AIN MEMS filters with extremely high bandwidth widening capability

The rapid expansion of 5G communications demands wider RF bands for higher data rates. RF bandpass filters must provide wide bandwidths while maintaining low loss and strong out-of-band suppression. Among candidate technologies, MEMS/acoustic filters are attractive due to compactness, cost, and sharp roll-off. Prior efforts have focused on three directions: using resonant modes with higher electromechanical coupling k^2, employing stronger piezoelectric materials (e.g., AlScN, LiNbO3), and hybridizing acoustic resonators with external networks to expand bandwidth. Although LiNbO3 modes such as SH0 and A1 can achieve high k^2, they suffer from power-handling and temperature-stability issues at wideband/high-frequency operation. AlN remains a preferred resonator material for its high power handling, thermal stability, and CMOS-compatible processing. This work proposes AlN MEMS filters that hybridize AlN Lamb wave resonators with surface-mounted lumped elements to realize greatly widened bandwidths compared with conventional ladder or lattice topologies. The study models and designs first- and second-order wideband filters and experimentally demonstrates a first-order AlN implementation.

• High-k^2 resonant modes: SH0 and A1 modes in LiNbO3 have demonstrated k^2 values of approximately 50% and 22%, enabling potential wideband filters; however, these modes face limitations in power handling and temperature stability for wideband/high-frequency use.
• Strong piezoelectric materials: Scandium-doped AlN (e.g., Sc0.12Al0.88N) increases the piezoelectric coefficient d33 by ~50% and reduces stiffness constant C33 by ~10%, improving k^2 by ~1.7×, but typically degrades resonator quality factor, which directly worsens filter insertion loss.
• Hybrid filters with SAW and lumped/microwave elements have been explored but often exhibit small bandwidth, poor roll-off, and low out-of-band rejection.
• AlN is widely used for resonators and sensors due to high power-handling capability, thermal stability, and mature, CMOS-compatible processing, making it a strong candidate for wideband RF MEMS filters.

• Resonator modeling: The MEMS resonator is modeled by the modified Butterworth–Van Dyke (MBVD) equivalent circuit with static capacitance C0 and motional branch Lm, Cm, Rm. Resonant frequency fr and antiresonant frequency fa, quality factor, and coupling are extracted by fitting measured admittance to the MBVD model.

• First-order wideband filter topology: A single MEMS resonator is paralleled with an inductor L0 to generate two symmetric transmission zeros (fa1, fa2) about fr, producing sharp skirts and setting the center frequency. Two identical three-element matching networks (each with series inductor Ls, shunt inductor Lp, and shunt capacitor Cp) are connected to input and output to widen the bandwidth while matching to system impedance. The matching network must satisfy (1/Lp + 1/Ls) Ls Cp = a^2 at fr; typically Ls and Lp are chosen and Cp determined by this constraint. When L0 is chosen such that ωs = ω0, with ωs = 1/√(Cm Lm) and ω0 = 1/√(C0 L0), L0 = Lm Cm / C0, leading to closed-form expressions for fa1,2, which depend on Cm/C0 (proportional to coupling). Thus, larger coupling increases the maximum achievable bandwidth via wider spacing of transmission zeros. Design insights: increasing Ls and Lp increases FBW; decreasing Lp improves out-of-band rejection; insertion loss depends on the Q of both the resonator and the lumped elements.

• Second-order wideband filter topology: Two first-order sections are cascaded in an image-symmetric configuration to simultaneously match both ports and further enhance roll-off and rejection. Each section has its own Ls, Lp, Cp network (Ls1, Lp1, Cp1 on one side; Ls2, Lp2, Cp2 on the other), with L0 and the matching element constraints as above. Four tuning inductors (Ls1, Lp1, Ls2, Lp2) provide expanded design freedom to trade off FBW and out-of-band rejection.

• Simulation and tuning studies: Parametric simulations illustrate how varying Ls and Lp controls FBW and rejection for first-order filters (example resonator with k^2 ≈ 1.7%, fr ≈ 230 MHz). Three example first-order designs (Filters A, B, C) and three second-order designs (Filters D, E, F) demonstrate the tunability and performance trade-offs. Bandwidth widening factor (BWF = FBW / resonator coupling) is used to benchmark against conventional ladder and lattice topologies.

• Experimental demonstration: An AlN SO Lamb wave first-order wideband filter was fabricated and measured, achieving FBW = 5.6% with resonator coupling of 0.94% and insertion loss (IL) = 1.84 dB. The extracted bandwidth widening factor (BWF) is 6, approximately 12× that of conventional ladder or lattice filters.

• First-order simulation examples (k^2 = 1.7% resonator):

  • Filter A: FBW 7.2%, BWF 4.2, out-of-band rejection ~22 dB.
  • Filter B: FBW 5.4%, BWF 3.2, out-of-band rejection ~29 dB.
  • Filter C: FBW 3.7%, BWF 2.2, out-of-band rejection ~39 dB.
  • Conventional references (same resonator): Ladder FBW ≈ 0.78%, BWF ≈ 0.45; Lattice FBW ≈ 0.95–0.98%, BWF ≈ 0.56.

• Second-order simulation examples (k^2 = 1.7% resonator):

  • Filter D: FBW 7.8%, BWF 4.6, rejection ~55 dB.
  • Filter E: FBW 6.7%, BWF 3.9, rejection ~65 dB.
  • Filter F: FBW 3.7%, BWF 2.2, rejection ~76 dB.
  • Second-order filters exhibit markedly improved roll-off; an example shows >60 dB out-of-band rejection and a −30 dB shape factor of ~1.05.

• Design trends: Larger Ls and Lp increase FBW; smaller Lp improves out-of-band rejection; insertion loss depends on the Q of the resonator and lumped elements. Cascading to a second-order topology significantly enhances rejection and roll-off with flexible trade-offs in bandwidth.

The proposed hybrid topology decouples filter bandwidth from the intrinsic limits of resonator k^2 more effectively than conventional ladder/lattice architectures by leveraging a parallel inductor to create transmission zeros around fr and using tailored three-element matching networks to widen passband while maintaining impedance match. As a result, filters can achieve FBWs several times the resonator's coupling (high BWF), enabling practical wideband operation using robust, thermally stable AlN Lamb wave resonators. Simulations and measurements validate that substantial bandwidth widening is achievable with low insertion loss and strong out-of-band rejection. The second-order configuration further improves roll-off and rejection (>60 dB) with modest impact on bandwidth, supporting stringent spectral masks. The demonstrated performance indicates strong potential to meet wide bandwidth requirements for 5G NR bands (n77, n78, n79) while preserving power handling and stability advantages of AlN. The approach is general and applicable to various acoustic resonator types modeled via MBVD.

This work introduces first- and second-order wideband RF MEMS filter topologies that hybridize AlN Lamb wave resonators with surface-mounted lumped-element matching networks. The method strategically generates transmission zeros and employs three-element networks to widen bandwidth while maintaining matching and sharp skirts. Simulations show BWF up to ~4.6 with high rejection (>60 dB) for second-order designs, and an experimentally realized first-order AlN filter achieves FBW 5.6% at 0.94% coupling with 1.84 dB IL and BWF 6, far surpassing ladder/lattice references. These results demonstrate an exceptional bandwidth widening capability and strong out-of-band suppression, making the approach promising for wideband 5G NR applications. Future research could include experimental realization of second-order filters, optimization of lumped-element Q to further reduce IL, extension to higher frequencies and different resonator platforms, and integration toward compact, multi-band front-end modules.

• The experimentally validated device is the first-order filter; second-order results are based on simulation only in the provided content.
• Out-of-band rejection and roll-off for the first-order design, while good, are inferior to the second-order approach.
• Insertion loss is sensitive to the quality factors of both the resonator and the discrete lumped components; practical component losses may limit performance.
• Some material approaches to boosting k^2 (e.g., Sc-doped AlN) can degrade resonator Q, potentially offsetting bandwidth gains if applied.