ω_s. At this point, the entire magnon band moves above the driving frequency, terminating the parametric process, as indicated by the sharp drop of parametric magnons (red curve) in Fig. 1g. This threshold behavior—where the uniform magnon population grows explosively—occurs when N₀ surpasses the parametric threshold N_par = Γ/ω_M, effectively reducing the high-amplitude transition power from $P_{\max }$ to a threshold that is close to the parametric excitation threshold P_par. For large detunings, the corresponding threshold P_par is much lower than $P_{\max }$ and remains well within the experimentally accessible power range, providing a practical route to reach the high-amplitude state.

Using Eqs. (1, 2), it is possible to obtain an explicit expression for the threshold power P_th = P_par of the MFC formation [see Eq. (S11) in Supplementary Note 1], expressed as a function of the detuning Δω_s

$$P_{\mathrm{th}}=P_{\mathrm{par}}=\frac{\mathrm{\Gamma }}{{\omega }_{\mathrm{M}}}{\left(\mathrm{\Delta }{\omega }_{\mathrm{s}}\right)}^2.$$ 5

In the absence of plane-wave parametric excitation (e.g., for small detuning Δω_s ~ Γ), the MFC excitation threshold is much larger and has the form [see Eq. (S3) in Supplementary Note 1]:

$$P_{\max }=\frac{4}{27{\omega }_{\mathrm{M}}}{\left(\mathrm{\Delta }{\omega }_{\mathrm{s}}\right)}^3.$$ 6

Thus, the reduction of the MFC excitation threshold due to the excitation of plane spin waves can be evaluated as $\frac{P_{\mathrm{par}}}{P_{\max }}=\frac{27\mathrm{\Gamma }}{4\pi \mathrm{\Delta }{\omega }_{\mathrm{s}}}$, which, in our experimental regime (where Δω_s ≈ 100 Γ), is approximately 0.07. For such device parameters, the effective pump-field amplitude is about 1.4 mT based on Eq. (6), which is difficult to achieve experimentally, whereas Eq. (5) corresponds to an accessible pump-field amplitude of only 0.37mT, which is comfortably within the experimentally accessible microwave drive range.

Dynamics of ultra-broadband frequency comb generation

Based on these nonlinear processes, the two-tone microwave pump (Fig. 2a, inset) drives the uniform magnon mode, taking the system through a cycle of bistable transitions (Fig. 1c). At low drive powers (t₁ and t₂), the amplitude of uniform magnons N₀ follows the lower branch until the positive feedback described above reduces the transition power to P_par at t₃, triggering a jump to the upper branch and a rapid increase in amplitude. The system remains in this high-amplitude state until the periodically varying drive decreases and the magnon power falls to $P_{\min }$ at t₆, returning to the lower branch. With reduced magnon power, the upper transition point resets to $P_{\max }$. Under periodic modulation, this cycle repeats, producing sustained oscillations in magnon power (Fig. 2b). The transient decaying oscillations observed after the jump (from t₃ to t₆) are characteristic of a forced nonlinear oscillator.

Fig. 2. Dynamic process of magnon frequency comb generation.

a Time trace of the microwave drive signal. $P_{\min }$, $P_{\max }$, and P_par correspond to the turning-point powers shown in Fig. 1d. b Time trace of the uniform magnon mode power. c Instantaneous states of the system during one modulation cycle at large detuning. Blue dots and lines: lower branch; red dots and lines: upper branch; arrows indicate sweep direction. d Calculated spectrum of the magnon temporal response under periodic modulation. e Calculated spectrum with the nonlinear dissipation via coupling to higher-order magnon modes is accounted for. Amp, Amplitude; arb. units, arbitrary units.

This temporal evolution produces an ultra-broadband frequency comb in the theoretical spectrum (Fig. 2d) calculated based on Eqs. [(1, 2)], spanning over 1 GHz, and comprising hundreds of comb lines. The broad spectral width originates from the sharp transition between the lower and upper branches of the bistable response. In our device, this transition is very abrupt—approaching step-like behavior—due to the large magnon amplitude achieved under the large detuning. Such rapid switching introduces high-frequency components into the temporal dynamics, which translates into a comb extended in the frequency domain. In practical implementations, however, nonlinear coupling between the uniform magnon mode and higher-order spin-wave modes must be considered. These modes, located at higher frequencies (inset of Fig. 2e), act as additional loss channels that limit the effective bandwidth of the generated comb. Due to this effect, the MFC spectrum becomes asymmetric with respect to the driving frequency, with the majority of comb lines concentrated below ω_s, as shown in Fig. 2e. The calculated bandwidth of the comb is approximately 500 MHz, which is consistent with the experimental results presented below.

Experimental demonstration of the ultra-broadband frequency comb

In our experiments, the MFC device is driven by a microwave pump formed by coherently combining two signals at frequencies ω₁/2π = ω_s/2π = 10.050 GHz and ω₂/2π = ω_s/2π + δ/2π = 10.052 GHz, each with an amplitude of P_s = − 3 dBm. The pump frequencies are fixed just below the cutoff of the slow-wave waveguide to maximize slow-wave enhancement, while different pump detuning Δω_s is achieved by tuning the uniform magnon mode frequency ω₀. Fixing the pump frequency rather than the magnon frequency is critical because it ensures that the pump power after slow-wave enhancement (which is strongly frequency-dependent) remains constant across detuning conditions, enabling a fair comparison of comb generation under different detunings. We examine two detuning conditions by biasing the magnon mode with different magnetic fields: (i) small detuning, ω₀/2π = 10.050 GHz (Δω_s = 0); and (ii) large detuning, ω₀/2π = 9.860 GHz (Δω_s/2π = 190 MHz). For the small-detuning case, a low sweep power of −30 dBm reveals the intrinsic magnon resonance as a narrow absorption dip with a linewidth of about 2.0 MHz at 10.05 GHz (Fig. 3a). In contrast, the large-detuning case uses a higher sweep power of −7 dBm, producing a broad resonance spanning 170 MHz—nearly two orders of magnitude wider than the magnon linewidth—and clear hysteresis between upward and downward sweeps (Fig. 3b). The span of this nonlinear resonance increases further with higher power and can eventually cover the pump frequency in our experiment.

Fig. 3. Experimental spectrum of the MFC.

a, b Device transmission spectra measured at low power (−30 dBm) and high power (−7 dBm), respectively. The high-power spectrum shows pronounced hysteresis, absent in the low-power regime. Periodic oscillations at high power originate from bulk acoustic phonon modes, a common feature in thin-film YIG magnonic devices^39,47. c MFC spectrum measured with a pump power of P_s = − 30 dBm at ω₁/2π = ω₀/2π = 10.05 GHz. d MFC spectrum measured with a pump power P_s = −3 dBm at ω₁/2π = 10.05 GHz and ω₀/2π = 9.86 GHz. e Measured time trace of the MFC signal. f Zoomed-in plot of the time trace in (e), showing the temporal oscillation within each time cycle. g Output spectra recorded at varying pump powers. h Device output power versus pump power, showing a threshold at −7 dBm. i Total MFC power as a function of pump power P_s and frequency ω₁. Inset: extracted threshold power P_th as a function of pump detuning Δω_s/2π. Yellow dots indicate extracted threshold points; the solid black line shows the theoretical P_par curve, which shows a quadratic dependence on Δω_s; dashed blue line plots the theoretical $P_{\max }$ curve, which has a cubic dependence on Δω_s. arb. units, arbitrary units.

In both cases, MFCs with the comb line spacing equal to δ/2π = 2 MHz are produced, but their spectral characteristics differ significantly. In the small-detuning case, the resonance shifts away due to the strong nonlinear effect under a strong pump, limiting the magnon amplitude and the corresponding comb bandwidth. This prediction matches the measured MFC in Fig. 3c, which spans only 10 MHz with six comb lines. In the large-detuning case, the comb spans 9.87–10.09 GHz (Fig. 3d), corresponding to 220 MHz and about 100 comb lines, which is an order-of-magnitude increase. The comb shows a relatively flat envelope and pronounced asymmetry around the driving frequency, with more than 70 comb lines below and fewer than 20 above. The typical power per comb line is about −45 dBm (≈30 nW). All reported power values refer to on-device levels after accounting for input and output losses.

Figure 3e shows the time trace of an MFC with δ/2π = 3 MHz measured using homodyne detection (see Supplementary Note 2 for details), which shows up as a train of pulses with a period T = 2π/δ = 330 ns. Zoomed-in plot of the pulse train (Fig. 3f) shows that each pulse consists of a periodic oscillation, matching the theoretical calculation in Fig. 2b.

As predicted by our theoretical analysis, MFC generation exhibits a clear threshold behavior. Figure 3g shows the measured output spectra with the input pump power (P_s) sweeping from −20 to 0 dBm, and Fig. 3h plots the extracted total comb power versus input power. A distinct threshold appears at P_s = −7 dBm, where the output power abruptly increases by about 20 dB. Beyond this point, the output rapidly evolves into a broadband MFC with a bandwidth exceeding 200 MHz. Furthermore, the output power saturates above the threshold, consistent with the theoretical prediction in Fig. 2c, which indicates that the comb originates from periodic sharp transitions between the lower and upper branches of the bistable nonlinear magnonic resonance curve, while the power variation within the upper branch after the transition plays only a minor role. Our analysis also predicts that the threshold power depends on the driving frequency detuning Δω_s, which is confirmed by the measured comb power (color coded in Fig. 3i) as a function of both the pump power and pump frequency. When ω_s < ω₀, no MFC generation occurs. When ω_s exceeds ω₀, a clear input power threshold emerges, above which the comb generation begins. The extracted threshold power P_th is plotted as a function of Δω_s in the inset of Fig. 3i (yellow squares), which follows a parabolic relation according to our theoretical calculation of P_par (solid black line) and deviates significantly from the cubic relation of the calculated $P_{\max }$, showing excellent agreement with theoretical predictions introduced earlier.

Magnonic frequency comb bandwidth expansion

The versatility of magnons enables flexible control of the MFC properties through multiple parameters for a given device. Owing to the magnetic tunability of the magnon frequency and the design flexibility of the slow-wave waveguide, MFCs can be generated across different frequency bands. In our experiments, ultra-broadband MFCs have been reliably achieved within the 3–10 GHz range. Beyond the operating frequency, key MFC characteristics—including the comb span, the number of comb lines, and the total comb power (excluding pump tones)—can be precisely controlled by adjusting the bias magnetic field. Figure 4a shows MFC parameters obtained at different bias magnetic fields under fixed pumping conditions: ω₁/2π = 10.005 GHz, δ/2π = 2 MHz, and P_s = −3 dBm. As the bias field decreases, the corresponding reduction in the magnon frequency ω₀ increases the pump detuning Δω_s, which, in turn, broadens the MFC span to as much as 300 MHz with approximately 150 comb lines. Under these conditions, the total on-chip comb power reaches 1.8 μW, corresponding to an average power of about 12 nW per comb line.

Fig. 4. Tuning of the MFC bandwidth.

a, b MFC characteristics--span, line count, and total on-chip comb power–as functions of the bias magnetic field and pump frequency difference δ, respectively. c Comparison of comb line count achieved in this work (red star) with state-of-the-art results, including experimental demonstrations (red circles) and theoretical proposals (blue squares), extracted from Refs. ^23–34. d Measured MFC spectrum with a comb line spacing of 2 MHz and a total of 200 lines, obtained with uniform magnon mode frequency ω₀/2π = 9.35 GHz, pump frequency ω₁/2π = 10.0232 GHz and ω₂/2π = 10.0252 GHz, and pump power P_s = −3 dBm. Right inset: magnified view of the comb spectrum. Left inset: magnified view of a single comb line with Gaussian fitting, revealing a linewidth of 26 Hz. e MFC spectrum exhibiting period doubling, obtained with uniform magnon mode frequency ω₀/2π = 9.35 GHz, pump frequency ω₁/2π = 10.0232 GHz and ω₂/2π = 10.0265 GHz (δ/2π = 3.3 MHz), and pump power P_s = −3 dBm. The comb line spacing is 1.65 MHz (corresponding to δ/2), with a total line count of 225. Inset: magnified view of the comb spectrum. Exp, Experimental results; Amp, Amplitude.

The MFC properties can also be effectively controlled by the pump spacing δ, as shown in Fig. 4b. For instance, the comb span increases monotonically as δ decreases. When δ/2π = 1.3 MHz, the comb span extends to 450 MHz with a total of 350 comb lines. However, the dependence of the comb power on δ is non-monotonic. As δ decreases from 10 MHz to 4.3 MHz, the comb power increases. Beyond this point, although the comb line count continues to grow, the power per line diminishes, leading to a reduction in total comb power. The maximum total on-chip comb power achieved in our experiments is 2.58 μW at δ/2π = 4.3 MHz, corresponding to an average of 31 nW per comb line.

The 350 comb line count obtained in our experiment represents the highest achieved to date in MFCs. Figure 4c provides a summary of recent progress in this field (see Supplementary Note 3 for details), serving as a benchmark for comparison. The comb-line count reported in this work exceeds all previous experimental demonstrations by at least one order of magnitude and even far surpasses existing theoretical predictions and simulations. This establishes our ultra-broadband MFC platform as the most advanced among reported microwave MFC systems, regardless of operating frequency, material platform, or physical mechanism. A parallel study³⁹ was reported independently during the preparation of this manuscript, which demonstrated an MFC with 130 comb lines by exploiting a series of equidistant, low-loss bulk acoustic resonances in the substrate of the YIG-film resonator. This approach relies on the substrate and imposes a fixed comb spacing determined by the FSR of the mechanical resonances. Our method is fundamentally different: it does not depend on the mechanical resonance in the substrate and can, in principle, be implemented on arbitrary substrates, offering great versatility for on-chip integration. Furthermore, our approach enables continuous tuning of comb line spacing, a unique advantage of MFCs compared with other frequency comb platforms.

Figures 4d shows the spectrum of the widest MFC obtained in our experiments, along with a close-up view of its comb lines, clearly revealing individual lines spaced by 1.3 MHz. Gaussian fitting yields a single comb-line linewidth of 26 Hz (left inset of Fig. 4d), limited by the resolution bandwidth of the spectrum analyzer, indicating excellent coherence with an actual linewidth below this value. The comb lines are superimposed on a broadband continuum background, which is likely attributed to spontaneous four-magnon scattering from the pump tones²⁵. This magnon continuum is more pronounced at the lower end of the MFC band, where it lies closer to the intrinsic frequency of the uniform magnon mode. Consequently, the extinction ratio of the comb lines is higher at the upper end of the span and gradually decreases toward the lower frequency end.

Another practical approach to further increase the comb line count is to exploit the period-doubling effect, a well-established feature in frequency comb systems. For a given MFC, this phenomenon typically occurs prior to the onset of chaotic dynamics, and it can approximately double the number of comb lines. In our system, this effect is observed at δ/2π = 3.3 MHz. As shown in Fig. 4e, it results in a comb line count of 225 with a halved line spacing of 1.65 MHz.

Discussion

We present an approach for realizing ultra-broadband MFCs by engineering the nonlinear response of a nonlinear magnonic YIG-film microresonator. Leveraging the interplay between nonlinear frequency shift and parametric pumping, our platform achieves low-threshold comb generation with a bandwidth of up to 450 MHz and a comb line count of 350. To the best of our knowledge, this represents the highest experimentally reported magnonic comb-line count, exceeding previous demonstrations by more than an order of magnitude and surpassing existing theoretical proposals (Fig. 4c). The MFC properties—including bandwidth, line density, and power distribution—can be tuned via external parameters such as magnetic bias and pump detuning, revealing rich nonlinear dynamics and enabling precise control of comb structures through bistability engineering.

The large comb-line count and broadband spectral extent originate primarily from the strong intrinsic nonlinearity of the YIG microresonator and the associated bistable response, which together facilitate efficient energy transfer across widely separated frequency components. Operation at large frequency detuning further broadens the accessible bandwidth by activating high-order comb lines. In contrast, parametric excitation and slow-wave enhancement act chiefly as enabling mechanisms that reduce the threshold and increase pump power density, rather than fundamentally determining the achievable comb span.

Our theoretical framework, based on direct numerical integration of Eqs. (1, 2), shows excellent agreement with the experimentally observed dynamics of ultrabroadband MFC generation. The model reproduces all key features of the measured behavior, including the multi-hundred-megahertz comb bandwidth, the quadratic threshold scaling with detuning, and the time-domain pulse-train structure, demonstrating a coherent and self-consistent physical picture. This multi-point correspondence across spectral, threshold, and temporal domains provides strong validation of the underlying mechanism and confirms that the essential physics of MFC formation is captured by the above-presented theoretical description.

The relatively low absolute power observed in the experimental spectra is mainly attributable to impedance and mode mismatch between the small magnetic element and the microwave resonator used for excitation. Because the YIG film occupies only a small fraction of the resonator mode volume, only a limited portion of the incident microwave energy is converted into comb power. This limitation is not fundamental: replacing the single element with a synchronized array of magnetic nano-elements would increase the interaction volume and substantially enhance the output power of the comb lines, improving suitability for practical applications.

Beyond performance metrics, this work establishes a scalable, hardware-efficient platform for integrating frequency comb functionality into magnonic circuits. The spectral richness of MFCs enables access to high-dimensional mode spaces, which can be harnessed for advanced computing and sensing applications. In particular, the inherent parallelism of the comb structure supports operations such as vector-matrix multiplication^43–46, a key primitive in the neuromorphic and analog computing. These capabilities are achieved without increasing the physical complexity or footprint of the device by exploiting the frequency domain as an additional degree of freedom. By uniting compactness, tunability, and computational potential, our MFC platform offers a compelling alternative to earlier established photonic, acoustic and magneto-acoustic systems. Its integration into magnonic architectures opens new directions for microwave signal processing, high-dimensional information processing, and precision sensing, laying the groundwork for next-generation functional devices driven by nonlinear magnetic phenomena.

Methods

Device description

The magnonic micro-resonator is a 200-nm yttrium iron garnet (YIG) thin film patterned into a 50 μm × 50 μm square on a 500-μm-thick gadolinium gallium garnet (GGG) substrate. The YIG chip is biased by an out-of-plane magnetic field, supporting magnons in the forward magnetostatic volume mode. This YIG structure is flip-bonded onto the slow-wave waveguide, which consists of a periodically corrugated microstrip fabricated on a printed circuit board (PCB). Microwave signals propagating along the corrugated microstrip excite spoof surface plasmon polaritons, which exhibit significantly reduced group velocity and enhanced mode confinement. This leads to a substantial enhancement in the pumping efficiency of magnon modes compared to conventional transducers such as microstrips or coplanar waveguides (CPWs), thereby boosting the nonlinear dynamics in the YIG resonator. Notably, due to the propagating nature of spoof surface plasmon polaritons, this enhancement is broadband, which is crucial for the broadband generation of MFCs.

Author contributions

Y.J. prepared the device, performed the measurements, and analyzed the data. V.T. and A.S. developed the theoretical model and performed numerical simulations. Y.H. assisted with data analysis. Z.Y. and A.P. assisted with measurements. X.Z. conceived the idea, supervised the project, and assisted with measurements and data analysis. Y.J., V.T., A.S. and X.Z. wrote the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

A.S. discloses support for the research of this work from the Department of Defense [STTR W911NF25P0003]. Y.J. and X.Z. discloses support for the research of this work from the National Science Foundation [ECCS 2337713]. V.T., Y.H., Z.Y. and A.P. declare no relevant funding.

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74143-4.