A wideband tunable, nonreciprocal bandpass filter using magnetostatic surface waves with zero static power consumption

Abstract

Modern wireless systems demand compact, power-efficient radio frequency (RF) front-end components that support wideband tunability and nonreciprocity. We present a class of miniature bandpass filter that achieves both continuously tunable frequency operation (4.0–17.7 GHz) and high nonreciprocity ( > 25 dB), all within a compact size of 1.07 cm³. The filter employs a microfabricated 18 µm thick Yttrium Iron Garnet (YIG) waveguide with meander-line aluminum transducers, enabling low-loss unidirectional propagation via magnetostatic surface waves. Leveraging a benzocyclobutene planarization fabrication process, this study enables a dispersion profile unique to thick YIG films, resulting in enhanced filter skirt performance with minimal spurious modes. Frequency tuning is enabled by a zero-static-power magnetic bias circuit using transient current pulses, eliminating continuous power consumption. The filter demonstrates low insertion loss (3–5 dB), high out-of-band rejection ( > 30 dB), narrow bandwidth (100–200 MHz), robust power handling ( > 10.4 dBm), and high linearity (IIP3 > 26 dBm).

This work demonstrates the first miniature RF filter combining ultrawide tunability (4–18 GHz) and high nonreciprocity, achieving low loss, zero static power, and compact size (1 cc).

Introduction

Modern wireless communication systems face increasing demands, creating significant challenges for radio frequency (RF) front-end design, especially in achieving wideband tunable filtering^1,2 and nonreciprocity^3,4. Figure 1a shows that integrating both capabilities permits simultaneous control of the signal’s desired passband and directional isolation. Consequently, RF front-end module design can be substantially simplified and optimized.

Fig. 1. Overview of the wideband tunable nonreciprocal filter.

a Block diagram of a transceiver system incorporating wideband tunable, nonreciprocal filters. b 3D schematic of the filter, which integrates a YIG (yttrium iron garnet) filter with a zero-static-power magnetic bias circuit. An input aluminum meander-line transducer excites a unidirectional magnetostatic wave (MSW), which propagates through the YIG and is reconverted back to an electrical signal at the output transducer. The magnetic bias structure includes two permanent magnets, two shunt magnets wound with coils, and magnetically permeable yokes that focus the magnetic flux at the filter. c Optical microscope image of the fabricated device assembly with a magnified view of the white box region showing the YIG filter and yoke gap. d Fabrication process flow of the YIG filter, incorporating an innovative benzocyclobutene (BCB) planarization technique for thickness-scaled YIG filters.

Wideband tunable filtering enables dynamic adjustment of operating frequencies across a broad spectrum. This capability is essential for modern RF systems supporting diverse applications, from sub-6 GHz 5 G New Radio for urban coverage⁵, to satellite communication downlinks at 10.7-12.7 GHz⁶. As frequency allocations grow, managing interference becomes increasingly critical. For example, spectral overlap between sub-6 GHz 5 G and C-band Very Small Aperture Terminals (VSATs) systems used in maritime and fixed satellite services poses a significant risk of unpredictable interference⁷. To address this, a combination of tuned and switched filter banks are often employed; however, this approach can incur additional loss from the additional signal paths and often hinders the creation of a multi-band filter response^8,9. Continuously tunable filters provide greater flexibility and are inherently upgradable to new bands and evolving interference environments. Some designs achieve tuning from 3.4 GHz to 11.1 GHz^10–12, but they still fall short of covering the full bandwidths supported by today’s wideband receivers^13–15.

An equally critical challenge lies in the implementation of nonreciprocal components, which are essential for protecting high-power RF sources from unwanted reflections and for isolating different sections of an RF transceiver^3,4. As shown in Fig. 1(a), a nonreciprocal bandpass filter in the transmitter path selectively allows only in-band frequency components to reach the power amplifier, while blocking both out-of-band signals in the forward direction and reflected in-band signals in the reverse direction from the amplifier or antenna. In the receiver path, the filter functions dually as a bandpass filter and an isolator, attenuating any reflected signals that could otherwise propagate back toward the antenna. Moreover, nonreciprocal components enable other applications, including full-duplex systems where the same frequency is being used for transmit and receive^4,16, quantum information processing^4,17, radars^18,19, and biomedical sensing⁴. Traditional ferrite-based isolators and circulators are bulky, while alternative approaches using transistors, nonlinearity, or temporal modulation involve performance compromising trade-offs between power consumption, insertion loss, noise, and power handling^4,19,20.

In this study, we introduce a class of RF components that simultaneously address the challenges of both wideband, continuously frequency tunable filtering and nonreciprocity. Unlike conventional solutions that require multiple discrete elements such as RF switches, filters, and isolators, our approach combines these functions into a single device, significantly reducing overall system size and complexity. The filter operates with zero static power consumption and offers continuous tunability from 4.0 GHz to 17.7 GHz, with an insertion loss of 3–5 dB, out-of-band rejection exceeding 30 dB, and isolation (S₁₂–S₂₁) greater than 25 dB. An overall schematic is shown in Fig. 1(b), and an optical microscopy image of the fabricated device is shown in Fig. 1(c). The entire device occupies a compact footprint of just 20.0 mm × 16.7 mm × 3.2 mm and consists of two key components: a nonreciprocal Yttrium Iron Garnet (YIG) bandpass filter and a zero-static-power magnetic bias circuit.

The nonreciprocal YIG filter employs a thick YIG waveguide with meander-line aluminum transducers. It operates using magnetostatic surface waves (MSSW), where an in-plane magnetic bias is applied perpendicular to the direction of wave propagation in the YIG. The meander-line transducers efficiently excite and collect MSSW, while the thick YIG waveguide supports low-loss unidirectional propagation. We demonstrate a fabrication technique in Fig. 1(d) that enables scaling the YIG thickness from 3 µm, as reported in previous literature^10,21,22, to 18 µm. This substantial increase significantly improves power handling, sharpens the filter skirt, broadens the bandwidth, and enables unidirectional MSSW propagation. Additionally, the meander-line aluminum transducers further enhance nonreciprocity, out-of-band rejection, suppress spurious modes, and produce a flatter passband response.

The magnetic bias circuit leveraged our previous work^10,23, but with improved materials and a redesigned structure. This updated design achieves a peak magnetic field of 5700 Gauss within a reduced volume of ~1.07 cm³— a substantial improvement over the previously reported 3170 Gauss in 1.68 cm^3 10. The magnetic field, and therefore the filter center frequency, is tuned by applying current pulses to AlNiCo magnets, modifying their nonvolatile remanence. This method enables dynamic frequency tuning with only transient power consumption²³, eliminating the need for steady-state power typically required by the electromagnets in YIG filters to maintain the required magnetic field for filter operation^24,25.

Result

Thickness scaled yttrium iron garnet (YIG) filter

Increasing YIG film thickness offers several advantages, including stronger MSSW coupling^26,27, reduced theoretical propagation loss²⁷, improved power handling^28,29, and enhanced unidirectionality³⁰. Thicker films improve excitation and detection efficiency due to the larger volume of magnetic material interacting with the transducer antenna’s magnetic field. In the setup shown in Fig. 1b, MSSWs propagate along the y-axis. At the top surface, waves are expressed as $A_{+y}{e}^{i\left(\omega t-\mid k_y\mid y\right)}$ for +y and $B_{-y}{e}^{i\left(\omega t+\mid k_y\mid y\right)}$ for -y directions. At the bottom surface, they are $C_{+y}{e}^{i\left(\omega t-\mid k_y\mid y\right)}$ and $D_{-y}{e}^{i\left(\omega t+\mid k_y\mid y\right)}$. Wong et al. showed that for a given spin wave wavelength, the amplitude ratio ${\mathrm{A}}_{+\mathrm{y}}/{\mathrm{B}}_{-\mathrm{y}}$ and ${\mathrm{D}}_{-\mathrm{y}}/{\mathrm{C}}_{+\mathrm{y}}$ are much higher in thick YIG films, indicating dominant MSSW propagation along $\overrightarrow{\mathrm{H}}\times \overrightarrow{\mathrm{n}}$³⁰.

Figure 2a shows the calculated dispersion relationship for 3 µm and 18 µm thick YIG films using the equations from³¹, with demagnetization effect neglected. The calculation details and the results for other thicknesses can be found in Supplementary Note #1. For an infinitely wide YIG film with an infinitely far away ground plane, the MSSW frequency is bounded by ${\omega }_{\min }=\gamma {\left[H\left(H+4\pi M_s\right)\right]}^{1/2}$ and ${\omega }_{\max }=\gamma \left(H+2\pi M_s\right)$, where $\gamma =2.8$ MHz/Gauss, $H$ is the magnetic field applied in Gauss, and $M_s=1780$ Gauss is the saturation magnetization of YIG. When the film has finite width, the wavevector in the width (z) direction becomes quantized, leading to discrete standing wave modes. Because the transverse quantization introduces a wavevector component parallel to the magnetization, the effective dipolar energy is reduced, causing these confined modes to appear at lower frequencies than the corresponding plane-wave MSSW mode. This finite-width confinement allows MSSWs to occur below ${\omega }_{\min }$, entering the volume wave frequency range. As YIG thickness increases, these width effects become more pronounced, with higher-order modes showing larger frequency separation from lower-order modes and the wavenumber increases more sharply with frequency at small wavenumbers. Since the group velocity ($V_g$) is defined as $V_g=\frac{\partial \omega }{\partial k}$, the theoretical group delay can be calculated as $1/V_g$, as plotted in the Fig. 2b. Thicker YIG exhibits a steeper dispersion, resulting in higher group velocity and thus lower group delay. As prior studies indicate, propagation loss is nearly proportional to group delay^27,32. Notably, with a magnetic bias field of 2325 Gauss, the 18 µm YIG shows a flat and low group delay between 8.41 GHz and 8.95 GHz, followed by a sharp drop at ~8.96 GHz. In contrast, the 3 µm film shows a gradual increase in group delay from 8.65 GHz to 9.0 GHz. This sharp transition can be used to create steep filter skirts, achieving a “brick-wall” response at the upper band edge.

Fig. 2. Thickness effect on YIG filters.

a Comparison of the dispersion relationships for 3 µm and 18 µm thick YIG films under a 2325 Gauss bias field for different width modes (n). The ${\omega }_{\min }$ and ${\omega }_{\max }$ frequencies represent the minimum and maximum allowable frequencies for an infinitely wide YIG film with a ground plane located at an infinite distance. b Calculated group delay per micrometer for the first width mode across different YIG film thicknesses with a bias field of 2325 Gauss. c Optical image of the fabricated YIG filter with an 18 µm thick film using BCB planarization. BCB is transparent and is not visible; aluminum transducers are 10 µm wide. d Optical image of the YIG filter with a 3 µm thick film. No BCB is used due to the thin YIG layer. Aluminum transducers are 5 µm wide. e Measured S₁₂ frequency response with applied magnetic flux density of around 2500 Gauss. f Measured S₁₂ group delay with applied magnetic flux density of around 2500 Gauss.

Despite the advantages of thicker YIG films, no prior studies have demonstrated microfabricated MSSW filters using etched YIG thicker than 3 µm, primarily due to challenges in microfabrication. Conventional MSSW filters are fabricated by patterning metal transducers, via lift-off or subtractive etching, on top of etched YIG waveguides^10,21. However, in 15–18 µm thick YIG, steep sidewalls hinder conformal metal deposition, leading to breaks (open) or incomplete removal (shorts) of the metal, as detailed in Supplementary Note #2. To overcome this, we implemented a planarization method using Benzocyclobutene (BCB), illustrated in Fig. 1d. After etching the YIG, a BCB layer was spin-coated and patterned via photolithography to remove BCB from the top surface of the YIG, minimizing the distance between the YIG and the aluminum transducers to ensure maximum energy coupling. This BCB process smooths the steep YIG sidewalls into gradual slopes. Supplementary Note #2 also confirms that the BCB layer does not introduce signal loss or alter MSSW propagation.

Figures 2c and 2d show optical images of fabricated YIG filters with identical YIG waveguide shapes but YIG thicknesses of 18 µm and 3 µm, respectively. Both filters in (c) and (d) have a width (W) of 200 µm, length of 140 µm (L) and pitch (P) representing the distance between two aluminum transducers of 70 µm. Figures 2e and 2f present the measured S₁₂ frequency response and group delay. The 18 µm-thick YIG filter exhibits significantly lower group delay and a sharp increase near 8.95 GHz. This results in a much steeper filter skirt at the upper passband edge compared to the 3 µm-thick YIG filter, indicating superior filtering performance. Supplementary Note #2 compares these two filters under various frequencies.

YIG filter with meander line transducers

Although a thicker YIG offers improved filter skirt performance at the upper frequency band, further optimization is needed to achieve a flat filter passband. Meander-line transducers have been proposed in previous studies, showing improved radiation impedance compared to single straight-line transducers through theoretical calculations^33,34, and demonstrating low insertion loss in filters utilizing 3 µm thick YIG¹². However, a comprehensive study of YIG filters employing meander-line transducers is still needed, and their application in 18 µm thick YIG remains unexplored. A previous study determined the radiation resistance for a meander-line transducer as follows³³:

$$T_m=\frac{R_m}{R_1\times l}={\left[\frac{\sin \left(\frac{W_T{k}_y}{2}\right)}{\frac{W_T{k}_y}{2}}\right]}^2{\left[\frac{\sin \left(\frac{pN{k}_y}{2}\right)}{\cos \left(\frac{p{k}_y}{2}\right)}\right]}^2$$ 1

Here, $T_m$ is the transducer specific radiation impedance per unit length, $R_m$ is the radiation impedance, and $R_1$ is a function of YIG parameters and ground-plane spacing and is independent of transducer parameters, l is the length of the meander-line transducer’s arm, which also serves as the width of the YIG waveguide, $W_T$ is the width of the transducer, $p$ is the pitch, and $N$ is the number of arms in the meander-line.

Applying the previously calculated dispersion relation into Eq. (1), the $T_m$ for different YIG thicknesses with meander-line transducers can be determined, as shown in Fig. 3a. In this case, $N=$ 4, $W_T=10\mu m$, and $p=70\mu m$. Within the first width mode of 3 µm thick YIG, multiple peaks corresponding to the different order length mode harmonics are evident. These higher-order modes appear at higher frequencies and shorter wavelengths with reduced amplitude. However, the unique dispersion characteristics of the 18 µm thick YIG pushes the frequency of all the larger wavenumber modes close to ${\omega }_{\max }$, beyond which MSSW propagation is forbidden. This significantly reduces the presence of higher-order length modes. An 18 µm thick YIG filter, detailed in Fig. 3b with a pitch of 70 µm, width of 200 µm, YIG waveguide total length of 420 µm, and meander-line with three arms and two bridges, demonstrates this effect. The arm is the section where the transducer is positioned on top of the YIG, and the bridge section connects the two arms. The measured S₁₂ frequency responses between 3 µm and 18 µm thick YIG waveguides is compared in Fig. 3c. The same YIG waveguide shape and transducers are utilized for both designs, except the 3 µm thick YIG filter has $W_T=5\mu m$, while the 18 µm thick YIG filter has $W_T=10\mu m$. The measurements confirm the substantial suppression of higher-order length mode spurious responses in the 18 µm thick YIG filter. Furthermore, the 18 µm thick YIG increases the 3 dB bandwidth from 83 MHz to 142 MHz. This broader bandwidth and flatter passband are attributed to the flat region in the dispersion relationship between ~200-300 cm^-1, while the steep dispersion at larger wavenumbers contributes to the reduced length modes. Figure 3d compares 18 µm thick YIG waveguides with meander-line and straight-line transducers, with an expanded comparison shown in Supplementary Note #4. This demonstrates that the meander-line YIG filter achieves an improved filter shape by selectively exciting specific wavelengths while attenuating others, whereas the straight-line YIG filter excites MSSWs over a broader range of wavelengths. Moreover, Supplementary Note #4 also compares the nonreciprocity of 3 µm and 18 µm thick YIG meander-line filters. It highlights that, with increasing film thickness and the resulting increased sidewall roughness, the YIG structure no longer supports resonant circulation. Instead, the spin waves excited at one port predominantly propagate in a single direction, leading to enhanced nonreciprocal behavior.

Fig. 3. YIG filters with meander-line transducers.

a Comparison of the calculated transducer specific radiation impedance for meander-line transducers on 3 µm, 10 µm, and 18 µm thick YIG films (first width mode, n = 1) at ~1500 Gauss. b Optical microscope image of a YIG filter with a meander-line transducer with a pitch of 70 µm and width of 200 µm. c Comparison of measured S₁₂ frequency responses between the 3 µm and 18 µm meander-line transducer filters under applied magnetic flux density of approximately 1500 Gauss. d Comparison of measured S₁₂ frequency responses between the meander-line and straight-line transducer 18 µm thick YIG filters under applied magnetic flux densities of approximately 1500 Gauss. e Optical image of a YIG filter with two parallel meander-line transducers with a pitch of 70 µm and width of 200 µm. f Comparison of measured S₁₂ frequency responses between filters with two parallel meander-line transducers and a single meander-line transducer at around 1500 Gauss.

Previous studies^10,35,36 and Eq. (1) indicate a linear increase in radiation impedance with the width of the YIG waveguide (or overlap length between the YIG and Al transducer). Consequently, the meander-line YIG filter, featuring a 3x longer total arm length compared to a straight-line counterpart, exhibits approximately 3x higher impedance. This results in superior performance at 6.3 GHz under ~1500 Gauss, due to the better impedance matching to the 50 Ω port impedances. The meander-line with 18 µm thick YIG shows a low insertion loss of 4.4 dB compared to 7.4 dB for the straight-line YIG filter. The maximum filter input impedance, Z₁₁, values are 40.9 dB and 33.2 dB for the meander-line and straight-line YIG filters, respectively. The minimum return losses, determined by the full impedance matrix, are 21.5 dB and 10.4 dB for the meander-line and straight-line YIG filters, respectively. However, since the radiation impedance also increases with frequency^10,35,36, at 11.8 GHz under an applied magnetic flux density of ~3500 Gauss, the higher peak Z₁₁ observed in the meander-line filter (47.4 dB vs. 42.1 dB) results in a substantial impedance mismatch and increased insertion loss of 8.3 dB, compared to just 4.1 dB for the straight-line filter.

To reduce insertion loss and enable high-frequency filtering, a two-parallel meander-line transducer design was implemented, as shown in the Fig. 3e. The input current is evenly split between two identical meander-line transducers. Figure 3f compares the performance of the one and two-parallel meander-line filters, while Supplementary Note #5 presents measurement results across different frequencies. At 6.3 GHz, introducing the parallel branch reduces the maximum Z₁₁ from 40.9 dB to 33.5 dB. This results in improved impedance matching and a reduction in insertion loss from 4.4 dB to 2.9 dB. This improvement is even more pronounced, at 11.8 GHz, where the single meander-line filter exhibits greater impedance mismatch and the insertion loss is reduced from 8.3 dB to 3.8 dB. In addition to lower insertion loss, the two-parallel transducer design also enhances out-of-band rejection by ~8 dB. This improvement arises because, at any moment, the currents in the upper and lower transducers flow in the opposite direction. The stray magnetic fields produced by these opposing currents cancel each other out far from the transducers, thereby increasing overall electromagnetic isolation. Furthermore, the use of a larger YIG waveguide in the two-parallel design contributes to 1-3 dB lower insertion loss by reducing sidewall propagation loss. Supplementary Note #6 compares the performance of two-parallel meander-line filters with single and dual YIG waveguides, showing that wider YIG waveguides exhibit lower insertion loss and reduced demagnetization effects. However, one drawback is the appearance of a more pronounced spurious mode between 5.7–5.9 GHz, as seen in Fig. 3f. This is likely due to the center bridge in the meander-line transducers, which can also excite magnetostatic backward volume waves (MSBVW), thereby introducing this spurious response. The pitch of the meander-line transducers can significantly influence the shape of the YIG filter. As discussed in Supplementary Note #7, a pitch of 70 µm is optimal for minimizing insertion loss and suppressing spurious modes.

As discussed earlier, meander-line transducers exhibit wavelength selectivity, providing enhanced excitation efficiency when the transducer period corresponds to an odd integral multiple of the MSSW wavelength, resulting in a higher radiation impedance^33,34. This property enables more effective control over the excited MSSW modes compared with straight-line transducers. They also contribute to improved out-of-band rejection. This enhancement arises because the currents in adjacent aluminum transducer lines flow in opposite directions, which helps confine the electromagnetic fields within the transducer region and reduces leakage into surrounding areas. Figure 4a shows a meander-line YIG filter with increased spacing between the input and output transducers, introducing an additional separation of up to 210 µm. The measured S₁₂ frequency responses with different extra spacings are shown in the Fig. 4b. The out-of-band rejection at 9.5 GHz improves with increased spacing, measured at 32 dB, 42 dB, 48 dB, and 53 dB for additional spacings of 0 µm, 70 µm, 140 µm, and 210 µm, respectively. Meanwhile, the insertion loss remains relatively constant at 3.1 dB, 3.2 dB, 3.0 dB, and 3.4 dB for these spacings, respectively. This constant insertion loss is attributed to the high group velocity of MSSWs in thick YIG films, resulting in negligible propagation loss. Figure 4c shows the group delay in the passband increases almost linearly with the extra spacings. A detailed comparison of different spacings at different frequencies are discussed in Supplementary Note #8.

Fig. 4. Meander-line transducer YIG filters with improved out-of-band rejection and non-reciprocity.

a Optical microscope image of a two parallel meander-line transducer YIG filter with pitch (P) of 70 µm, width (W) of 200 µm, and extra spacing (S) of 210 µm. b Measured S₁₂ frequency responses for different meander-line transducer spacings with magnetic flux density of ~2500 Gauss. c Measured passband group delay for different meander-line transducer spacings with magnetic flux density of ~2500 Gauss. d Optical microscope image of a dual-hexagon shaped YIG with two parallel meander-line transducers YIG filter with a pitch of 70 µm, width of 200 µm, and no extra spacings. e, f Comparison of measured (e) S₁₂ and (f) S₂₁ frequency responses between two parallel meander-line transducer YIG filters with dual-hexagon shaped YIG and rectangular shaped YIG at around 2500 Gauss.

Figure 4d depicts a dual-hexagon YIG design that enhances nonreciprocity and suppresses the spurious mode around 8.4 GHz, while maintaining a nearly constant insertion loss. The measured S₁₂ and S₂₁ are shown in Fig. 4e and Fig. 4f, respectively. As discussed in the previous section, MSSWs propagate unidirectionally along the top and bottom surfaces of the YIG. By modifying the YIG edge from a straight edge to a triangular shape, the MSSWs on the top surface are effectively blocked from reflecting to the bottom surface, reducing internal MSSW circulation within the YIG waveguide. The oblique edges of the triangular YIG structure suppress standing-wave formation by scattering the propagating MSSW into directions that follow the MSBVW condition. Because these scattered waves lie within the stopband of the MSBVW, they rapidly decay^37–39. Additionally, in the triangular regions, a large portion of the YIG lies near the etched sidewalls at both ends. The increased surface roughness of these sidewalls introduces extra damping and attenuation of MSSWs, preventing reflection toward the bottom interface and further reducing internal wave circulation^40–43. This added loss mechanism enhances unidirectional propagation, leading to a significant improvement in nonreciprocity ($∣S_{12}-S_{21}∣$). Moreover, in a rectangle shape YIG filter, a strong spurious mode peak is observed near 8.4 GHz. This is attributed to the excitation of a higher-order width mode and MSBVW generated by the center bridge of the meander-line transducer, as this frequency lies below ${\omega }_{\min }$ for MSSW waves in Fig. 1a. The dual-hexagon design mitigates this issue by introducing width variation along the YIG waveguide, which disrupts the formation of these spurious modes⁴⁴. Supplementary Note #9 provides a comparison of these improvements across a range of operating frequencies.

Integrated devices

The tunable magnetic bias circuit comprises NdFeB permanent magnets, soft magnetic yokes, AlNiCo programmable magnets and solenoid coils wound around the AlNiCo magnets. The NdFeB magnets provide a constant magnetic flux, while the AlNiCo magnets, with lower coercivity, serve as a tunable source that can be magnetized or demagnetized by current pulses through the coils, retaining remanent flux after each pulse. A 3D magnetic field simulation of our previously reported bias circuit shows that only approximately 10% of the magnetic flux generated by the NdFeB and AlNiCo magnets passes through the yoke poles and is used for biasing the YIG device, while the remainder leaks into the air surrounding the yokes²³. To improve the tuning range of the magnetic bias circuits, a possible approach is to reduce leakage flux. In this study, the surface area of the yoke pieces has been reduced by narrowing the width of the yokes, corresponding to the vertical dimension of the rectangular yoke base in Fig. 1c. At the same time, to prevent magnetic saturation in the smaller yokes, an iron-cobalt magnetic alloy with a high saturation flux density of 2.4 T is employed. The designed magnetic bias circuit is shown in Fig. 1c. Figure 5a shows the measured tuning range of −25 Gauss to 5700 Gauss achieved, which is approximately twice that of a previous report¹⁰, while the total volume of the bias circuit is reduced to 1.07 cm³.

Fig. 5. Integrated device.

a Measured maximum magnetic flux density at the center of the magnetic bias circuit gap under different applied current pulses. b Measured S₁₂ frequency responses with the magnetic field supplied by the zero-static-power magnetic bias circuit. The YIG filter used in this study features a dual-hexagon design with an 18 µm thick YIG film and two-parallel meander-line transducers. The filter has a pitch of 70 µm, a waveguide width of 200 µm, and transducer widths of 10 µm, with no additional spacing between the input and output transducers. c Magnified measured S₁₂ and S₂₁ frequency responses showing the flat passband and large non-reciprocity of the tunable filter.

The YIG filter shown in Fig. 4d has been integrated into the magnetic bias circuit, with the complete integrated device shown in Fig. 1c. Figure 5b shows the measured S₁₂ frequency response spanning from 4 GHz to 17.7 GHz. Figure 5c focuses on the passband responses at four different frequency points, demonstrating a flat passband with over 25 dB of isolation. Supplementary Note #10 discusses the power handling capabilities of the integrated filter. The filter achieves a 1 dB compression point (P1dB) more than 10.4 dBm, significantly improved from the P1dB of −17 dBm for a YIG filter with a 3 µm thick film¹⁰. This enhancement results from the increased YIG thickness and a larger waveguide area, as similar meander-line YIG filters realized in 3 µm thick films, exhibit P1dB of around 0 dBm. Supplementary Note #11 also shows the in-band linearity of the filter, which exhibits an input-referred third order intermodulation intercept point (IIP3) of more than +26 dBm. Supplementary Note #12 shows the details of the measured frequency response, including S₁₁, S₂₂, S₂₁, and group delay at various frequencies.

Supplementary Note #13 details the time and energy requirements for frequency tuning of the integrated device. The maximum switching duration is ~100 µs, requiring ~116 mJ to tune across the full 4–18 GHz range (0–5700 Gauss). Supplementary Note #14 examines the repeatability of the tunable filter. Although the AlNiCo magnet exhibits hysteresis, the filter demonstrates excellent frequency tuning repeatability when a reset (demagnetization) process is performed prior to setting the desired frequency.

Discussion

In this work, we have demonstrated a class of miniature, narrowband, and tunable bandpass filters that offer zero static power consumption, low insertion loss ( < 5 dB), high out-of-band rejection ( > 30 dB), and substantial nonreciprocity ( > 25 dB). Figure 6 benchmarks the performance of this study with prior work on nonreciprocal filters^{25,38,45–55}, with a detailed comparison table available in Supplementary Note #15. To the best of our knowledge, all previous nonreciprocal filters have been limited to frequencies below 9 GHz. This study marks the first demonstration of nonreciprocal filters operating beyond 9 GHz—extending up to 18 GHz, while also introducing wideband frequency tunability. This advancement paves the way for potential deployment in the emerging FR3 Band (7 GHz to 24 GHz), which has already been used for satellite communication and is expected to be the band for future 6 G wireless communications networks^5,6.

Fig. 6. Comparison of this study with nonreciprocal filters based on YIG, Ferrites, Spatiotemporal Modulation (STM), PIN diode switches (Switch), and Transistors.

a Insertion loss (b) Bandwidth (c) Area (d) DC power consumption (e) Minimum out-of-band rejection—for each filter in references, this is plotted versus its center frequency. For this study, the out-of-band rejection is shown across the 2–18 GHz frequency range. f Center frequency isolation ($∣S_{12}-S_{21}∣$).

This nonreciprocal bandpass filter plays an important role in both the transmit and receive paths, as illustrated in Fig. 1a. In the receive path, wideband LNAs and wideband antennas often exhibit significant impedance mismatches across the 4–18 GHz range, resulting in partial reflections that can form standing waves. The reflected power can re-enter the LNA input, causing ripple and oscillations, ultimately degrading receiver sensitivity. By incorporating the nonreciprocal filter, reflected in-band signals from the amplifier or antenna are absorbed within the device rather than reflected, ensuring that unwanted energy is dissipated internally and does not re-enter the signal chain. This is confirmed by the high return loss (S₁₁) observed at the filter’s output port. In the transmit path, the nonreciprocal bandpass filter effectively isolates the DAC from the PA by absorbing backward-propagating energy and suppressing out-of-band components. This absorptive isolation stabilizes the impedance between stages, minimizing distortion and feedback. The filter thus provides a compact and efficient solution that simultaneously achieves frequency selective transmission and absorptive isolation. Although placing the nonreciprocal filter between the PA and antenna could further enhance system stability, this configuration requires higher power handling capability. Future work will therefore focus on increasing the device’s power tolerance from the current ~11 dBm to above 20 dBm to enable direct integration at the PA output.

While some prior works using active devices have reported lower insertion loss or even power gain (i.e., negative insertion loss)^53,54, those solutions are not frequency tunable and suffer from the noise of the active circuits. To achieve wide frequency coverage, they must rely on multiple RF switches—an approach that significantly increases system complexity and contributes to additional insertion loss^8,9. Furthermore, our study offers the highest selectivity within the 2–18 GHz range, with a narrow bandwidth of 100–200 MHz. This level of selectivity is well-suited for modern communication systems. For instance, 5 G systems specify a maximum bandwidth of 100 MHz in the sub-6 GHz range and up to 400 MHz in the millimeter-wave bands⁵⁶. Moreover, although transistor based nonreciprocal filters can offer a smaller footprint, they typically require hundreds of milliwatts of DC power for operation^52,54. In contrast, the magnetic bias circuit developed in this study achieves a compact form factor, occupying 4 times less area and 82 times less volume than the electromagnets reported in ref. ²⁵, while eliminating the need for power-hungry electromagnets that consume 57.6 W of static power.

Even when disregarding the benefits of nonreciprocity, this study outperforms previous YIG-based reciprocal filters, as shown in Supplementary Note #16. It demonstrates lower insertion loss, superior out-of-band rejection, a sharper filter skirt, larger power handling and substantially higher linearity. Notably, this study achieved almost 30 dB of rejection when the frequency is 1.6 times the bandwidth above the center frequency and higher frequency spurious modes are eliminated. The enhanced filter skirt is attributed to the unique dispersion characteristics of the 18 µm thick YIG and the spurious mode attenuation provided by the dual-hexagon shaped YIG.

Given that the filter demonstrated in this work simultaneously achieves a narrow frequency band, excellent out-of-band rejection, a flat passband, and minimal group delay variation, the most comparable existing technology is the surface acoustic wave (SAW) delay line. Similar to how SAWs propagate from input-to-output interdigitated transducers (IDTs) on piezoelectric substrates, this study utilizes MSSW excited by aluminum meander-line transducers on a YIG film to achieve signal delay. However, this work offers several advantages beyond what SAW delay lines can provide: (1) Frequency turnability. SAW delay lines have fixed operational frequencies determined by lithographically defined IDTs and lack electrical tunability. (2) Nonreciprocity. While SAW delay lines can exhibit nonreciprocal behavior via the acoustoelectric effect⁵⁷ and helicity mismatch⁵⁸, these require additional fabrication steps and typically involve DC power consumption. (3) Low insertion loss. Even though SAW filters have been demonstrated with low insertion loss⁵⁹, state-of-the-art SAW delay lines with low loss are generally limited to below 6 GHz and exhibit insertion losses around 7 dB⁶⁰. In contrast, this study demonstrates operation up to 18 GHz while maintaining an insertion loss below 5 dB.

Although only one thin YIG film (3 µm) and one thick YIG film (15−18 µm) were examined in this study due to limits on commercially available YIG thicknesses, the BCB planarization fabrication method demonstrated here can be readily adapted to YIG films of other thicknesses. Furthermore, this work provides the first comprehensive experimental and theoretical analysis of the thickness effect in microfabricated YIG films. The theoretical calculation results show no abrupt transitions as thickness varies. Instead, all thickness effects evolve gradually. Moreover, the methodologies used here, including the dispersion relation, power handling, linearity, nonreciprocity, and propagation loss, can be directly applied to future investigations of YIG devices with intermediate or alternative thicknesses.

Methods

MSSW filter fabrication

Initially, a < 111> oriented 15–18 µm thick YIG was grown on a GGG substrate using liquid epitaxy, prepared by MTI Corporation, with a ferromagnetic resonance linewidth of 0.5–2.0 Oe. The difference between 15 µm and 18 µm YIG films is minimal and does not significantly influence the device performance. The YIG film used in Supplementary Note #2 (Thick YIG fabrication process) was ~15 µm thick, while all other experiments in the main text utilized the 18 µm film. In the first step, a 500 nm thick SiO₂ layer was deposited as a hard mask via atomic layer deposition using Bis(diethylamino)silane (BDEAS) and O₃ (Cambridge Nanotech S200). The sample was then annealed at 600 °C for 30 min in a nitrogen environment. The hard mask was patterned using standard photolithography and CHF₃ dry etching (Oxford 80 Plus RIE). After removing the photoresist, the mask pattern was transferred to the YIG through wet etching, utilizing phosphoric acid at 140 °C. The remaining SiO₂ hard mask was stripped using Buffered Oxide Etchant (BOE) 6:1, a mixture of a buffering agent and hydrofluoric acid. To achieve surface planarization of the YIG substrate, photosensitive benzocyclobutene (BCB) (CYCLOTENE 4026-46) was spin-coated onto the YIG surface. Following spin coating, the BCB layer was exposed to UV light and patterned using AP3000 developer. The regions of BCB not covering the YIG were crosslinked upon UV light exposure and remained, while the unexposed BCB directly on top of the YIG was dissolved in the developer (see Fig. 1d). The patterned BCB was then thermally annealed at 350 °C for 2 h to complete cross-linking and fully cure the film, forming a stable dielectric layer. To pattern RF electrodes, a 2 μm thick Al layer was deposited by sputtering at 1000 W from a 100 mm diameter Al target at a base pressure of 1e^-7 mbar (Evatec Clusterline 200 II) at 150 °C. This was followed by depositing a 300 nm thick SiO₂ layer using plasma-enhanced chemical vapor deposition (Oxford PlasmaLab 100). Standard photolithography was again employed to pattern the photoresist, followed by etching the SiO₂ using CHF₃ RIE dry etching. The photoresist layer was removed with 1165 solvent, and the Al layer was patterned using an ICP Etcher (Oxford Cobra ICP Etcher) with Cl₂/BCl₃ gases. Finally, the SiO₂ hard mask was removed using CF₄ RIE with the same RIE tool.

Magnetic circuit fabrication

The yoke pieces, the NdFeB magnets and the AlNiCo magnets were prepared separately and then assembled. The yokes were cut from a 3.2 mm thick Hiperco 50 A cobalt-iron sheet using a wire EDM (electric discharge machining) process. Each yoke has a 20.0 mm × 2.0 mm rectangular base, with a 5.85 mm tall, tapered pole piece extending from the center. The pole piece is 3.0 mm wide at the base and narrows to 2.0 mm at the top. The NdFeB permanent magnets have a cylindrical shape with a diameter of 3.18 mm and a length of 12.7 mm, and were purchased from K&J Magnetics, Inc. The left and right NdFeB magnets are grades N42 and N52, respectively. The AlNiCo 5 magnets, identical in size to the NdFeB magnets, were purchased from DigiKey. A 49-turn coil of 32-gauge copper wire, coated with a polyimide insulation layer, was manually wound around each AlNiCo magnet. After preparing all the parts, the yokes, NdFeB magnets, and AlNiCo magnets were assembled on a 3D-printed substrate and fixed using cyanoacrylate adhesive.

Measurement setup

The YIG sample was first characterized using a magnetic probe station (MicroXact’s MPS-1D-5kOe). The magnetic field was generated by electromagnets inside the magnetic probe station. A Gaussmeter (Model GM2, AlphaLab Inc) was used to calibrate the magnetic probe station. Due to potential variations in sample placement across measurements, the magnetic field experienced by the device may not have been entirely consistent in this study.

For the integrated device, the magnetic field was produced by the magnetic bias circuit, eliminating the need for the electromagnets. To tune the fabricated magnetic circuit, current pulses with different amplitudes and a consistent duration of 0.5 ms were applied to the coils wound around the AlNiCo magnets using a DC electronic load (EL34243A, Keysight) and a pre-charged large 65 F supercapacitor (XVM-16R2656-R, Eaton). The two coils are connected in series, allowing a single current pulse to simultaneously magnetize both AlNiCo magnets. After each current pulse, the magnetic field within the air gap between the yoke tips was measured using a gaussmeter.

Filter performance was measured using a Keysight vector network analyzer (VNA) P5026B with a power level of −20 dBm with 50 Ω port impedances, unless otherwise noted. Prior to measurement, a two-port calibration to the probe tips was performed within the desired frequency range using the Short-Open-Load-Through (SOLT) method. The ground-signal-ground (GSG) probe used had a pitch of 150 μm from GGB industries. There is no on wafer de-embedding utilized in this study. Filter parameters were manually extracted from the measured S-parameter data.

Author contributions

X.D. and R.O. developed the device concepts and experimental implementations. X.D., S.Y., Yijie D., S.W., and C.C. fabricated the YIG filters under the supervision of R.O. Yixiao D. designed and fabricated the magnetic bias circuit under the supervision of M.A. Yixiao D., D.L., and X.W. performed the magnetic bias circuit measurements, and M.A. supervised them. X.D., S.Y., D.L., and S.W. performed the filter measurements, and R.O. supervised the measurements. X.D., Yixiao D. and R.O. analyzed all data and wrote the manuscript. All authors have given approval to the final version of 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.

Data availability

All data supporting the findings of this study are available within the article and its supplementary files. Any additional requests for information can be directed to, and will be fulfilled by, the corresponding author.

Competing interests

The authors declare the following competing interest: a provisional patent application related to the YIG filter design has been filed in the United States by Xingyu Du, Shun Yao, Shuxian Wu, Roy H. Olsson III. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at (10.1038/s41467-026-68289-4).