Spatially tailored spin wave excitation for spurious-free, low-loss magnetostatic wave filters with ultra-wide frequency tunability Authors: Shuxian Wu, Shun Yao, Xingyu Du, Chin-Yu Chang, Roy H Olsson Published: 2026 ABSTRACT AbstractYttrium iron garnet (YIG) magnetostatic wave (MSW) radio frequency cavity filters are promising for sixth-generation (6G) communication systems due to their wide frequency tunability. However, the presence of severe spurious modes arising from the finite cavity dimensions severely degrades the filter performance. We present a half-cone transducer that spatially tailors spin wave excitation to selectively enhance the primary cavity modes comprising the YIG-based MSW filter passband, while strongly suppressing the undesired spurious modes. Theoretical analysis, numerical simulations and experiments verify the effectiveness of the spatially tailored technique. We utilize the half-cone transducer to demonstrate a spurious-free, single-cavity half-cone MSW filter (HC-MSWF) with an insertion loss (IL) of 2.4–3.2 dB over a frequency tuning range of 6.3–16.8 GHz. Extending our study, we further demonstrate a spurious-free, dual-cavity HC-MSWF with an unprecedented tuning range of 21.7 GHz (9.8–31.5 GHz) while maintaining a low IL of 2.9–3.8 dB. This significant advance in performance will enable highly reconfigurable and robust 6G networks. Subject terms: Electrical and electronic engineering, Structural properties Introduction State-of-the-art reconfigurable radios, such as cellular handsets, rely on a switched bank of more than 50 filters to operate across multiple bands while suppressing interference1. With the rapid evolution and continuous expansion of wireless communication technologies, especially with the emerging sixth-generation (6G) networks, significantly more fixed frequency radio-frequency (RF) filters are required due to the continuous increase in the number of frequency bands in communication systems2. Tunable RF filters have drawn widespread attention as an alternative to switched banks of fixed frequency filters due to their multi-frequency compatibility across different bands3. Specifically, advanced software defined radio systems4, radar technologies5, base stations6, satellite communications7, and Internet of Things (IoT) devices8 demand tunable filters with low insertion loss, spurious-free responses and high selectivity to ensure reliable and efficient spectrum operations. Tunable RF filters dynamically select operational frequencies by employing various tuning mechanisms, thus significantly enhancing system flexibility and spectral efficiency. Varactor-based tunable filters typically leverage the voltage-dependent capacitance of semiconductor devices such as varactor diodes, PIN diodes, or field-effect transistors (FETs) to alter the resonant frequency of the filters. Tunable filters based on varactors offer advantages such as small size, low power consumption and fast tuning speed9–11. However, their frequency tuning ratio is generally limited to below 2:1, and the insertion loss is relatively high due to the low resonator quality factor (Q). Low-Q and high insertion loss are particularly problematic at millimeter-wave and sub-millimeter-wave frequency bands, restricting their applicability in advanced applications. Microelectromechanical systems (MEMS) technology enables the fabrication of miniature mechanical structures that can be actuated to change the physical dimensions or shapes of resonators, thus achieving frequency tuning12–15. MEMS-based tunable filters exhibit potential for small size, high integration and low power consumption, but challenges remain in realizing high tuning speed and wide frequency tuning ratio of more than 2.5:1. Surface acoustic wave (SAW) and bulk acoustic wave (BAW) filters are widely used in RF front-ends due to their compact dimensions, low fabrication costs and high Q value16,17. Frequency tuning can be achieved by incorporating variable capacitor or switching elements onto SAW/BAW resonators18,19. However, SAW/BAW filters typically experience significant performance degradation at high frequency bands, and their frequency tuning range is extremely narrow, being only a few percent of the center frequency. Research on the generation and propagation of spin waves in ferromagnetic films has promoted the development of spin-based device technology20. Tunable filters based on spin waves using ferrite materials such as yttrium iron garnet (YIG) exhibit low insertion loss, high Q, and extremely large frequency tuning ranges from several GHz to tens of GHz when subjected to tunable external bias magnetic fields21–24. These unique properties enable YIG-based tunable filters to provide outstanding performance, especially at microwave and millimeter-wave frequencies, making them highly attractive for next-generation communication systems. Commercially available YIG filters typically rely on the manual alignment of coupling transducer loops around YIG spheres, which significantly hinders scalability25. Additionally, the traditional design results in large volume and high power-consumption due to the substantial currents required by the electromagnets used to tune the magnetic bias field and filter frequency. Recently, single-crystal thin-film YIG resonators and filters with small magnetic damping have been demonstrated, resulting in lower filter insertion loss and smaller size when compared with YIG sphere-based filters26–33. An edge-coupled magnetostatic forward volume wave bandpass filter (MSFVW) was demonstrated with an insertion loss of 6.94 dB from 4.5 GHz to 10.1 GHz28. A microfabrication deep-etching methodology was studied to enlarge the coupling of thin film YIG magnetostatic wave resonators29. To solve the problem of high power consumption, a YIG film-based magnetostatic surface wave (MSSW) filter was proposed with relatively wide frequency tuning range from 3.4 GHz to 11.1 GHz, insertion loss of 3.2–5.1 dB, and zero static power consumption by developing a novel tunable magnetic bias circuit31. YIG tunable filters have demonstrated frequency tunability from 4 to 18 GHz along with nonreciprocal behavior32. In addition, thick YIG-based two-cavity tunable filters have achieved wide tunability with insertion loss below 5 dB33. However, a significant challenge in the advancement of YIG-based MSSW filters is the presence of severe spurious responses. The unwanted modes in MSSW filters primarily arise from several sources: a) For finite width YIG films, magnetostatic waves not only propagate along the length direction of YIG (x-direction in Fig. 1a) but also form standing wave modes across the width direction of the film (y-direction in Fig. 1a), which may generate unwanted width modes34. b) The strong dispersion of magnetostatic waves in YIG-based bandpass filters can lead to the unintended excitation of higher-order modes due to the frequency-dependent variation in phase and group velocities, resulting in the excitation of spurious responses35. c) Reflections of MSSW at the input/output ports and the edges of the film may interfere, causing some spurious responses36. Such spurious responses degrade filter selectivity and spectral purity, severely impacting their practical applicability in advanced RF systems. Design and realization of half-cone magnetostatic wave filters (HC-MSWF). a 3-D geometry and cross section of the proposed HC-MSWF. A 15 μm thick YIG film grown on top of a Gadolinium Gallium Garnet (GGG) substrate is patterned by wet etching using phosphoric acid, and a negative photoresist Benzocyclobutene (BCB) is laid on top of the YIG for planarization. The Al transducers are patterned by dry etching for the conversion between electromagnetic waves and magnetostatic waves. An external bias magnetic field (H⃗0) is applied to the YIG thin film. b SEM image of the fabricated single-cavity reference straight-line bandpass filter. c SEM image of the fabricated single-cavity bandpass filter with half-cone spatially selective transducer. d The measured frequency responses of the reference bandpass filter using straight-line transducers. It can be observed that there are many spurious modes on the left side of the filter passband, and the insertion loss increases significantly at higher frequency bands. e The measured frequency responses of the HC-MSWF. All spurious modes in the HC-MSWF are effectively suppressed and the insertion losses at all frequency bands are lower than that of the reference filters Researchers have explored various techniques to mitigate the adverse effects of spurious modes in magnetostatic wave (MSW) filters. High-quality YIG films grown by liquid phase epitaxy can reduce intrinsic losses and spurious mode generation37. Designing specialized input/output coupling structures, such as multi-section couplers or rotated YIG slabs with a microstrip transducer pair, can suppress the excitation of spurious modes38,39. Although these techniques have demonstrated success in reducing spurious modes, they often complicate filter design and fabrication procedures, and inadvertently increase insertion loss, thus presenting challenging trade-offs, particularly across extremely wide tuning ranges. Previous studies exploited the unique dispersion characteristics of 15–18 μm thick YIG thin films and demonstrated partial attenuation of spurious modes at high frequency bands, achieving 28 dB rejection at frequencies 0.9 times the bandwidth above the center frequency32,33. However, spurious suppression over a wide tunable range remains a critical unresolved challenge. Moreover, there is a research gap in numerical analysis of the MSSW modes and the corresponding spurious suppression techniques for micromachined thin-film YIG filters to date. Effective mitigation of these spurious responses is crucial for realizing the full potential of thin film YIG-based tunable filters. Apodization, also known as aperture or electrode weighting, is a technique employed to optimize the frequency response or spatial radiation characteristics of transducers in acoustic applications such as SAW and BAW filters40–42. Through the deliberate design of the electrodes, the amplitude/phase or acoustic velocity distribution of excitation waves along the transducers is modified to guide the coupling of waves. This technique effectively suppresses the spurious responses of acoustic filters, significantly enhancing the selectivity without notably increasing insertion loss or manufacturing complexity43–46. The proven effect of apodization techniques in acoustic filters provides strong motivation for addressing the similar spurious suppression challenges in MSSW filters. In this study, inspired by the successful application of apodization techniques in acoustics, we propose a novel spatially tailored half-cone transducer for MSSW bandpass filters, demonstrating suppression of spurious modes over an unprecedented frequency tuning range of 6.3–31.5 GHz. Figure 1a depicts a 3-D schematic geometry and a cross-section of the half-cone magnetostatic wave filter (HC-MSWF), where L represents the length of the cavity in the direction of MSSW propagation (x-direction), and W represents the width of the cavity along the length direction of the transducers (y-direction). A 9 μm thick negative photoresist Benzocyclobutene (BCB) is laid on top of YIG for planarization. To comprehensively analyze and visualize the mode shapes in thin-film YIG filters, finite element method (FEM) simulations are conducted in Ansys High Frequency Structure Simulator (HFSS) software platform, providing a powerful guide for the spatially selective transducer technique. By combining Maxwell’s equations and Landau-Lifshitz-Gilbert equation analysis, a half-cone transducer is introduced to tailor the excitation of spin waves. The current-density distribution of the half-cone transducers is modulated to excite the desired primary modes and to suppress the spurious modes. Meanwhile, the half-cone transducer can reduce in-band insertion loss due to better impedance matching and lower ohmic loss. Simulation results demonstrate that the half-cone transducer can effectively suppress undesired and strong width modes in conventional MSSW filters using straight-line (unapodized) transducers. To further validate the feasibility and effectiveness of this novel approach, single-cavity and dual-cavity MSSW bandpass filters using the half cone spatially selective transducer are fabricated. Scanning electron microscope (SEM) images of the manufactured single-cavity reference straight-line and half-cone bandpass filters are shown in Fig. 1b, c, where the two pads located on either side of the YIG cavity implement the RF input/output transducers. The white area indicates the YIG thin film cavity which is surrounded by a BCB layer with 9 μm thickness, and the Al layer is routed on the top of the YIG layer successfully. The measured frequency responses of single-cavity filters at different bias magnetic field intensities implemented with traditional straight-line transducers and spatially tailored half-cone transducers are depicted in Fig. 1d, e. Experimental results closely match the simulations, highlighting significant improvements in spurious mode suppression. The half-cone transducer technique successfully suppresses all spurious modes and reduces the insertion loss across multiple frequency bands. The insertion loss of filters in this study is 2.4–3.8 dB across the extremely broad frequency tuning range of 6.3–31.5 GHz, indicating superior performance compared to switched filter banks or alternative YIG thin film-based tunable filter technologies. Results Magnetostatic surface wave cavity behavior MSWs in ferrite media exhibit multiple distinct propagation characteristics depending on the orientation of an external bias magnetic field (H⃗0) relative to the wave vector (k⃗) and the normal vector (n⃗) of the surface of the ferrite slabs. Damon and Eshbach originally categorized the MSWs into three distinct propagation modes, including MSSW, magnetostatic forward volume waves (MSFVW), and magnetostatic backward volume waves (MSBVW), which are foundational in understanding MSW dynamics47. For a straight MSSW resonator based on thin-film YIG, when the bias magnetic field is perpendicular to the wave propagation direction, surface waves propagate along the surface of the YIG film and reflect at the straight boundaries onto the other surface48. A circular wave pattern is established when the following resonance condition is satisfied49 where kxn is the average wavenumber on the top and bottom surfaces, and n is an integer representing the order of the spin waves in the x direction (primary modes). As L increases, the spacing between the different primary modes gets smaller due to the strong dispersion of the MSSWs. When the thickness of the YIG thin film is much smaller than the width, a width mode can be excited in the y-direction. The wavenumber (kym) can be described as a simple expression: where m represents the order of spin waves in the y direction (width modes). In YIG thin-film–based filters, the interaction between transducer-induced currents and magnetization dynamics is governed by the coupled Maxwell’s equations and the Landau–Lifshitz–Gilbert (LLG) equation50–52. Current density (J) in patterned transducers generates an oscillating magnetic field (Hrf), as described by Ampère’s law in Maxwell’s equations: where r is the spatial domain variable, t is the time domain variable, J is the electric current density, and D is the electric displacement. This field penetrates the YIG film and acts as the primary excitation source for magnetostatic spin waves. The temporal evolution of the local magnetization (M) is then dictated by the LLG equation under the influence of an effective magnetic field (Heff): where μ0 is the vacuum permeability, γ represents the gyromagnetic ratio, and α is the Gilbert damping constant. The first term of the LLG equation provides an expression for the frequency dependence of the magnetizing response. By combining the expression of magnetic flux density (B = μ0(H + M)), the component form of the lossless permeability tensor (μ) in the YIG film can be characterized as: where μ1=1−ΩHΩ2−ΩH2, μ2=ΩHΩ2−ΩH2, Ω=ωγMs, ΩH=H0Ms, ω is the angular frequency of the magnetostatic waves, γ is the gyromagnetic ratio (γYIG = 2.8 × 106 Hz/Gauss), and Ms represents the saturation magnetization of the YIG thin film. In free space, the magnetic permeability is described by a scalar quantity with a normalized value of unity; accordingly, as expressed in the notation of Eq. 6 this corresponds to μ1=1 and μ2=0. For the MSSW bandpass filters in this study, a 15 μm thick YIG thin-film is grown on top of a Gadolinium Gallium Garnet (GGG) substrate. Two transducers are adopted to convert energy between electrical currents and magnetostatic waves. The external bias magnetic field parallel to the transducers is applied to the YIG thin film. In such a setup, MSSW will be excited and propagate in a direction perpendicular to the transducers. To clearly observe the frequency changes with wavenumber in the x-direction for the primary modes and width modes in the YIG thin-film, the theoretical derivations of MSSW dispersion and analysis at a magnetic field intensity of 2500 Gauss are described in Part 1 of the supplementary materials. Magnetostatic wave cavity mode shapes In the study of acoustic filters, modal analysis helps guide the design and thus improves the performance of filters. However, there has been little work on modal analysis for MSW cavity devices to date. In this work, a three-dimensional (3-D) electro-magnetostatic wave simulation model is built for the cavity bandpass filters using a finite element method in the Ansys HFSS platform. The top view of the 3-D simulation model is depicted (Fig. 2a), where the resonant body, including a GGG layer, a YIG thin film layer (yellow), and an Al layer (dark blue), is placed in the middle area. The BCB layer is not included in the model to reduce computational complexity, as the BCB layer has almost no effect on the magnetostatic wave propagation. The length and width of the YIG are set as 280 μm and 400 μm to maximize the propagation of the primary modes. The thickness of the GGG substrate is 500 μm and the thickness of the YIG thin film is 15 μm. The thick YIG thin film is chosen due to the demonstrated advantages to minimize spin-wave propagation loss, enhance magnetic coupling efficiency, and improve power handling capacity, owing to the increased interaction volume of magnetic material with the magnetic fields produced by the transducers and thermal robustness of thicker films32,33. A 2 μm thick Al film with a width of 10 μm is located at the L/4 position of the YIG to optimize the coupling efficiency and out-of-band rejection capability of the filters. Simulations and magnetostatic wave mode shapes of the YIG thin film cavity MSWF. a Top view of the simulation model of the MSWF cavity filters, including a GGG layer, a YIG cavity layer (yellow) and an Al layer (dark blue). Two wave ports with perfect electric conductors (PEC) are applied on both sides of YIG for the radio-frequency signal input and output. b Simulated frequency response of the conventional straight-line transducer (reference) cavity bandpass filter. The different magnetostatic wave modes, including primary modes and width modes are labeled in the frequency response, where “p” refers to the order of the primary modes and “w” refers to the order of the width modes. c Magnetic field vector shapes in the z-direction are shown for the primary and width magnetostatic wave modes with different orders in the reference cavity MSWF, where the units of all color bars are Gauss Two wave ports and ground-signal-ground (GSG) pads are located on both sides of the resonant body for input/output RF electrical signals. Compared to lumped port simulation, wave port simulation is more appropriate to observe the wave propagation that needs to be suppressed to enhance filter performance. To facilitate identification of different modes, we denote the primary modes as “p”, the width modes as “w”, and the subsequent numbers express the order of the modes. For instance, “p2w3” refers to the third-order width mode of the second-order primary mode. The simulated frequency response of the conventional straight-line transducer (reference) cavity bandpass filter is shown in Fig. 2b. The different magnetostatic wave modes, including primary modes and width modes are labeled in the frequency response. The magnetic field vector distributions of the nine resonance modes along the thickness direction (z-direction) of the YIG thin film are observed (Fig. 2c). It can be observed that the desired primary modes which form the passband response are predominantly concentrated in the central region in the y-direction, whereas the width modes forming the undesired spurious modes extend towards the two edges of the YIG film. Consequently, a spatially selective excitation technique by tailoring the shape of the transducers is proposed to enhance the primary mode coupling (w = 1) while suppressing generation of width modes (w > 1). Half-cone spatially selective MSW excitation In conjunction with Eqs. (3)–(5) the current density of transducers affects M indirectly by generating Hrf, which precesses the magnetization vector and excites spin waves in YIG. The spatial distribution of current density directly determines the local excitation efficiency and modal selectivity of the spin waves. Therefore, a spatially tailored half-cone transducer is designed to modulate the current density distribution for selectively exciting the desired spin waves (w = 1) while suppressing spurious responses (w > 1). Through the previous simulations, various propagation modes can be clearly observed. The magnetic field vector distribution of the fourth-order primary mode with first-order width mode (p4w1) in the reference cavity bandpass filter is illustrated (Fig. 3a). The propagation of magnetostatic waves predominantly concentrates in the middle region of the YIG thin film along the width direction. To suppress the width modes as mentioned earlier, three spatially tailored transducer techniques, including a full-cone, an extended cone, and a half-cone are studied by tailoring the shapes of the transducers in this work. Here we focus on the introduction of full-cone and half-cone techniques, and the extended cone technique is introduced in the supplementary materials. By precisely adjusting the electrode width of the transducers, the current density distribution changes along the length of the transducers, consequently affecting the dynamic magnetization within the YIG film and the mode coupling characteristics. Detailed information about the current density distribution is introduced in Part 2 of the supplementary materials. When the current density at the transducer edges is lower than that at the central region, the width modes generated across the width direction are weakened, thereby effectively suppressing the width spurious modes. Comparison of the simulated magnetic field vector shapes for the high-order primary mode (p4w1) in the reference straight-line cavity filter, FC-MSWF and HC-MSWF, where the units of all color bars are Gauss. a Mode shape of the reference straight-line cavity bandpass filter. b Mode shape of the full-cone transducer cavity bandpass filter. c Mode shape of the half-cone transducer cavity bandpass filter. d Simulated frequency responses of the three cavity bandpass filters. It can be observed that both the full-cone and half-cone transducers have a strong spurious suppression ability. The full-cone transducer increases the insertion loss and worsens the passband performance of filters, while the insertion loss of filters with the half-cone transducer matches the reference straight-line filters while suppressing the width spurious modes below the passband frequency First, a full-cone transducer is adopted at both edges of the two transducers. In this case variation of the electrode width causes the spacing between the two transducers to reduce, resulting in a difference of the MSSW phase at the receiver transducer. Consequently, high-order primary modes (e.g., p4w1 and p5w1) propagating from the edge of the full-cone region of one transducer toward another transducer are affected. The p4w1 mode shape of the full-cone MSWF is destroyed with weaker magnetostatic wave coupling compared to the reference straight-line filter (Fig. 3b), leading to increased insertion losses for the high-order primary modes that make up the upper frequency region of the filter passband. Fortunately, the half-cone transducer effectively addresses this issue (Fig. 3c). Due to the consistent spacing between the two transducers, the half-cone transducer not only suppresses the generation of width modes but also preserves the propagation characteristics of the higher-order primary modes. To compare the effects of the three transducers, the simulated frequency responses of the reference straight-line filter, full-cone MSWF (FC-MSWF) and HC-MSWF are presented (Fig. 3d). As anticipated, the full-cone transducer suppresses width spurious modes but introduces a higher insertion loss at the high end of the passband. While the half-cone transducer achieves effective spurious mode suppression and reduces the insertion loss simultaneously, thereby improving the overall filter performance. To further achieve a better spurious suppression and obtain a flatter passband response, the dimensions of the cavity HC-MSWF are optimized. The detailed information about HC-MSWF optimization is shown in Part 3 of the supplementary materials. The positions of width modes are shifted for the FC-MSWF and HC-MSWF cavities owing to the difference in the metal coverage of the two designs, as metalized surfaces shift the dispersion and MSSW velocity. Detailed information is shown in Part 4 of the supplementary materials. Experimental results A scanning electron microscope (SEM) image of the fabricated single-cavity bandpass filter featuring half-cone transducers is shown in Fig. 1c. The white area indicates the YIG thin film cavity which is surrounded by a BCB layer with 9 μm thickness, and the Al layer is put on the top of the YIG layer successfully. The fabricated devices are measured on a magnetic probe station using a vector network analyzer (VNA), which is described in detail in Part 5 of the supplementary materials. The measured S₁₂ frequency responses of the two bandpass cavity filters are analyzed, where the blue traces correspond to the reference straight-line bandpass cavity filter and the red traces represent the half-cone bandpass cavity filter (Fig. 4a–d). It can be observed that the half-cone transducer design effectively suppresses all width spurious modes (w > 1) of the cavity filter. The suppression level of spurious cavity modes is up to 15 dB. A comparison of the insertion loss (IL) for the two bandpass cavity filters is depicted (Fig. 4e). The insertion loss of the HC-MSWF (2.4–3.2 dB) is lower than that of the reference bandpass filter (2.6–5.2 dB), exhibiting a maximum reduction in IL of about 2 dB. The reduction of insertion loss arises from lower ohmic losses and a decrease of the equivalent electrode impedance (Zs) in the filter with the half-cone geometry, leading to improved impedance matching for the filter input/output ports, which is especially pronounced at high frequencies. A more detailed analysis comparing the loss of filters with half-cone and straight-line transducers is given in Part 4 of the supplementary materials. Measured frequency responses (S12) of the reference straight-line and the half-cone bandpass filters at different magnetic field intensities. a 1500 Gauss. b 2500 Gauss. c 3500 Gauss. d 4500 Gauss. e Insertion losses of the reference straight-line and half-cone cavity bandpass filters. The half-cone transducer effectively suppresses the width modes of the filters, and the insertion loss of the half-cone-based single-cavity bandpass filter is lower than that of the reference bandpass filter due to better impedance matching, especially at high frequency bands Additionally, bandpass filters employing the full-cone and extended-cone transducers discussed before were also fabricated and analyzed in this work, and the comparison of the measured results with the reference straight-line cavity filter is shown in Part 4 of the supplementary materials. By comparing the performance of different structures, it is concluded that the half-cone transducer technique exhibits superior performance in spurious mode suppression and effectively reduces the insertion loss of the bandpass filters, enabling exceptional performance over an extremely wide tuning range. In the single-cavity reference straight-line bandpass cavity filters, the Zs is best matched with standard 50 Ω input/output port impedances around 9 GHz. However, as the filter is tuned to higher frequency bands, the Zs increases accordingly, leading to impedance mismatch. Such mismatches inevitably result in increased insertion loss for the bandpass filters. To achieve low insertion loss bandpass filters at higher frequency bands, a reference straight-line and a half-cone dual-cavity bandpass filter are designed and fabricated. SEM images of the filters are shown, where two YIG cavities are connected in parallel (Fig. 5a, b). The two cavities are not physically connected or magnetically coupled; they are linked together through two transducers and appear electrically in parallel. The direction of the transducer connection between the two cavities (gap = 56 μm) is the width mode resonance direction of the dual-cavity filters, and the main MSW propagates along the direction perpendicular to the transducers. However, the amplitude of the width modes is relatively weak, and the spacing is large, so the magnetic coupling between the cavities can be ignored. The design parameters of the dual-cavity filter are identical to those of a single cavity filter except for the number of cavities. By connecting two cavities in parallel, the total Zs of the filter is effectively halved, and the impedance matching is improved, thereby reducing the IL, especially at high frequencies. Moreover, the half-cone spatially tailored transducer is employed to suppress the spurious responses and to further reduce the IL of the dual-cavity bandpass filters. Fabricated dual-cavity bandpass filters and measured results. a SEM image of the reference straight-line dual-cavity bandpass filter. b SEM image of the spatially tailored half-cone dual-cavity bandpass filter. c The measured S12 frequency responses of the reference dual-cavity bandpass filters. d The measured S12 frequency responses of the half-cone dual-cavity bandpass filters. The half-cone transducer technique effectively suppresses spurious responses at all the frequency bands. e Insertion loss comparison of the reference and half-cone dual-cavity bandpass filters. The insertion loss of the half-cone dual-cavity filter above 15.75 GHz is lower than that of the reference dual-cavity straight-line filter. Using the spatially tailored transducer technique, a spurious-free and low insertion loss bandpass filter is achieved with a wide center frequency tuning range from 9.8 GHz to 31.5 GHz The measured frequency responses of the dual-cavity reference straight-line filters and HC-MSWF under different bias magnetic field intensities are illustrated (Fig. 5c, d). It is evident that the half-cone transducer effectively suppresses spurious responses across all the frequency bands, significantly enhancing the selectivity and improving the performance of the filters. The out-of-band rejection of the dual-cavity HC-MSWF is up to 41 dB at 9.8 GHz. The minimum insertion losses of the two filters over the tunable frequency range from 9.8 GHz to 31.5 GHz are compared in Fig. 5e. For frequency bands above 15.75 GHz, the insertion loss of the HC-MSWF is much lower than that of the reference straight-line bandpass filter. The maximum difference in insertion loss between the two dual-cavity bandpass filters is 2.5 dB. The improved performance for the half-cone design at high frequency bands is primarily attributed to superior impedance matching between the Zs and the port impedance. To highlight the impact of the proposed half-cone structure on filter performance, responses of the fabricated single-cavity bandpass filters before and after conjugate matching are described in Part 6 of the supplementary materials. Furthermore, Smith charts of the measured S11 for the reference straight-line transducer and the HC transducer dual-cavity bandpass filters are depicted and analyzed in Part 7 of the supplementary materials. A performance comparison of the tunable bandpass filters developed in this study and other works is presented in Fig. 6a–d, including single-cavity and dual-cavity configurations compared with previously reported varactor, MEMS, MEMS capacitor, and YIG thin film-based technologies11,14,28,31,53–58. The half-cone transducer bandpass filter has a much larger frequency tuning range and smaller dimensions than other works based on varactor, MEMS, and MEMS capacitors. Moreover, the bandwidth of this work is larger, the frequency tuning range is much larger, and the out-of-band rejection is higher, especially in the higher frequency bands, when compared to other YIG thin film-based tunable passband filters. The single-cavity HC-MSWF in this work demonstrates a wide frequency tuning range of 6.3–16.8 GHz, low insertion losses between 2.4 dB and 3.2 dB and a relatively large 3 dB bandwidth of approximately 245 MHz. Meanwhile, the dual-cavity configuration extends the large frequency tuning capability to 9.8–31.5 GHz with low insertion losses of 2.9–3.8 dB. The out-of-band rejection performance is comparable to that of most existing YIG thin film-based filters in the low-frequency range, and superior in the high-frequency range. To compare the level of spurious responses, the measured S12 of the single-cavity HC-MSWF and a previously reported YIG film-based cavity filter are plotted together in Fig. 6e. The half-cone transducer technique in this study effectively suppresses the spurious modes of the filters and achieves a relatively larger bandwidth, which outperforms existing state-of-the-art works in terms of performance. Performance comparison of varactor, MEMS, MEMS capacitor, and YIG thin film-based tunable bandpass filters. a Insertion loss. b 3-dB bandwidth. c Out-of-band rejection. d Area of the fabricated tunable filters. e Frequency response of the single-cavity HC-MSWF at a magnetic field intensity of 2500 Gauss compared with ref.58. The HC-MSWF has a much larger tuning range and a smaller dimension than other works based on varactor, MEMS and MEMS capacitors. Tunable filters with half-cone transducers effectively suppress spurious modes, in contrast to other reported filters based on YIG thin films. Moreover, the frequency tuning range and bandwidth of this work are larger, and the out-of-band rejection is higher, especially for high frequency bands, when compared to other YIG thin film-based tunable passband filters The experimental results underscore the effectiveness of the half-cone spatially tailored transducer for spurious-free, low insertion loss, wide frequency tuning range, and comparable out-of-band rejection characteristics, positioning the tunable filters described herein as highly suitable candidates for next-generation wireless and RF communication applications. The detailed performance comparison of this study and other works is listed in Table S1 in Part 8 of the supplementary materials. Discussion In this study, we present a novel half-cone transducer for YIG film-based MSW filters, achieving spurious-free and low-loss performance across an extremely wide frequency tuning range of up to 21.7 GHz. This transducer design is informed by Maxwell’s equations and the LLG equation, combined with numerical simulations that clearly identify the mode shapes of the MSSW cavity filters. By tailoring spatial current distributions elaborately, the half-cone transducer can suppress the generation of width spurious modes and spatially couple more energy into the desired spin waves making up the passband of the MSW cavity filter. Both numerical simulation and experiments validate that the half-cone transducer effectively suppresses spurious modes and reduces the insertion loss of single-cavity bandpass filters. Furthermore, a dual-cavity bandpass filter employing half-cone transducers is scaled to higher frequency bands with spurious-free performance. The insertion loss of 2.9–3.8 dB is lower than that of a co-fabricated reference straight-line bandpass filter. This study establishes a framework for tunable bandpass filter design, providing new insights into magnetostatic wave mode analysis and demonstrating the concurrent achievement of spurious modes suppression, broad frequency tuning, and low-loss performance through spatially tailored excitation of spin waves. While the HC transducer can be utilized to suppress spurious modes in smaller YIG cavities, there are two main reasons that smaller YIG cavity MSW filters were not ultimately pursued in this study. First, a smaller YIG cavity increases the proportion of the etched sidewall region with higher roughness, which leads to stronger MSW scattering and degraded performance32. Second, reducing the width increases the fraction of the area influenced by demagnetization near the YIG cavity edges, thereby shrinking the usable region of the YIG cavity32. These effects lead to a lower Q factor when reducing the cavity width, as reported in ref. 31. More details are provided in Part 9 of the supplementary materials. The half-cone transducer suppresses higher-order width modes by spatially tailoring the excitation field, thereby minimizing the overlap between the radio-frequency magnetic field and the profiles of higher-order width modes while preserving efficient excitation of the primary modes. A detailed description of the physical mechanism underpinning width mode suppression is described in Part 10 of the supplementary materials. Spin-wave dispersion does not impose a strict limit on physical size. However, it does set clear limits on the operating frequency. At lower frequencies, such as below ~4 GHz in 15 µm-thick YIG films, the magnetostatic surface wave exhibits strong subsidiary absorption, which significantly reduces power handling and the 1 dB power-compression point of the filter59. Using a material with lower saturation magnetization could help mitigate this issue. At higher frequencies, there is no fundamental cutoff, but the propagation loss becomes higher due to the shorter lifetime of spin waves24. Also, the bandwidth becomes severely restricted at high frequencies. For instance, commercial YIG-sphere filters from MicroLambda exhibit bandwidths of only ~30 MHz at 50 GHz25, which limits their usefulness in some wider bandwidth applications. The spatially tailored spin wave technique in this work can be extended to other magnetostatic wave RF signal processing devices, such as notch filters, delay lines, isolators, and circulators. The suppression of spurious modes in magnetostatic wave filters is indispensable for achieving high selectivity, minimal insertion loss, and reliable performance in advanced communication systems and sophisticated signal-processing applications. Materials and methods MSWF Fabrication The fabrication steps of the YIG cavity filters follow the procedures in32. The starting substrate was a 15 µm thick YIG thin film grown using liquid epitaxy on top of a 500 µm thick GGG substrate. GGG with <111> orientation has an excellent lattice coefficient and thermal expansion coefficient matching with YIG. First, a 500 nm SiO2 film was deposited as a hard mask for YIG cavity etching using atomic layer deposition (ALD) using Bis (diethylamino) silane and trioxygen (O3). After annealing the wafer at 600 °C in a nitrogen atmosphere, the SiO2 layer was patterned using photolithography, development, and dry etching with Trifluoromethane (CHF3) and oxygen (O2) gases. Then, the YIG layer was wet etched at around 150 °C in Phosphoric Acid. The etch rate is approximately 200 nm/min and the etch selectivity of YIG to SiO2 was approximately 10:1. After etching the YIG layer, the remaining SiO2 layer was stripped using 6:1 Buffered Oxide Etchant (BOE). For the 2 µm thick Al electrodes to be successfully laid on top of the 15 µm YIG cavities, a 9 μm thick negative photoresist benzocyclobutene (BCB, CYCLOTENE 4026-46) was spin coated on the YIG and was patterned and developed using DS3000 developer, then was annealed at 300 °C for better stability and planarization. Next, a 2 μm thick Al film was sputter deposited at 150 °C. Then, a 350 nm SiO2 layer was deposited as a hard mask with 40 W RF power using Plasma Enhanced Chemical Vapor Deposition (PECVD). Next, the SiO2 layer was patterned using photolithography and etched with CHF3 and O2 gases by reactive-ion etching (RIE). Finally, the Al layer was etched with boron trichloride (BCl3) and chlorine (Cl2) by inductively coupled plasma (ICP), and the remaining SiO2 was removed by RIE. Measurement setup The characterization of the YIG samples was carried out using a magnetic probe station (MicroXact, model MPS-1D-5kOe). The required magnetic fields were produced via integrated electromagnets within the probe station. To ensure accuracy, the magnetic field strength was calibrated with a precision Gaussmeter. The frequency responses of the filters were obtained using a vector network analyzer (VNA, Keysight P5026B), operating under a standardized measurement condition with an input power level of -20 dBm and a consistent port impedance of 50 Ω. Before conducting the measurements, a Short-Open-Load-Thru (SOLT) calibration method was implemented to ensure accurate two-port calibration for the RF probes across the targeted frequency span. Ground-Signal-Ground (GSG) probes, featuring a probe gap of 150 μm, were employed to measure and extract the S-parameters of the MSSW filters. It should be noted that no additional on-wafer de-embedding corrections were applied in this study, meaning each filter was directly treated as the device under test (DUT). Further comprehensive details regarding the experimental setup and procedures are provided in Part 5 of the supplementary information. Supplementary information Supplemental material Acknowledgements We would like to thank Dr. Todd Bauer, Dr. David Abe, and Dr. Tim Hancock of the Defense Advanced Research Projects Agency (DARPA) and Dr. Michael Page of the Air Force Research Laboratory for their guidance and support of this work under the DARPA Wideband Adaptive RF Protection (WARP) program, contract FA8650-21-1-7010. The fabrication of devices was performed at the Singh Center for Nanotechnology, supported by the NSF National Nanotechnology Coordinated Infrastructure Program (No. NNCI-1542153). Author contributions S.W. and S.Y. contributed equally to this work. S.W., S.Y., and R.O. developed the device concepts. S.W. and S.Y. built the simulation models of the devices. S.W. and X.D. fabricated the YIG filters. S.W., S.Y., and C.-Y. C. analyzed the measured data under the supervision of R.O. S.W. wrote the original draft, and S.W., S.Y., and R.O. modified the manuscript. All authors have given approval to the final version of the manuscript. Data availability All data supporting the findings of this study are available within the article and the supplementary materials. Any additional requests for information can be directed to and will be handled by the corresponding author. Competing interests The authors declare no competing interests. Footnotes These authors contributed equally: Shuxian Wu, Shun Yao Supplementary information The online version contains supplementary material available at 10.1038/s41378-026-01260-x. References Associated Data Supplementary Materials Supplemental material Data Availability Statement All data supporting the findings of this study are available within the article and the supplementary materials. Any additional requests for information can be directed to and will be handled by the corresponding author.