A TSV-packaged SP6T RF MEMS switch with ultra-thin profile and suppressed substrate loss for DC to Ku band application Authors: Yulong Zhang, Jiangtao Wei, Huiliang Liu, Jianwen Sun, Qiannan Wu, Mengwei Li, Zewen Liu Published: 2026 ABSTRACT AbstractThe evolution of wireless communication imposes stringent requirements on RF front-end modules, specifically demanding switches with ultra-thin profiles and wideband coverage up to Ku band. However, conventional packaged MEMS switches often result in excessive device height, creating a bottleneck for compact system integration. To address these challenges, this study presents a TSV-packaged Single Pole Six Throw (SP6T) RF MEMS switch that simultaneously achieves an ultra-thin profile and superior signal integrity. The device features a total height of only 300 μm, achieved through a substrate grinding process. To compensate for the impedance variations induced by the ultra-thin substrate and air cavity, two types of Heterotypic Microstrip (HMS) transmission lines are proposed and optimized for signal matching from DC to Ku band. The switches are fabricated using a combination of surface and bulk micromachining, including gold-alloy electroplating, Au-Sn bonding, and TSV processes. Measurement results verify the design strategy: the switch with a 300-μm height exhibits significantly improved insertion loss (IL) suppression compared to thicker counterparts. The best IL reaches 2.0 dB at 18 GHz, with return loss better than 14 dB and isolation exceeding 18 dB across the DC ~18 GHz band. Furthermore, mechanical and thermal simulations confirm that the substrate thinning does not compromise contact force or heat dissipation. These results demonstrate that the proposed packaging scheme successfully solves the conflict between miniaturization and high-frequency performance, offering a promising solution for next-generation mobile RF systems. Subject terms: Electrical and electronic engineering, Engineering Introduction Wireless communication plays an important role in the world, and it shows a long-term evolution from GSM to 5G/6G1–3 and satellite communication4 during the past decades. Terminals in these communication systems need to cover frequencies from L to Ku bands, which is a great challenge for RF front-end (RFFE) modules. As a core component in RFFE, switch works for signal receiving, transmitting, and antenna tuning, which needs performances of low loss, high linearity, high RF voltage tolerance, and wide band coverage. The Micro-Electro-Mechanical-System (MEMS) switch meets the requirements above well5–7, showing good potentials as a candidate for RFFE application, compared with Silicon on Insulator (SOI), wide band-gap semiconductor (GaAs, GaN), phase change materials (PCM) technologies8–11. For example, cellphones, the most common terminals in daily life, support 2G, 3G, 4G, 5G, WiFi 6E/7 protocols1,12–14. The switch in cellphone should have a single-pole-multi-throw (SPMT) topology for multiple channels, and it need to cover the frequency from 0.6 to 7.12 GHz, even higher to X and Ku band. Meanwhile, the height of the switch is constrained, particularly in modern 5G smartphones and foldable devices. Antenna tuners and switches are required to be placed at the PCB edges or within high-density system-in-package (SiP) modules to optimize radiation efficiency. Due to the stringent vertical stack-up budget, the height of these components is increasingly demanded to be thinner. Current leading-edge solutions from suppliers such as Maxscend15, Infineon16, Qorvo17 and Skyworks18 have transitioned to ultra-low profiles, typically ranging from 0.36 to 0.45 mm, to accommodate the shrinking height budget of mobile terminals. Devices exceeding these industry-standard profiles face significant challenges in next-generation compact assemblies. Studies on MEMS SPMT switch have been reported in previous literature. Lee et al.19,20 proposed a low-loss SP6T switch based on compact metal contact MEMS switch, which performed well with an isolation from 62.4 to 39.1 dB and an insertion loss of 0.19~0.70 dB from DC to 6 GHz. The switch was fabricated on a GaAs substrate of 625 μm, packaged by a chip on board (CoB) method with wire-bonding, which limited its working band. Rebeiz et al.21–23 presented a series of SPMT RF MEMS switches, based on a symmetric circular switch topology. The SP7T switch achieved isolation level of 50~19 dB, and insertion loss of 0.3~1.2 dB at 0.1~10 GHz. The switches were fabricated on a 550-μm high-resistivity silicon substrate, without package. Dey et al.24–27 reported several SPMT RF MEMS switches, where the output number varies from 3 to 14 throws. The SP6T switching network demonstrated a measured isolation of 18 dB and insertion loss of 0.7 dB up to 12 GHz. The switches were fabricated on a 635-μm alumina substrate, without package. These SPMT switches in previous studies show passable RF performances. However, they work with thick profiles, and without proper package, which make these devices not applicable in modern compact mobile terminals. Packaging is essential for MEMS devices to provide structure protection and signal connection, and it is vital for RF MEMS switch to exhibit best performance28. Researchers and engineers have developed numerous studies in the field of MEMS switch package. There are two different package methods for RF MEMS switch, conventional chip scale package, and advanced wafer level package. Early MEMS switches were always packaged by chip scale method29,30, which showed narrow frequency band, large device scale, and limited production. In contrast, the wafer level package methods bring in several advantages, including wider bands, compact size, and higher yield. Göritz et al.31 reported a wafer-level silicon-dioxide thin-film encapsulated mm-Wave RF MEMS switch in back-end-of-line (BEOL) of a 130-nm SiGe BiCMOS technology. The encapsulated switch was measured at D-band without a significant performance drop. Jiang et al.32 showed an epoxy adhesive silicon cap package for an SPST switch with a height of 0.8 mm, in which the package method can support working frequency up to 20 GHz. Zhou et al.33 proposed an 8×8 single-chip sequential rotation crossbar (SR-crossbar) RF MEMS switch matrix with a height of 0.86 mm, in which Au-In bonding, silicon cap and TSV are employed for its wafer-level package. The matrix showed an insertion loss better than 19 dB, while the single SPST switch unit showed an insertion loss better than 2 dB within 0~26.5 GHz. Menlo Microsystems34–37 introduced several packaged SPMT switches, with glass cap and through glass vias (TGVs), and the heights are about 0.9 mm. The switch achieved a de-embedded insertion loss better than 1.3 dB up to 18 GHz at normal SP4T mode. Analog Devices38 also offers RF MEMS SP4T devices, which has a 0.95-mm-height SiP form with a cap-protected MEMS part and a charge pump integrated circuit part. The SP4T device demonstrates an insertion loss of 3 dB up to 12 GHz. Most of the packaged MEMS switches have high profiles due to the two bonded wafers, showing a limited prospect in some applications with thin profile. In this paper, we develop a TSV packaged SP6T RF MEMS switch with ultra-thin profile down to 0.3 mm, which is suitable for applications in confined space. The packaged switch consists of two parts: the device wafer is ground from 400 μm to 50 μm, while the cap wafer with TSVs has a height of 250 μm. Due to the ultra-thin HR-Si substrate, loss of TML is suppressed significantly. The switch shows IL better than 2.0 dB, RL better than 14 dB, and Isolation better than 18 dB @DC ~ 18 GHz. The simulation, tested results and discussion shows that the switch have excellent performances, and great potentiality for DC to Ku band applications. The paper is organized as follows. The next section shows design of the switch in detail, and the fabrication processes are introduced step by step. Then, figures of the switches are characterized, and test results are analyzed. Meanwhile, several post-simulations are conducted for further performance discussion. Finally, we conclude the study in the last section. Device design and fabrication Device design Figure 1a illustrates the schematic of the TSV-packaged SP6T RF MEMS switch, in which RFc represents the input port, and RFx (x = 1~6) represents output port, respectively. The switch consists of two parts, device part and cap part, as shown in the exploded view in Fig. 1b. The thickness of the cap substrate is 250 μm. Although the initial device wafer thickness is 400 μm, it is ground to 50 μm in the final optimized ultra-thin configuration, resulting in a total switch height of 300 μm. These dimensions are optimized to balance performance and reliability. The 50-μm device substrate is essential for minimizing substrate loss, while the 250-μm cap provides necessary mechanical rigidity to serve as a support handle during the back-grinding process. The impact of the device substrate thickness variation (from 400 to 50 μm) on the switch performance will be discussed in detail in the following sections. Figure 1c shows the core of the switch, with a heptagon structure for the SP6T topology, in which one side works for input signal and other six sides work for output signal. The close view of a single switch is shown in Fig. 1d, with anchor, beam, and contacts. When the actuation voltage is applied into the bottom electrode, the beam is pulled down, and the contacts allow the signal passing through, resulting in an ON state for this channel; when the voltage is turned off, the beam recovers to the original position, the channel turns into an OFF state. Figure 1e, f shows the close view of the match lines, sealing ring, and TSVs in the switch. The width and length of the match lines are designed carefully for a proper matching performance, which is described as follows. Structure design and working principle of the switch. a Overview of the TSV packaged SP6T RF MEMS switch. b Exploded view of the switch, with device part and cap part. c Heptagon structure for the SP6T, protected by a cap, which is HMS-Type I region. d Close view of the single switch, with anchor, beam, and contacts. e Close view of the HMS-Type II match lines, sealing ring, and TSVs. f TSVs for RF signal, control, and GND Due to the silicon substrates and air cavity, the TMLs in the switch belong to heterotypic microstrip (HMS) family, as illustrated in Fig. 2a, b, which are different with conventional microstrip in Fig. 2c. The Type I HMS TML in Fig. 2a works with an air layer below the signal line and a silicon layer above the signal line, while the Type II HMS TML in Fig. 2b works with only a silicon layer above the signal line. The electromagnetic performances of the characteristic impedances (Z0) of the HMS lines were simulated using ANSYS Electronics Desktop 2021. In the simulation model, the high-resistivity silicon substrate was defined with a dielectric constant of 11.9 and a loss tangent of 0.004. The gold material for TMLs was assigned a conductivity of 4.1 × 107 S/m. The geometric parameters were initially tuned at the center frequency of 10 GHz. The characteristic impedance Z0 is fundamentally determined by the effective dielectric constant εeff, which is modulated by the signal line width (wsignal), substrate thickness (tdevice-sub), and air cavity height (hcavity). As summarized in the parametric analysis of Fig. 2d–f, increasing tdevice-sub or wsignal raises εeff and lowers Z0, while increasing hcavity has the opposite effect. Guided by these mechanisms, a step-impedance matching network can be implemented. To validate the design robustness, the frequency dependence of Z0 for the optimized HMS line was simulated up to Ku band. As plotted in Fig. 2g and Supplementary Fig. S1, the impedance remains stable with minimal dispersion across the bandwidth. This is attributed to the high-frequency quasi-TEM operation of the TML, where Z0 is asymptotically dominated by the distributed inductance and capacitance. As these distributed parameters are primarily fixed by the physical cross-sectional dimensions and substrate properties, the impedance remains highly stable over the wideband spectrum. This design ensures good signal transmission, which is experimentally confirmed by the measured return loss in following sections. Analysis and design of the match lines: heterotypic microstrip and conventional microstrip transmission lines. a Schematic diagram of heterotypic microstrip – Type I, with another substrate above the signal line and an air cavity below the signal line. b Schematic diagram of heterotypic microstrip – Type II, with a substrate above the signal line, but without an air cavity below the signal line. c Schematic diagram of conventional microstrip – Type III. d Simulated characteristic impedance (Z0) of the TMLs versus width of signal line, under the condition of 50-μm cavity and 50-μm device substrate. e Simulated characteristic impedance (Z0) of the TMLs versus thickness of device substrate, under the condition of 50-μm cavity and 120-μm signal line. f Simulated characteristic impedance (Z0) of the Type I TMLs versus height of air cavity, under the condition of 120-μm signal line and 50-μm device substrate. g Simulated characteristic impedance (Z0) of the Type II TMLs versus frequency, under the condition of 120-μm signal line On the basis of the analysis above, the length and width of the TMLs in the switch are designed carefully. As shown in Fig. 3a, the cap part has six beam pairs, and they belong to HMS-Type I lines; when the beam pair is turned on or turned off, the RF signal is transferred to different output path. The input and output ports are matched by step-impedance lines; and each match line consists of two cascaded sections of HMS-Type II transmission lines with different characteristic impedances. The geometric parameters of the HMS matching lines were optimized using 3D electromagnetic simulations to achieve a system-level impedance match. Notably, this design carefully considers the parasitic loading effects of the subsequent TSV vertical transitions. By slightly adjusting the characteristic impedance Z0 of the horizontal lines, the overall reflection of the integrated signal path is minimized. The parasitic parameters in the circuit are listed in Fig. 3b. Due to the symmetrical topology of the SP6T switch, the parasitic parameters for output ports RF4, RF5, and RF6 are identical to those of RF3, RF2, and RF1; thus, only the unique parameters for half of the channels are presented for conciseness. Besides these TMLs, there are also several parasitic capacitors in this circuit, including feedthrough capacitor, core-part capacitor, and off-state capacitor. The equivalent circuit can describe the switch properly. Working principle of the SP6T switch. a Equivalent circuit. b Parasitic parameters extracted from the electromagnetic simulation model Device fabrication The switch is fabricated by surface and bulk processing technology with 14 masks, as shown in Fig. 4. The whole fabrication process flow consists of three parts: (I) device wafer processes, (II) cap wafer processes, and (III) package processes, which are described in details as following. (I-a) HR-Si substrate cleaning, (I-b) Poly silicon bottom electrode and bias resistor depositing and patterning, (I-c) Silicon nitride insulation layer depositing and patterning, (I-d) Gold TML electroplating and seed layer removing, (I-e) Silicon dioxide sacrificial layer depositing and patterning, (I-f) Gold-alloy beam electroplating and seed layer removing, (I-g) Silicon dioxide sacrificial layer releasing. (II-a) HR-Si substrate cleaning, (II-b) TSV etching, passivating and side wall metallizing, (II-c) SiO2 passivating and Ni/Au pads fabricating, (II-d) Au/Sn electroplating and Cavity etching. (III-a) AuSn bonding, (III-b) Device wafer grinding, (III-c) Dicing. Fabrication process flow of the SP6T switch. (I) Device wafer processes. (II) Cap wafer processes. (III) Package processes The fabrication flow includes several featured processes, such as gold-alloy electroplating, sacrificial layer releasing, gold-tin bonding, and TSV processes. The gold-alloy beams are deposited by electroplating in a cyanide bath39,40. The gold element is provided by KAu(CN)2, and the gold concentration is about 6 g/L. The alloying element is Nickel (Ni), and its concentration is about 1.5 g/L. The temperature is 50 °C. The electroplating current density is 1.0 ampere per square decimeter (ASD). The thickness of beams is electroplated to be about 5.0 μm. The sacrificial layer releasing is conducted via a wet etching process using Buffered Oxide Etch (BOE). The etchant is a 7:1 mixture of 40% NH4F and 49% HF. The etching is performed at 25 °C for approximately 90 min to ensure the complete removal of the silicon dioxide layer underneath the beams. To prevent stiction induced by the capillary forces of liquid surface tension, the wafer is extensively rinsed with deionized (DI) water, followed by solvent exchange with isopropyl alcohol (IPA). Finally, the devices are dried using a Supercritical CO2 Critical Point Dryer (CPD) process at 31.1 °C/7.38 MPa. The wafer-level packaging is achieved via Au-Sn eutectic bonding. The thickness of sealing ring is about 3 μm. And it consists of electroplated Au and Sn layers with a weight ratio of 80:20 to form the eutectic alloy. The bonding process is conducted in a nitrogen chamber with an atmospheric pressure. The wafer pair is pressed with a contact pressure of 2.0 MPa and heated to 290 °C for 10 min to form a reliable hermetic intermetallic compound. The TSVs are formed by Deep Reactive Ion Etching (DRIE) using the Bosch process. The via diameter is 65 μm with a depth of 250 μm. After etching, a SiO2 insulation layer is formed. Subsequently, a Ti/Au seed layer is sputtered, followed by electroplating to achieve a sidewall gold metallization for vertical electrical interconnection between the cap top and the bonding pads, in which the thinnest sidewall in via is about 2 μm. After that, a SiO2 insulation layer is deposited on the top side, and a directional anisotropic etching step was performed to selectively etched away the SiO2 to open the electrical contact window, while followed by the Ni/Au plating. Then, the Au/Sn bonding pads are fabricated as annular rings on the bottom silicon surface surrounding the TSV opening, and the cavity is formed by Si etching. Figure 5 shows the images of the fabricated SP6T switch. The switch is fabricated on a 6-inch wafer, as shown in Fig. 5a. Figure 5b provides the whole view of the diced switch with ultra-thin profile. The switch has a size of 4 × 4 × 0.3 mm3, which consists of a 250-μm cap substrate and a 50-μm device substrate. The visible grid-like metal patterns on the surface in Fig. 5b are the backside ground pads located on the cap substrate. These pads are designed to serve as the common ground reference and allow for direct electrical bonding to the external PCB ground. And, Fig. 5c shows the cross-section view of the TSV, with scanning electron microscope (SEM) photos. The gold metallization exhibits good continuity and coverage along the vertical sidewalls of the hollow via, ensuring reliable electrical pathways. The X-ray photographs of the SP6T switch are shown in Fig. 5d–f. The sealing ring and TSVs in the switch can be observed clearly. Figure 5g, h illustrates the optical microscope (OM) and SEM photographs of the heptagon structure for the SP6T, respectively. Figure 5i, j shows the SEM photographs of the beam and contact, in which the beam has a flat profile with a 2-μm gap above the substrate. Additionally, further FIB cross-sectional analysis can be found in Supplementary Information Note S2, to clarify the integration mechanisms and experimentally confirm the structural integrity of the device. As illustrated in Supplementary Fig. S2, the tenting effect of the dry film photoresist successfully enabled pad fabrication on the top of wafer with hollow vias. Meanwhile, the intact sidewall metallization and the empty micro-cavity without underfill are also shown in Supplementary Fig. S2. Switches in the wafer have good uniformity, and these images prove that the switch has been fabricated successfully, as required in the design. Images of the SP6T MEMS switch. a 6-inch wafer photo. b Whole view of the diced switch. c Cross section view of the TSV with SEM photos. d–f X-ray photographs of the SP6T switch. g Optical microscope photograph of the heptagon structure for the SP6T. h–j SEM photographs of the heptagon structure, beam, and contact in the SP6T Results and discussion Device measurement The fabricated switches were characterized using a Keysight N5244A PNA-X vector network analyzer connected to a Cascade Microtech Summit 11000 probe station. The measurements were performed under an atmospheric environment at room temperature. Prior to testing, a standard Short-Open-Load-Thru (SOLT) calibration was performed using an impedance standard substrate to move the reference plane to the probe tips. Resistance of the TSV was evaluated before the SP6T switch measurement, as shown in Supplementary Note S3. The measured resistance is approximately 0.173 Ω per TSV. This value, including the contact resistance at the bonding interface, confirms that the gold sidewall is continuous and capable of reliable signal transmission. The measured S-parameters of switches with different thicknesses are shown in Fig. 6. Due to the symmetry of the SP6T structure, only half of the RF output ports (RF1, RF2, and RF3) are plotted in these measured figures. Measurement results. a Insertion losses for different channels. b Insertion losses for switches with different thicknesses. c Return losses for switches with different thicknesses. d Smith chart for switches with different thicknesses. e Isolations for different channels. f Isolations for switches with different thicknesses. g, h Insertion losses and isolations uniformities of the switches with 50-μm device substrate Insertion losses for different channels in switch with 50-μm device substrate are plotted in Fig. 6a. The IL of RFc to RF1 is better than 2.0 dB at DC to Ku bands, while the ILs of RFc to RF2 and RF3 are better than 1.7 dB at same frequencies. The ILs at 0.1 GHz are about 0.10–0.12 dB, which means that parasitic resistances in signal pathways are 1.16–1.39 Ω with an acceptable uniformity. The ILs differences at higher frequency are caused by the discontinuity in SP6T structure, in which the angles between RFc and RFx ports are 51.4°, 102.8°, and 154.3°, respectively. The acute angle results in more mismatch than the obtuse angle. ILs for switches with different thicknesses are compared in Fig. 6b. For a clear comparison, only RFc to RF3 ILs are plotted. The switch with 400-μm device substrate shows a higher loss and noticeable fluctuations at 15~20 GHz. These fluctuations are attributed to the parasitic substrate resonance modes supported by the thick 400-μm silicon bulk. In contrast, the 50-μm ultra-thin profile effectively suppresses these modes by shifting their cutoff frequencies to a higher band, resulting in a much smoother and lower-loss transmission curve. This proves the loss suppression effect of substrate grinding process. Meanwhile, the return losses for switches with different thicknesses are plotted in Fig. 6c. The switch with 50-μm device substrate have a return loss better than 14 dB. And, the return loss of switch with 400-μm device substrate is only better than 10 dB, owing to the Z0 variation for thicker substrate as discussed in Fig. 2. Figure 6d illustrates the Smith chart at ON state for data in Fig. 6b; and the switch with ultra-thin profile shows better matching performance. Isolations for different channels in switch with 50-μm device substrate are plotted in Fig. 6e. The Isolation of RFc to RF1 is better than 18 dB at DC to Ku bands, while the Isolations of RFc to RF2 and RF3 are better than 20 dB at same frequencies. The isolation differences between RF1 and RF2/RF3 are caused by the feedthrough effect; the RF1 port is closer to RFc port, resulting a higher feedthrough capacitor of 15 fF, as listed in Fig. 3b. RFc to RF3 Isolations for switches with different thicknesses are compared in Fig. 6f. There are no significant Isolation differences between these switches, because of the similar parasitic capacitors. To address the uniformity, measurement results of 8 randomly selected switches from different dies on the same wafer are shown in Fig. 6g, h. The results show consistent performance with a variation of less than ±0.2 dB in insertion loss and a variation of less than ±2 dB in isolation, demonstrating good fabrication uniformity. Discussion The thinning of the device wafer from the original 400 μm to 50 μm is a critical optimization for Ku-band performance. At high frequencies such as 18 GHz, a thick silicon substrate acts as a significant volume for energy dissipation due to dielectric losses and parasitic substrate modes. By drastically reducing the substrate thickness, the interaction volume between the high-frequency electromagnetic fields and the lossy silicon medium is minimized, which effectively suppresses the substrate-induced insertion loss. Furthermore, the 50-μm profile provides superior electromagnetic field confinement for the HMS lines. In a thinner substrate environment, the electric field lines are more tightly coupled within the intended signal path rather than spreading into the bulk substrate, thereby reducing parasitic capacitance and ensuring the stability of the characteristic impedance Z0. This structure allows the switch to maintain a low insertion loss while achieving an ultra-thin form factor. For a further insight of the switch, several post-simulations are conducted as illustrated in Fig. 7. Figure 7a, b plotted the simulated RFc to RF3 insertion loss and return loss versus thickness of the device substrate at different frequencies. At frequency higher than 10 GHz, the ILs increase with the thickness of device substrate significantly, which means that grinding process can suppress the substrate loss effectively. All of the simulated RLs are better than 15 dB. It is worth noting that the RL trend at 5 GHz differs from that at higher frequencies. At lower frequencies, the variation in RL is primarily attributed to the shift in characteristic impedance (Z0) caused by substrate thickness changes (as analyzed in Fig. 2e), rather than substrate absorption. In contrast, at frequencies higher than 10 GHz, substrate loss becomes the dominant factor, leading to a consistent trend. Figure 7c plots the contact force versus thickness of device substrate with a subplot of the device substrate displacement. The contact forces (Fcontact) pre- and post-grinding process are almost identical. The 50-μm ultra-thin substrate does not cause degradation of switch contact, which is important for the IL performance. Figure 7d illustrates the simulated temperature distribution at steady state with 1 W 10 GHz RF signal passing RF3 channel. The boundary condition of the model bottom is set to be 20 °C. The highest temperature appears at the beams in on-state pathway. Owing to the excellent thermal conductivity of silicon substrate and the ultra-thin profile, the switch shows excellent heat dissipation performance, resulting in a low temperature rise, which is less than those switches with glass substrate. Furthermore, it is beneficial for the lifetime improvement of the switch. Post-simulations of the switch. a, b Simulated RFc to RF3 insertion loss and return loss versus thickness of the device substrate at different frequencies. c Simulated contact force versus thickness of the device substrate with subplot of the device substrate displacement. d Simulated steady state temperature distribution with 1 W 10 GHz RF signal loaded The performance of the proposed SP6T RF MEMS switch is summarized in Table 1, alongside other state-of-the-art switches reported in the literature. Facilitated by the optimized substrate grinding process, this work achieves an ultra-thin profile of 0.3 mm, which provides a critical advantage for space-constrained terminals. Table 1Comparison of SPMT switch performances based on different technologiesRef.Technology/substrateSPMTPackage typeFrequency (GHz)ILRLIsolationHeight (mm)CostProcess Complexity21MEMS/SiliconSP7Tw/o package0.1 ~ 101.2 dB@10 GHz17 dB@10 GHz19 dB@10 GHz0.55LowMedium25MEMS/AluminaSP6Tw/o packageDC ~ 120.7 dB@12 GHz18 dB@12 GHz18 dB@12 GHz0.64LowMedium37MEMS/GlassSP4TTGV WLPDC ~ 180.4 dB@6 GHz15 dB@6 GHz25 dB@6 GHz0.90MediumHigh0.5 dB@10 GHz14 dB@10 GHz22 dB@10 GHz1.3 dB@18 GHz10 dB@18 GHz18 dB@18 GHz38MEMS/SiliconSP4TLF CSPDC ~ 100.4 dB@6 GHz18 dB@6 GHz23 dB@6 GHz0.95HighHigh1.0 dB@10 GHz9 dB@10 GHz12 dB@10 GHz41SOI/ SapphireSP4TLGA0.000009 ~ 181.5 dB@6 GHz14 dB@6 GHz42 dB@6 GHz0.91MediumHigh2.1 dB@10 GHz14 dB@10 GHz38 dB@10 GHz3.2 dB@18 GHz14 dB@18 GHz25 dB@18 GHz42NA/GaAsSP6TSMTDC ~ 182.5 dB@6 GHz12 dB@6 GHz47 dB@6 GHz0.90MediumMedium2.4 dB@10 GHz9 dB@10 GHz42 dB@10 GHz3.5 dB@18 GHz10 dB@18 GHz35 dB@18 GHz43pHEMT/GaAsSP10Tw/o packageDC ~ 182.0 dB@6 GHz11 dB@6 GHz40 dB@6 GHz/MediumMedium2.4 dB@10 GHz18 dB@10 GHz30 dB@10 GHz3.1 dB@18 GHz11 dB@18 GHz40 dB@18 GHzThis workMEMS/SiliconSP6TTSV WLPDC ~ 180.9 dB@6 GHz18 dB@6 GHz21 dB@6 GHz0.3MediumHigh1.1 dB@10 GHz22 dB@10 GHz19 dB@10 GHz2.0 dB@18 GHz14 dB@18 GHz18 dB@18 GHz Comparison of SPMT switch performances based on different technologies Achieving such a compact thickness is a great challenge for existing technologies. For other RF MEMS packaging methods, the mandatory protective cavity, combined with the vertical clearance required for wire-bonding or the standoff height for flip-chip bumps, severely restricts the wafer thinning. Conversely, while solid-state switches do not require cavities, aggressive wafer thinning process often induces severe mechanical rigidity and structural warpage issues. Furthermore, the presence of a packaging substrate inherently limits the further reduction of the overall device height. By integrating TSVs with a localized micro-cavity, this monolithic approach effectively circumvents these bottlenecks above. While offering higher integration density with more throw channels compared to standard SP4T counterparts, the device maintains competitive RF characteristics comparable to those of thicker alternatives. When benchmarked against mainstream commercial solid-state switches operating in the same frequency band, the proposed device demonstrates a distinct efficiency edge. For instance, while high-end commercial RF SOI and GaAs switches exhibit typical insertion losses of 3.1–3.5 dB at 18 GHz, the proposed TSV-packaged MEMS switch achieves a significantly lower loss. This 1.0 dB performance advantage, combined with the superior linearity of metal-to-metal contacts, highlights the potential of the proposed ultra-thin switch as an irreplaceable solution for high-efficiency, reconfigurable RF front-ends. Conclusion In this work, we proposed a novel SP6T MEMS switch based on HR-Si substrate and packaged by a cap with TSVs. The HMS TMLs are designed carefully for signal matching from DC to Ku band. The switches are fabricated by surface and bulk processing technologies, including gold-alloy electroplating, sacrificial layer releasing, gold-tin bonding, substrate grinding, and TSV processes. With the help of design and fabrication, the switch has an ultra-thin profile down to 0.3 mm, covering the frequency bands from DC to Ku band. The test results reveal that the switch shows insertion loss (IL) suppression performance at ON state. The IL reaches 2.0 dB@ 18 GHz, the return loss (RL) is better than 14 dB @DC ~ 18 GHz, the Isolation at OFF state is better than 18 dB @DC ~ 18 GHz. These performances pave the way for applications with higher frequency and limited space, such as emerging future wireless communication. Further, these researches in antenna tuning and RF channel switching will be conducted to promote the development of practical applications. Supplementary information Supplemental material Acknowledgements This research is supported by The National Natural Science Foundation of China No. 62504212, Foundation of Shanxi Key Laboratory of Graphene Sensing Materials and Devices No. SMX2025003, and National Key R&D Program of China No. 2018YFB2002801. Author contributions Y.Z. performed design, fabrication, and analysis; J.W. contributed to the device FEA simulation; H.L. and J.S. contributed to the analysis and paper revising; Q.W., M.L. and Z.L. discussed the manuscript and analysis. All authors contributed to the preparation of the manuscript. Data availability All data are fully available without restriction. Conflict of interest The authors declare no competing interests. Contributor Information Mengwei Li, Email: lmw@nuc.edu.cn. Zewen Liu, Email: liuzw@tsinghua.edu.cn. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41378-026-01417-8. References Associated Data Data Availability Statement All data are fully available without restriction.