Sub-femtosecond figure-of-merit millimeter-wave switches via solution-processed MoS2 for 6G radio-frequency front-ends Authors: Changwoo Pyo, Sungmoon Park, Seungchan Lee, Dohyun Lee, Juho Son, Joohoon Kang, Deji Akinwande, Myungsoo Kim Published: 2026 ABSTRACT Next-generation 6G communication requires radio-frequency components capable of operating above 100 GHz with low loss, high isolation, and zero static power—requirements that challenge complementary metal-oxide-semiconductor (CMOS) and microelectromechanical systems (MEMS) technologies. Here, we report transfer-free, large-area millimeter-wave (mmWave) switches based on solution-processed MoS2. While solution-processed 2D materials are often viewed as inferior to their crystalline counterparts due to high defect densities, we demonstrate that their edge-rich morphology is, in fact, a performance enabler. These edge defects act as intrinsic templates that confine Cu-filament pathways, enabling rapid (76 ns) and low-energy switching (1.57 nJ) with > 2000 cycles and uniform zero static-power operation, as corroborated by Kelvin probe force microscopy, conductive atomic force microscopy and low-temperature studies. The resulting switches achieve an low insertion loss ( < 0.1 dB) and high isolation ( > 35 dB) at 67 GHz. Notably, they exhibit a switching figure-of-merit (RON·COFF) of ~ 0.8 fs (fco ~ 187 THz), surpassing previously reported 2D switches. Importantly, by adopting an inverse-state operational scheme in a SHUNT architecture, we mitigate self-switching and achieve improved power handling (P0.1dB > 10 dBm) and linearity (IIP3 > 42.1 dBm). Finally, we demonstrate the platform’s circuit-level viability by integrating the switches into a true-time-delay and hybrid-coupled phase shifters targeting 30 GHz mmWave applications. Introduction The advancement of wireless communication toward 6G standards necessitates high data rates (up to 100 Gbits−1) and operation in the mmWave spectrum above 100 GHz1–3, which demands high-performance, low-power radio-frequency (RF) components4. Analog switches serve a function in signal routing and processing to reconfigure communication systems, yet conventional technologies fall short5–8. Field-effect transistors (FETs) are hindered by intrinsic parasitic capacitance that degrades high-frequency performance, while microelectromechanical systems (MEMS) suffer from high actuation voltages and large footprint issues9–11. A notable drawback for both is the requirement for static power. To address these limitations, emerging memory-based switches are gaining prominence for mmWave communication. Considering the conventional communication architecture used in cellular networks, their nonvolatile nature enables low-power operation by consuming only a few nanojoules during each Tx (UL)/Rx (DL) transition, in contrast to conventional switches that continuouslyconsume milliwatts of power. However, leading memristive candidates face significant trade-offs. Phase-change memory (PCM), despite achieving complementary metal-oxide-semiconductor (CMOS) integration, is hampered by the thermal crosstalk of its heaters and a performance ceiling of RON·COFF ~ 10 fs (fco = 1/2πRONCOFF ~ 15 THz)12–17. Oxide-based (e.g., HfOx, TiOx, VOx) and polymer-based (e.g., Nafion, pV3D3) memristors are likewise limited at high frequencies, where large parasitic capacitance arising from high dielectric constants compromises isolation, and high ON-state resistance penalizes insertion loss5,18–23. In contrast, switches based on two-dimensional (2D) materials like hBN and MoS2 have demonstrated figures of merit (FoM), achieving RON·COFF ~ 2 fs (fco ~ 120 THz) in small unit devices3,24–27. Although many of these studies have employed mechanically exfoliated or CVD-grown films, solution processing offers a distinct materials route for memristive RF switches. Beyond its compatibility with low-cost and wafer-scale deposition, the flake-assembled microstructure of solution-processed films naturally provides abundant edge-rich defect sites, which can promote ion migration and conductive filament formation28. Nevertheless, challenges remain for this emerging platform, including an incomplete understanding of configuration-dependent performance, scaling implications, and reliability bottlenecks such as limited endurance and uniformity, as well as practical integration hurdles for RF front-ends. A key bottleneck for 2D switches remains their limited cycling endurance, typically capped at a few hundred cycles29. This work overcomes these challenges by developing a low-power and sub-femtosecond mmWave switch platform based on a transfer-free, large-area Au/solution-processed MoS2/Cu device. We demonstrate that the vacancy-rich edge sites inherent to solution-processed MoS2, where sulfur vacancies (VS) guide Cu ions, serve as preferential pathways for Cu ion migration, effectively engineering the filament formation energy30. We identified this mechanism as electrochemical metallization (ECM) via comprehensive analysis using conductive atomic force microscopy (cAFM), Kelvin probe force microscopy (KPFM), and low-temperature I–V analysis. This mechanism facilitates rapid and low-power operation, confirmed via real-time pulse measurements to be approximately 76 ns and 1.57 nJ per switching event with 2000 cycles of endurance, while improving switching uniformity by 77% compared to the exfoliated sample26,29,31. We demonstrate the platform’s versatility across series, shunt, and series-shunt (SS) configurations. We further demonstrate that implementing an inverse-state operational scheme in a shunt topology for memristive switches resolves the self-switching reliability bottleneck3,24,25. By mapping the memristor’s low resistance state to the mmWave switch’s OFF-state, we render the device immune to accidental triggering by high-power RF signals, a known reliability issue in conventional memristive switches. The shunt configuration exhibits a low insertion loss (<0.1 dB), high isolation (>35 dB) at 67 GHz, a P0.1dB > 10 dBm, high linearity (IIP3 > 42.1 dBm), and an extrapolated RON·COFF ~ 0.8 fs (fco ~ 187 THz). This sub-femtosecond performance represents an advancement over previously reported 2D material-based RF switches25. Furthermore, the series-shunt architecture surpasses the scaling limits of the series configuration, enabling higher isolation. The practical viability of this platform is demonstrated by integrating these switches into a functional single-pole double-throw (SPDT), 2-bit true time delay (TTD) phase shifter operating over a 16 GHz bandwidth with low error (RMS insertion loss error ≤ 0.3 dB, RMS group delay error ≤ 4 ps), and a 2-bit mmWave hybrid-coupled phase shifter (HCPS) targeting 30 GHz, confirming their suitability for essential components in next-generation communication systems. Results Device structure and DC characteristics Figure 1a presents a schematic diagram of the solution-processed MoS2 mmWave switch. The device’s active layer was formed by spin-coating a solution-processed MoS2 ink, which was synthesized via an electrochemical intercalation method using a tetraethylammonium bromide (THAB) (Supplementary Figs. 1 and 2). The inset of Fig. 1a shows the scanning electron microscope (SEM) image of the MoS2 film’s surface, confirming full coverage on a 1 × 1 cm2 sapphire substrate. The uniformity of the photoluminescence (Fig. 1b) and Raman spectra (Fig. 1c), acquired from eight arbitrary points, confirms high spatial homogeneity. Furthermore, AFM topology (Supplementary Fig. 3) verifies the formation of a uniform film with a thickness of ~30 nm. The device’s active area is defined by the overlapping region of the top electrode (TE) and bottom electrode (BE) (Supplementary Fig. 4). Ellipsometry analysis, fitted to the Lorentz oscillator model, yielded a dielectric constant (εr) of 2.13, comparable to that of CVD-grown MoS2 (Supplementary Note 1, Supplementary Fig. 5)32.Fig. 1Device structure, material characterization, and DC switching performance of solution-processed MoS2 switches.a Schematic illustration of Au/solution-processed MoS2/Cu millimeter-wave (mmWave) switch fabricated on impedance-matched coplanar waveguide (CPW). The inset shows a scanning electron microscopy (SEM) image of the solution-processed MoS2 film morphology in the device overlap area. The scale bar is 5 μm. b Photoluminescence and c Raman spectra of solution-processed MoS2 nanosheets, measured at eight arbitrary points on a 1 × 1 cm2 sapphire substrate prepared under conditions identical to the mmWave switch. d Non-volatile bipolar resistive switching I–V curves obtained during the first 50 voltage-swept switching cycles of a 4 × 4 μm2 device, measured with a 1 mA compliance current (CC). e Retention and f endurance characteristics of the 4 × 4 μm2 device, demonstrating stable operation over 106 s and 2000 switching cycles, respectively, with an ON/OFF ratio of ~103. The endurance was measured using DC voltage sweeps with a 5 mA CC. Corresponding g cumulative distribution functions (CDFs) for retention, and h endurance with 0.88% and 28.6% of coefficient of variation (CV) for low resistance state (LRS), respectively. Real-time pulse switching performance under a 1 GHz, 0 dBm continuous-wave (CW) input, triggered by a 100 ns, 3 V programming pulse for i SET and j RESET processes. Here, tON and tOFF denote the RF power transition times extracted from the 10–90% rise and 90–10% fall of the RF output power during the SET and RESET processes, respectively. tpulse denotes the applied programming pulse width, and the grey shaded areas indicate the transition-time extraction windows. Corresponding voltage and current across the device and extracted power consumption and energy during transition for k SET and l RESET processes. Device structure, material characterization, and DC switching performance of solution-processed MoS2 switches. a Schematic illustration of Au/solution-processed MoS2/Cu millimeter-wave (mmWave) switch fabricated on impedance-matched coplanar waveguide (CPW). The inset shows a scanning electron microscopy (SEM) image of the solution-processed MoS2 film morphology in the device overlap area. The scale bar is 5 μm. b Photoluminescence and c Raman spectra of solution-processed MoS2 nanosheets, measured at eight arbitrary points on a 1 × 1 cm2 sapphire substrate prepared under conditions identical to the mmWave switch. d Non-volatile bipolar resistive switching I–V curves obtained during the first 50 voltage-swept switching cycles of a 4 × 4 μm2 device, measured with a 1 mA compliance current (CC). e Retention and f endurance characteristics of the 4 × 4 μm2 device, demonstrating stable operation over 106 s and 2000 switching cycles, respectively, with an ON/OFF ratio of ~103. The endurance was measured using DC voltage sweeps with a 5 mA CC. Corresponding g cumulative distribution functions (CDFs) for retention, and h endurance with 0.88% and 28.6% of coefficient of variation (CV) for low resistance state (LRS), respectively. Real-time pulse switching performance under a 1 GHz, 0 dBm continuous-wave (CW) input, triggered by a 100 ns, 3 V programming pulse for i SET and j RESET processes. Here, tON and tOFF denote the RF power transition times extracted from the 10–90% rise and 90–10% fall of the RF output power during the SET and RESET processes, respectively. tpulse denotes the applied programming pulse width, and the grey shaded areas indicate the transition-time extraction windows. Corresponding voltage and current across the device and extracted power consumption and energy during transition for k SET and l RESET processes. Figure 1d displays I–V curves, clearly demonstrating the device’s non-volatile resistive switching (NVRS) characteristics. Both the RESET (low-resistance state (LRS) to high-resistance state (HRS)) and SET processes (HRS to LRS) occur within a low operating voltage window of −1 to 2 V. The device exhibits a stable retention time exceeding 106 s at room temperature (Fig. 1e). Importantly, both the LRS and HRS also remained stable at 85 °C over 12 h without noticeable degradation, confirming the thermal robustness of the programmed states under representative communication-chip operating conditions (Supplementary Fig. 6). The device also shows an endurance of 2000 cycles (Fig. 1f), a improvement over previously reported MoS2-based RF switches24,29. To minimize stress, endurance measurements were conducted without back-tracing, an improvement over the standard double-sweep method (Supplementary Fig. 7)26. The stability of the resistance states is further evidenced by the cumulative distribution functions (CDFs) for the LRS and HRS (Fig. 1g and h). The low coefficients of variation (CV = σ/μ, where σ is the standard deviation and μ is the average) for both retention (0.88%) and endurance (28.6%) highlight the good uniformity of the LRS. This uniformity is particularly encouraging, as the LRS in memristive RF switches directly dictates the insertion loss (IL)26. Also, RLRS could be enhanced via compliance current (CC) without specialized techniques such as pulsed-voltage stress (Supplementary Fig. 8)26,33. With zero static power consumption owing to NVRS, the energy per switching event is the key performance metric. To be viable for modern communication systems, this non-volatility must be paired with high speed and low energy consumption (Supplementary Note 2). To quantify these real-time switching characteristics, a 1 GHz, 0 dBm continuous wave (CW) signal was applied to the device simultaneously with a voltage pulse (Supplementary Fig. 9). The device achieved rapid switching, with a SET time of 64 ns (Fig. 1i) and a RESET time of 76 ns (Fig. 1j), using ~3 V pulses, limited by the bandwidth of the measurement instrument. The measured switching time is sufficient for 6G-relevant RF applications, which generally require sub-µs-level switching34,35. Prior reports of faster nanosecond- and sub-nanosecond-scale switching in 2D-material devices further suggest room for improvement beyond the present instrument-limited values3,36. By analyzing the simultaneous I–V data (Supplementary Fig. 10), the switching energy was calculated to be 2.38 nJ for the SET (Fig. 1k) and 1.57 nJ for the RESET (Fig. 1l). The switch’s isolation (IS) performance is strongly correlated with OFF capacitance (COFF), which is dominated by the metal–insulator–metal (MIM) structure. The device’s overlap area is therefore a key design parameter, making it crucial to verify that the NVRS characteristics are preserved at smaller device dimensions24,37. To investigate this scalability, point switching was performed using cAFM with a tip radius of 32.20 nm (Supplementary Fig. 11). This demonstrates that the switching phenomenon can be confined to an area as small as ~0.03 μm2. The higher SET and RESET voltages observed during cAFM measurements (Supplementary Fig. 12) are attributed to an increased voltage drop from the higher contact resistance of the nanoscale contact area. The cAFM line confirms this localized NVRS characteristic (Supplementary Fig. 13). Additional cycling data and individual point switching graphs are presented in Supplementary Figs. 14–16. Conduction and switching mechanisms To clarify the charge-transport and resistive-switching physics of the solution-processed MoS2 mmWave switch, we measured temperature-dependent electrical characteristics. In the LRS, the I–V response is well described by Ohmic conduction, showing a near-unity dependence (I2–V2 slope ~1) across 100–300 K (Fig. 2a). The resistance decreases with increasing temperature, consistent with a Cu-dominated metallic channel similar to that reported for Au/MoS2/Cu systems30 which can be expressed as Eq. (1):1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho (T)={\rho }_{0}+{\rho }_{{{{\rm{ph}}}}}(T)$$\end{document}ρ(T)=ρ0+ρph(T)where ρ0 is the residual resistivity, which comprises all structural resistivity contributions like defect or surface scattering, and ρph is the resistivity contribution due to electron–phonon scattering. The temperature coefficient of resistance (TCR, α) is defined as Eq. (2):2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R\left(T\right)=\,{R}_{0}\left(1+\alpha \left(T-{T}_{0}\right)\right)$$\end{document}RT=R01+αT−T0where R0 is the resistance at a reference temperature T0, and α is TCR. The extracted α ~ 0.002 K−1 is smaller than the value expected for bulk Cu, suggesting that an additional transport contribution is likely present in parallel with the Cu-rich filamentary path. By contrast, the HRS (Fig. 2b) exhibits thermally activated conduction that is well described by the Schottky emission model (ln(I)–V1/2, R2 ~ 0.99) indicating that current flow is primarily limited by the interfacial Schottky barrier at the electrode/MoS2 junction38 as Eq. (3):3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {J}={A}^{\ast }{T}^{2} \exp \left(-\frac{q\left(\right.{\phi }_{{{{\rm{B}}}}}-\sqrt{\frac{{qE}}{4\pi {\varepsilon }_{r}{\varepsilon }_{0}}}}{{kT}}\right)\\ {{A}^{\ast }=\frac{120{m}^{\ast }}{{m}_{0}}}$$\end{document}J=A*T2exp−qϕB−qE4πεrε0kTA*=120m*m0where J is the current density, A* is the effective Richardson constant, m0 is the free electron mass, m* is the effective mass, q is the electron charge, T is the absolute temperature, E is the electric field across the dielectric, k is Boltzmann’s constant, ε0 is the permittivity of free space, and εr is the relative dielectric constant. This picture is further supported by the RESET behavior, where the current collapses abruptly in a single step, which is a hallmark of filament rupture driven by highly localized Joule heating39.Fig. 2Analysis of conductive filament formation and carrier transport in solution-processed MoS2 switches.I–V characteristics of the solution-processed MoS2 switch measured from 100 to 300 K in 25 K intervals for a the LRS and b the high resistance state (HRS), with inset graphs showing α fitting at LRS and Schottky fitting at HRS with extracted R2 results, respectively. RT and R0 denote the resistance at temperature T and at the reference temperature, respectively. c High-resolution conductive atomic force microscopy (cAFM) current map of a localized filament region acquired after switching with a 10 mA compliance current and removing the top electrode. d Spatial map of R2 values derived from Ohmic fitting performed at each pixel (400 independent points measured from 0 to 0.5 V on a 20 × 20 grid). Vread denotes the voltage applied to the cAFM tip during image acquisition. e Simultaneous 3D current and work function maps of the switch overlap area after 10 mA switching and top electrode (TE) removal. f 3D topology and surface potential maps of a pristine solution-processed MoS2 flake. g SET and RESET switching voltage distributions for devices based on exfoliated and solution-processed MoS2. h LRS and HRS resistance distributions for devices based on exfoliated and solution-processed MoS2. Both device types utilize the identical Cu/MoS2/Au structure and fabrication process. μ and σ denote the average value and standard deviation, respectively. Both cyclic endurance measurements were performed with a voltage sweep without backtracking. Analysis of conductive filament formation and carrier transport in solution-processed MoS2 switches. I–V characteristics of the solution-processed MoS2 switch measured from 100 to 300 K in 25 K intervals for a the LRS and b the high resistance state (HRS), with inset graphs showing α fitting at LRS and Schottky fitting at HRS with extracted R2 results, respectively. RT and R0 denote the resistance at temperature T and at the reference temperature, respectively. c High-resolution conductive atomic force microscopy (cAFM) current map of a localized filament region acquired after switching with a 10 mA compliance current and removing the top electrode. d Spatial map of R2 values derived from Ohmic fitting performed at each pixel (400 independent points measured from 0 to 0.5 V on a 20 × 20 grid). Vread denotes the voltage applied to the cAFM tip during image acquisition. e Simultaneous 3D current and work function maps of the switch overlap area after 10 mA switching and top electrode (TE) removal. f 3D topology and surface potential maps of a pristine solution-processed MoS2 flake. g SET and RESET switching voltage distributions for devices based on exfoliated and solution-processed MoS2. h LRS and HRS resistance distributions for devices based on exfoliated and solution-processed MoS2. Both device types utilize the identical Cu/MoS2/Au structure and fabrication process. μ and σ denote the average value and standard deviation, respectively. Both cyclic endurance measurements were performed with a voltage sweep without backtracking. Spatially resolved evidence for filament formation was obtained by conductive AFM (cAFM) following removal of the TE, where discrete, highly conductive spots appear within the active region. Successful TE removal with minimal damage to the switching layer was confirmed by AFM topography, which showed negligible difference between the as-deposited and TE-removed surfaces (Supplementary Fig. 17). Increasing the CC from 5 to 40 mA systematically enhances the conductivity of these spots, consistent with progressive filament reinforcement (Supplementary Fig. 18). Such CC-dependent evolution can be rationalized by positive feedback in ECM-type growth: once a conductive nucleus forms, the local resistance drops, current crowds into that region, and Cu incorporation is further accelerated. The proportional scaling between the peak current and the applied CC (Supplementary Fig. 19) indicates that filament development is reproducible and controllable, producing an effective ON-state resistance that decreases approximately inversely with CC. The filament geometries were further resolved using high-resolution cAFM (Fig. 2c) together with pixel-by-pixel Ohmic fitting maps (R2) (Fig. 2d) over a 140 × 140 nm2 area. The R2 value approaches ~1 at the filament center, indicative of nearly ideal metallic conduction, but gradually decreases toward the periphery, implying a transition to a mixed transport regime where Schottky-like components become increasingly influential. This spatial trend is consistent with a conical filament profile and agrees with the elevated Schottky-fitting R2 observed near the filament boundary (Supplementary Fig. 20). Complementary confirmation of both composition and three-dimensional structure was obtained by simultaneous cAFM/KPFM at 10 mA CC (Fig. 2e), where a pronounced work–function contrast emerges: the periphery remains MoS2-like (~4.36 eV), whereas the conductive core shifts toward a Cu-like value (~4.53 eV)40. This localized work-function gradient provides strong evidence that the conductive pathway is Cu-rich and adopts a conical growth morphology. Importantly, the simultaneously collected topography shows no discernible surface protrusion, indicating that the filament forms within the MoS2 layer rather than as a raised feature on the surface. (Supplementary Fig. 21). To understand the physical origin of the suppressed α at LRS, we analyzed pristine, isolated MoS2 flakes by KPFM (Fig. 2h). The edge regions exhibit higher surface potential (lower work function) compared to the basal plane, consistent with a higher density of VS that act as electron donors41. Based on this observation, we suggest that vacancy-rich flake edges provide energetically favorable channels for Cu ion migration42, thereby promoting more localized and efficient filament nucleation/growth and ultimately reducing the required switching voltages. This mechanism-level advantage is directly reflected in the switching statistics when comparing exfoliated and solution-processed MoS2 devices with the same Cu/MoS2/Au stack (Supplementary Fig. 22). As Fig. 2g and h shows the solution-processed devices switch at lower average voltage, with RESET decreasing from 1.04 V (exfoliated) to 0.71 V (solution-processed) and SET decreasing from 1.79 to 1.36 V. They also exhibit improved LRS resistance uniformity with the distribution narrowing from 128% to 28.6% when moving from exfoliated to solution-processed devices. This trend remained consistent even in comparison with CVD-grown 2D switches (Supplementary Fig. 23). Moreover, unlike the exfoliated counterpart, the solution-processed switch shows a clearly defined switching window with robust LRS/HRS separation, enabling more deterministic state assignment. Consistent with this enhanced stability, endurance improves from ~100 cycles (exfoliated) to ~2000 cycles (solution-processed). We emphasize that this reduction in LRS/HRS dispersion is not merely a device-level metric but a key enabler of RF reliability: tighter RON and COFF distributions suppress C2C drift in IL and IS, thereby stabilizing mmWave front-end performance over cycling. mmWave performance To characterize the mmWave performance, we evaluated the scattering parameters (S-parameters) from 10 MHz to 67 GHz using a vector network analyzer (VNA) with standard short-open-load-thru (SOLT) calibration and de-embedding procedure (Supplementary Figs. 24 and 25). We fabricated series, shunt, and SS switch architectures on characteristic impedance-matched coplanar waveguide (CPW)(Supplementary Fig. 26). As shown in Fig. 3a, all configurations demonstrate ON-state IL < 0.5 dB up to 67 GHz, surpassing leading technologies including PCM (−2 dB at 60 GHz)15, RF silicon-on-insulator (RF-SOI) metal–oxide–semiconductor field-effect transistors (−1.3 dB at 50 GHz)43 and a commercial switch (−3.6 dB at 67 GHz)44. The shunt achieved a low IL < 0.1 dB at 67 GHz, exceeding even hBN-based switches (−0.4 dB at 67 GHz)26. An equivalent lumped-element model (Supplementary Fig. 27) yields an extracted RON of ~1.17 Ω.Fig. 3Configuration-dependent mmWave radio-frequency (RF) performance of solution-processed MoS2 switches with flexible application.RF characteristics of the 4 × 4 μm2 Au/solution-processed MoS2/Cu mmWave switch in its a ON- and b OFF-states, characterized up to 67 GHz. The shunt configuration exhibits an ON-state resistance (RON) of 1.17 Ω and an OFF-state capacitance (COFF) of 0.72 fF, yielding a calculated cutoff frequency (fco) of 187.7 THz (RON .COFF ~ 0.8 fs). The SS structure achieves a high isolation of 40 dB at 67 GHz, while the shunt switch maintains a low insertion loss of <0.1 dB at 67 GHz, a figure competitive with state-of-the-art devices. Power handling capabilities in the c ON-state and d OFF-state. No 0.1 dB gain compression point (P0.1dB) or self-switching was observed up to the instrumental limit of 10 dBm for shunt. The red and blue indicate the output power for each state (right axis), and the gray indicates the normalized S21 (left axis). Device characteristics as a function of compliance current for e, f series and g, h shunt architectures, demonstrating tunable resistance and capacitance suitable for variable attenuator and varactor applications. The dashed lines represent fitting curves. i Intermodulation distortion measurement in the ON-state to evaluate device linearity, using a two-tone signal (f1 = 2.365 GHz, f2 = 2.415 GHz). The dashed lines represent linear extrapolations of the measured fundamental and third-order intermodulation output powers. The results reveal a high input third-order intercept point (IIP3) of 42.1 dBm and an output IP3 (OIP3) of 40.6 dBm. j S-parameter stability of the shunt device under static bending with radii ranging from 40 mm down to 5 mm. k Scaling of OFF-state capacitance for series and series-shunt (SS) devices, showing that the SS architecture improves the scaling limit by approximately an order of magnitude. Configuration-dependent mmWave radio-frequency (RF) performance of solution-processed MoS2 switches with flexible application. RF characteristics of the 4 × 4 μm2 Au/solution-processed MoS2/Cu mmWave switch in its a ON- and b OFF-states, characterized up to 67 GHz. The shunt configuration exhibits an ON-state resistance (RON) of 1.17 Ω and an OFF-state capacitance (COFF) of 0.72 fF, yielding a calculated cutoff frequency (fco) of 187.7 THz (RON .COFF ~ 0.8 fs). The SS structure achieves a high isolation of 40 dB at 67 GHz, while the shunt switch maintains a low insertion loss of <0.1 dB at 67 GHz, a figure competitive with state-of-the-art devices. Power handling capabilities in the c ON-state and d OFF-state. No 0.1 dB gain compression point (P0.1dB) or self-switching was observed up to the instrumental limit of 10 dBm for shunt. The red and blue indicate the output power for each state (right axis), and the gray indicates the normalized S21 (left axis). Device characteristics as a function of compliance current for e, f series and g, h shunt architectures, demonstrating tunable resistance and capacitance suitable for variable attenuator and varactor applications. The dashed lines represent fitting curves. i Intermodulation distortion measurement in the ON-state to evaluate device linearity, using a two-tone signal (f1 = 2.365 GHz, f2 = 2.415 GHz). The dashed lines represent linear extrapolations of the measured fundamental and third-order intermodulation output powers. The results reveal a high input third-order intercept point (IIP3) of 42.1 dBm and an output IP3 (OIP3) of 40.6 dBm. j S-parameter stability of the shunt device under static bending with radii ranging from 40 mm down to 5 mm. k Scaling of OFF-state capacitance for series and series-shunt (SS) devices, showing that the SS architecture improves the scaling limit by approximately an order of magnitude. Regarding IS, all devices achieved >20 dB up to 67 GHz, comparable to the benchmarks (Fig. 3b), suggesting that even a standalone series switch can approach the isolation levels typically targeted by CMOS SS cells, thereby enabling a more compact implementation with reduced footprint13. The shunt, leveraging a low RLRS via CC control, demonstrated even higher IS > 35 dB at 67 GHz. The lumped circuit model reveals a low COFF of 0.72 fF, making the signal leakage to ground virtually negligible and thereby alleviating a key practical drawback of the shunt configuration. This yields an FoM (RON·COFF) of ~0.8 fs. This sub-femtosecond metric corresponds to an fco of 187 THz, indicating that the intrinsic material response will not be the bottleneck even as carrier frequencies push deep into the terahertz regime required for 6G communications37. This high fco behavior was consistently reproduced in repeated measurements of the same shunt device (Supplementary Fig. 28). The extracted εr of 2.15 also aligns well with ellipsometry results. Furthermore, the SS achieved >35 dB of IS at 67 GHz, making it suitable for demanding applications such as phased-array antennas, low-power channel selectors, and reconfigurable circuits45. While the IL is inherently higher (0.3 dB at 67 GHz) due to its low-pass topology, this SS still vastly outperforms conventional CMOS equivalents. The measurement data before and after the de-embedding process are presented in Supplementary Fig. 29. The matching performance, characterized by the return loss (S11), showed favorable results up to 67 GHz (Supplementary Fig. 30). Device-to-device (D2D) variability was assessed for the series and shunt configurations (Supplementary Note 3, Supplementary Fig. 31), showing uniformity comparable to commercial PCM technology15,46,47. The devices demonstrated highly uniform fco with low CV of 5.2% and 4.8% for series and shunt, respectively26. A notable feature is that CC-tunable RLRS allows the device to function as a variable attenuator in the series (Fig. 3e and f) and as a varactor-like element in the shunt (Fig. 3g and h). Power handling capabilities were investigated for each configuration. The shunt, which routes the ON-state signal along the primary metal line, exhibited a 0.1 dB gain compression point (P0.1dB) > 10 dBm (the instrument limit) (Fig. 3c). This demonstrates linearity compared to the series (10 dBm) and series-shunt (6 dBm). In the OFF-state, the shunt avoided self-switching up to 10 dBm (Fig. 3d), unlike the series, which self-switched at ~8 dBm3,5,24,25. This robustness is attributed to the asymmetry of the switching voltages combined with an inverse-state mapping between the memristor (LRS/HRS) and the shunt switch (OFF-/ON-state), which mitigates self-switching under high RF power. As a separate stability test, the programmed state showed negligible change in the measured S-parameters under continuous-wave RF stress at 10 GHz and 0 dBm for ~6 h, confirming robust RF-state retention during prolonged signal loading (Supplementary Fig. 32). Nevertheless, because mmWave AC signals inherently include periodic negative half-cycles, the possible influence of long-term reverse-bias stress on device lifetime remains an important subject for future investigation. Linearity was further assessed via two-tone intermodulation distortion (IMD) measurements (f1 = 2.365 GHz and f2 = 2.415 GHz). The shunt achieved an input third-order intercept point (IIP3) of 42.1 dBm, corresponding to an ~22.4% improvement over the series (Fig. 3i). At an input power of 0.325 dBm, the shunt suppressed 3rd IMD (IM3) products to −84.6 dBc, better than the −75.7 dBc of the series (Supplementary Fig. 33). This linearity is a direct result of the shunt design, where the signal bypasses the nonlinear MoS2 channel. Ref. 3 these findings were consistent with COMSOL simulations, which visually confirm the power and electric field distribution (Supplementary Figs. 34, 41 and 42). Scaling characteristics of the OFF-state were also explored via COMSOL, projecting performance for devices scaled down to 0.03 μm2 (the minimum switchable active area identified by cAFM) (Fig. 3k). Simulations predict that performance converges due to CPW geometry limitations, a trend matching experimental data. While this issue has historically limited the series scaling37, our work shows the SS architecture extends this scaling limit by an order of magnitude, enabling further isolation improvements. Mechanical reliability for flexible applications was confirmed via cyclic and static bending tests. Under 103 bending cycles (30 mm radius), all configurations maintained stable ON- and OFF-state performance, with low endurance CV of 1.14% and 0.9%, respectively (Supplementary Fig. 35). Notably, observed failures during prolonged testing are attributed to the mechanical limits of the CPW metal lines, not degradation of the MoS2 switching element itself, confirming the intrinsic robustness of the active material (Supplementary Note 4, Supplementary Fig. 36). This interpretation is further supported by AFM topography of solution-processed MoS2 films on the same PI substrate before and after identical cyclic bending, which showed no noticeable cracking and only minor changes in surface roughness (Supplementary Fig. 37). Subsequently, the shunt device exhibited stable characteristics when subjected to a progressively decreasing static bending radius from 40 down to 5 mm (Fig. 3j). Memristive circuit for RF application To validate that these atomistic advantages translate to system-level performance, we integrated the switches into reconfigurable RF circuits. Leveraging the nonvolatile, zero-static-power operation of the switch, these circuits reduce standby power relative to actively biased diode- or transistor-based reconfigurable RF circuits. Separately, the low RON·COFF and low insertion loss of the present platform compare favorably with PCM-based implementations in dense arrays. First, we implemented a single-pole double-throw (SPDT) circuit, a fundamental topology for reconfigurable systems using both series and SS configurations (Supplementary Fig. 38)48. This circuit is commonly used to route signals between a Tx path (power amplifier) and an Rx path (low-noise amplifier) for a single antenna. Both configurations effectively routed and isolated the Tx and Rx path signals, which were consistent with electromagnetic (EM) simulations (Supplementary Fig. 43). However, dynamic metrics relevant to practical Tx/Rx operation, including path-switching settling time and channel-to-channel transient crosstalk, were not evaluated in the present setup and will be systematically investigated in future work. Notably, the series-only switches already provided sufficient IS, replacing the conventional SS topology—typically adopted to maximize IS—with the series-only configuration improves IL, increases integration density, and reduces switching energy (Supplementary Note 5, Supplementary Fig. 39). Next, we integrated the switches into phase shifter circuits: a 2-bit true-time-delay (TTD) phase shifter and a 2-bit hybrid-coupled phase shifter (HCPS). The TTD architecture is important for applications such as beamforming antennas and reconfigurable intelligent surfaces (RIS), as it prevents beam squint across wide bandwidths by ensuring all frequency components propagate coherently13,45,49. The zero static power consumption of our switch is a key enabler for such large-scale, energy-efficient time-space-coded RIS. Simultaneously, the HCPS integration demonstrates the switch’s value for the low-power edge devices, particularly Internet of Things (IoT) devices, where its efficiency is crucial for extending battery life. The TTD operates by routing a signal through one of several transmission paths with different physical lengths. This path length difference introduces a time delay, generating a corresponding phase shift (∆ϕ) governed by Eq. (4):4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \phi=\beta ({l}_{2}-{l}_{1})=\frac{2\pi f}{{V}_{{{{\rm{p}}}}}}({l}_{2}-{l}_{1})$$\end{document}Δϕ=β(l2−l1)=2πfVp(l2−l1)where β is the phase propagation constant, VP is the phase velocity, and l1 and l2 are the lengths of the reference and delay path, respectively. Figure 4a shows a schematic diagram of the TTD phase shifter that integrates our series and shunt switches into SP4T modules that route the RF signal through a reference line or one of three cascaded delay lines, engineered to produce phase shifts of 0° (reference, λ), 90° (λ + λ/4), 180° (λ + λ/2), and 270° (λ + 3λ/4) at 10 GHz. The optical image of the fabricated device is provided in Supplementary Fig. 40, and the overall area is 19.54 mm2. The measured results (Fig. 4c) confirm that the TTD achieves four distinct, linear-phase shift states relative to the reference, highlighting linear-phase shift with 90° per state at 10 GHz. These findings are well-supported by EM simulations (Supplementary Fig. 44). The measured maximum IL is 5.5 dB (series) and 4.7 dB (shunt), while the architecture supports phase tuning up to 270° (Fig. 4d). Using a widely adopted phase-shifter figure-of-merit (FoM), defined as Eq. (5):5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathrm{FoM}}}}_{{{\phi }}/{{\mathrm{IL}}}}=\frac{\triangle {{{\phi }}}_{{total}}}{{{{\mathrm{IL}}}}_{{{\mathrm{Max}}}}}[{\scriptstyle{\circ}\atop}/{{\mathrm{dB}}}]$$\end{document}FoMϕ/IL=△ϕtotalILMax[∘/dB]where the Δφtotal is total phase change and ILMax is maximum IL, our design achieves FoM ~ 49.1°/dB (series) and ~ 57.4°/dB (shunt) at 10 GHz. Figure 4e shows that the TTD phase shifter provides discrete delay states spaced by 25 ps across an operating band (2–18 GHz), maintaining nearly constant state-to-state phase steps and thereby enabling broadband beam steering with minimal frequency-dependent distortion. For phased arrays such as RIS, the device exhibits low state-dependent errors, with an RMS insertion-loss error ≤ 0.3 dB across all states and an RMS group-delay error ≤ 4 ps (shunt). These metrics directly map onto array-level fidelity: insertion-loss variation manifests as amplitude imbalance that reduces phased-array gain and introduces pattern ripple, while group-delay error contributes to timing/phase misalignment that can translate into beam distortion in wideband beamforming systems.Fig. 4RF circuit implementations of solution-processed MoS2 switches with typical performance.Phase shifters fabricated using solution-processed MoS2 switches, demonstrating two distinct design topologies with a CPW structure. a Circuit schematic of a 2-bit true-time delay (TTD) phase shifter on a sapphire substrate. This device employs single-pole 4-throw (SP4T) switches to selectively route the RF signal through a reference path or one of the delay lines, providing phase shifts in 90° increments up to 270° by utilizing delay lines with progressively increasing electrical lengths (from λ to a λ + 3/4λ). b Circuit schematic of a 2-bit, 90° hybrid coupled phase shifter (HCPS) on sapphire substrate. Z0 represents the characteristic impedance of the transmission line, while ZSn represents the equivalent impedance of the corresponding switch elements. This reflective-type design utilizes a 90° hybrid coupler terminated by identical reflective loads, each composed of a λ/8 transmission line stub and shunt-configured MoS2 switch to control the reflection phase. c Relative phase shift of the TTD phase shifter targeting 10 GHz, showing four well-defined states separated by stable 90° increments. d Corresponding S21 insertion loss of each state; the inset shows the root mean square (RMS) insertion loss error in dB. e Relative group delay variation of each state showing root mean square (RMS) group delay error ≤4 ps and a ~25 ps increment per state, confirming TTD functionality. The inset shows the RMS group delay error in ps. f Performance of the HCPS targeting 30 GHz, detailing four distinct operational phase states with ~90° steps. g Insertion loss maintained below 9 dB. h Voltage standing wave ratio (VSWR) is maintained minimum 1.8, demonstrating impedance matching. Collectively, these results underscore the versatility and high performance of solution-processed MoS2 switches for realizing advanced mmWave phase control circuits. RF circuit implementations of solution-processed MoS2 switches with typical performance. Phase shifters fabricated using solution-processed MoS2 switches, demonstrating two distinct design topologies with a CPW structure. a Circuit schematic of a 2-bit true-time delay (TTD) phase shifter on a sapphire substrate. This device employs single-pole 4-throw (SP4T) switches to selectively route the RF signal through a reference path or one of the delay lines, providing phase shifts in 90° increments up to 270° by utilizing delay lines with progressively increasing electrical lengths (from λ to a λ + 3/4λ). b Circuit schematic of a 2-bit, 90° hybrid coupled phase shifter (HCPS) on sapphire substrate. Z0 represents the characteristic impedance of the transmission line, while ZSn represents the equivalent impedance of the corresponding switch elements. This reflective-type design utilizes a 90° hybrid coupler terminated by identical reflective loads, each composed of a λ/8 transmission line stub and shunt-configured MoS2 switch to control the reflection phase. c Relative phase shift of the TTD phase shifter targeting 10 GHz, showing four well-defined states separated by stable 90° increments. d Corresponding S21 insertion loss of each state; the inset shows the root mean square (RMS) insertion loss error in dB. e Relative group delay variation of each state showing root mean square (RMS) group delay error ≤4 ps and a ~25 ps increment per state, confirming TTD functionality. The inset shows the RMS group delay error in ps. f Performance of the HCPS targeting 30 GHz, detailing four distinct operational phase states with ~90° steps. g Insertion loss maintained below 9 dB. h Voltage standing wave ratio (VSWR) is maintained minimum 1.8, demonstrating impedance matching. Collectively, these results underscore the versatility and high performance of solution-processed MoS2 switches for realizing advanced mmWave phase control circuits. For the HCPS, the design utilizes a 90° branch-line hybrid coupler (Fig. 4b). This reflective-type design splits the input signal into quadrature paths terminated by stubs. By integrating shunt switches to toggle the effective electrical length of these stubs (λ/8), a 2-bit phase shift in 90° increments is achieved at a center frequency of 30 GHz, based on Eq. (6):6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \phi={\phi }_{2}-{\phi }_{1}=4\pi \frac{{l}_{2}-{l}_{1}}{\lambda }$$\end{document}Δϕ=ϕ2−ϕ1=4πl2−l1λwhere l1 and l2 are the lengths of the stubs. This architecture also cancels reflections at the isolated port. The optical micrograph (Supplementary Fig. 40) confirms the compact 51 mm2 layout. Measurement results (Fig. 4f–h) validate the design, showing four distinct phase states with stable steps of 90° ± 14° at the 30 GHz target frequency. The voltage standing wave ratio (VSWR) reaches a minimum value of approximately 1.8 across the three states., indicating effective impedance matching. Moreover, owing to the compact switch footprint (4 × 4 µm2) and the resulting short channel length, the additional phase delay introduced by the switch itself is negligible over the operating bands, ensuring that the measured phase response is dominated by the intended transmission-line delays rather than the switch parasitics. Taken together, these results verify the practical circuit-level applicability of the proposed switches, and a benchmarking comparison with previously reported phase shifters based on conventional and emerging memory-based RF switch technologies is provided in Supplementary Table 2. Benchmarking of mmWave switches A comprehensive benchmarking of our device against other switching technologies is presented in Fig. 5 and Supplementary Table 1. The radar chart (Fig. 5a) highlights the key strengths of the present platform in a class-specific manner: zero static power compared with conventional active switch technologies, low switching energy and low actuation voltage compared with emerging memory-based RF switches, and high fco as an advantage that remains favorable across both conventional and nonvolatile RF switch platforms. A normalized comparison against leading conventional foundry technologies such as RF SOI, RF Silicon-on-Sapphire (SOS), high electron mobility transistor (HEMT), and GaN in Fig. 5b reveals a performance advantage for our device in all configurations9,43,50–64. Notably, the shunt achieves an RON·COFF FoM of ~0.8 fs, placing it firmly in the sub-femtosecond regime and confirming its potential for mmWave applications. This performance gap is fundamental: conventional switches are limited by substrate parasitic capacitance from their front-end-of-line (FEOL) fabrication. Furthermore, integrating these FEOL devices with back-end-of-line (BEOL) components like antennas and inductors introduces interconnect losses15. In contrast, our solution-processed switch is fully compatible with BEOL integration, enabling it to maintain intrinsic high performance and achieve more favorable, low-loss integration with the RF front-end without expensive SOI technologies. Figure 5c extends this comparison to other emerging memory-based RF switches, highlighting key trade-offs. While other 2D atomristors and conductive bridge random access memory (CBRAM) achieve high fco, they typically require material transfer processes, posing a barrier to CMOS integration3,24–27,29,37,65. Our device, however, achieves a high fco of 187 THz using a scalable, transfer-free solution process. Other technologies present their own limitations: oxide-and polymer-based memristors suffer from high IL due to large RON5,7,18,20,22,66–69; and PCM and insulator–metal transition (IMT) devices face heater-integrated barchitectural constraints6,12–14,17,31,47,70–73.Fig. 5Benchmarking solution-processed MoS2 switches against conventional and emerging memory-based mmWave technologies.a Radar chart comparing the key performance metrics of our device with representative conventional and emerging memory-based RF switch technologies, highlighting the characteristics of the 2D material-based approach, including high fco, minimal switching energy and delay, and negligible static power consumption. b Normalized RON versus COFF benchmark against leading RF SOI, GaN-on-SiC, and GaN-on-Si technologies from conventional foundries. Data are normalized by channel width to account for parasitic capacitances. Gray dashed lines indicate constant RON·COFF contours. Our device achieves an FoM in the sub-femtosecond regime (~0.8 fs), surpassing conventional counterparts. Comparative plots against other conventional and emerging memory-based mmWave technologies; c fco showing the higher bandwidth of our devices. d Switching voltage versus DC ON/OFF ratio; and e frequency vs. RF ON/OFF ratio. Notably, despite a modest DC ON/OFF ratio, the device achieves an RF ON/OFF ratio due to its optimized structural design. f Reliability assessment comparing endurance and data retention with other reported 2D material-based mmWave switches, demonstrating 2000 switching cycles of endurance and 106 s of retention. In all panels, data points for the device in this work are marked with red stars. Benchmarking solution-processed MoS2 switches against conventional and emerging memory-based mmWave technologies. a Radar chart comparing the key performance metrics of our device with representative conventional and emerging memory-based RF switch technologies, highlighting the characteristics of the 2D material-based approach, including high fco, minimal switching energy and delay, and negligible static power consumption. b Normalized RON versus COFF benchmark against leading RF SOI, GaN-on-SiC, and GaN-on-Si technologies from conventional foundries. Data are normalized by channel width to account for parasitic capacitances. Gray dashed lines indicate constant RON·COFF contours. Our device achieves an FoM in the sub-femtosecond regime (~0.8 fs), surpassing conventional counterparts. Comparative plots against other conventional and emerging memory-based mmWave technologies; c fco showing the higher bandwidth of our devices. d Switching voltage versus DC ON/OFF ratio; and e frequency vs. RF ON/OFF ratio. Notably, despite a modest DC ON/OFF ratio, the device achieves an RF ON/OFF ratio due to its optimized structural design. f Reliability assessment comparing endurance and data retention with other reported 2D material-based mmWave switches, demonstrating 2000 switching cycles of endurance and 106 s of retention. In all panels, data points for the device in this work are marked with red stars. Figure 5d and e reveals advantages of our platform. Although the solution-processed MoS2 exhibits a modest DC ON/OFF ratio, the architectural design (shunt and SS) and precise RLRS control via CC enable an RF ON/OFF ratio (S21,ON–S21,OFF). This result outperforms other emerging memory-based RF switches and demonstrates a design principle: intelligent architecture can compensate for non-ideal DC material properties to achieve good RF performance. This concept can be broadly applied to enhance other memristive switch systems. These results and their associated trade-offs are discussed further in Supplementary Note 6. Finally, Fig. 5f benchmarks the device reliability, revealing an advancement for MoS2-based mmWave switches24,26,27,29,74. Whereas previous reports were limited to ~600 cycles29, we demonstrate an extended endurance of 2000 cycles. This durability is comparable to that of the most reliable 2D-based switches reported, such as those using hBN26. Furthermore, the device exhibits data retention exceeding 106 s, confirming the enhanced stability of our solution-processed platform. Discussion In summary, we have developed a sub-femtosecond mmWave switch platform based on solution-processed MoS2. We demonstrated that a shunt delivers high performance, including a low insertion loss (<0.1 dB), high isolation (>35 dB), and an extrapolated RON·COFF time of ~0.8 fs (fco ~ 187 THz)—an advancement over prior 2D material-based switches. Comprehensive mechanistic analyses confirmed that the edge-rich morphology of the solution-processed film acts as a natural defect-mediated template, facilitating efficient filament formation via VS, achieving an endurance of 2000 cycles and low, nanojoule-scale switching energy with enhanced uniformity of resistance. We further demonstrated the platform’s versatility by realizing variable attenuators and varactors and introduced an operational scheme to ensure high linearity (IIP3 = 42.1 dBm) and power handling (P0.1dB > 10 dBm) by eliminating self-switching. The successful integration of these switches into complex SPDT, 2-bit TTD, and hybrid-coupled phase-shifter circuits confirms their viability for practical circuit integration. This work provides a scalable, high-performance solution for advancing future mmWave communication. Methods Device and RF circuits fabrication The fabrication process began with the definition of a 50 Ω characteristic impedance-matched CPW via photolithography on a 430 μm sapphire wafer. An electron-beam evaporator was used to deposit the BE and ground pads, consisting of a Cr/Au (5/55 nm) stack. For the active layer, a ~30 nm-thick MoS2 film was created by spin-coating a 0.5 mg/mL solution (synthesized via THAB-assisted electrochemical intercalation and mixed in isopropyl alcohol, see Supplementary Fig. 1) at 2500 rpm for 40 s, followed by a 5 min bake at 70 °C. This MoS2 layer was then patterned into the active region using photolithography and reactive-ion etching (RIE). The device was completed with the deposition of the TE stack of Cr/Cu (5/85 nm). An identical procedure was used for the RF circuits, with the addition of an in-house Cu electroplating process for the phase shifter pads with 2 μm thick (Supplementary Fig. 2). Material characterization Confocal Raman spectroscopy was performed using a WITec Alpha 300R system equipped with a 532 nm laser source (2 mW power) and an 1800 grooves/mm grating. Confocal Raman was calibrated using a Si/SiO2 sample. Samples were prepared for analysis by coating the material onto a sapphire substrate, following the same procedure used for device fabrication. At eight arbitrary points on the sample surface, spectra were acquired by accumulating 10 integrations with an exposure time of 1 s per integration. cAFM and KPFM analyses were performed using an NX7 (Park Systems). For cAFM measurements, an NSC18/Cr–Au tip was used. The TE removal was performed through diluted Cu etchant with 0.5 Å/s of etch rate. To ensure a reliable conductive pathway, the bottom electrode of the device was electrically connected to the sample disk with silver paste, and the bias voltage was applied directly to the sample disk. For KPFM measurements, MoS2 flakes were spin-coated onto a Si wafer, which was then electrically connected to a grounded sample disk with silver paste. An NSC36/Cr–Au C tip was utilized for KPFM measurements. Prior to scanning, the system was calibrated to 4.6 eV using a highly oriented pyrolytic graphite (HOPG) substrate. During the scan, topology and surface potential data were acquired simultaneously. The topology was measured via the first lock-in amplifier (modulation frequency 54 kHz), while the KPFM data was obtained from the second lock-in amplifier (drive voltage 1 V, AC tip bias 17 kHz). DC and low-temperature characterization A Keysight B1500A semiconductor parameter analyzer, supplemented with a B1530A WGFMU pulsing module, was the primary instrument for DC electrical characterization at room temperature in a dark shield environment. Low-temperature analysis was conducted with a Keithley 4200-SCS system with Lakeshore cryogenic probe station CRX-4K, where the temperature was swept from 100 to 300 K under high vacuum (10−6 torr). The top electrode was consistently biased for all electrical tests. Hot switching measurements Hot switching measurements were conducted on a probe station equipped with T-plus TP26 GSG RF probes. A Keysight P5024A vector network analyzer, operating in CW time mode with 1 GHz and 0 dBm, served as the continuous wave RF source, while a Keysight B1530A WGFMU supplied the input pulse to the device. A Tektronix PSPL5585 bias tee was employed to distinguish the pulse signal and the RF signal. To capture the transient switching event, a Keysight MSOX6004A oscilloscope was connected to monitor both the RF signal transmitted through the device and a reference signal from the WGFMU’s remote sensing unit (RSU) monitor line, using a trigger. Any residual pulse artifacts in the captured RF signal were subsequently removed through post-processing with custom-developed software. For a detailed diagram of the measurement setup, please refer to Supplementary Fig. 8. RF mmWave measurements On-wafer measurements are performed using the Cascade Infinity GSG probes and Keysight N5247A vector network analyzer from 10 MHz to 67 GHz. For small-signal measurements, the RF power was set to a nominal level of −20 dBm. Before actual device measurement, the SOLT calibration was carried out from 10 MHz to 67 GHz using a Cascade 138-356 Impedance Standard Substrate and Cascade WinCal software. Scattering parameters were measured for each device in both the ON- and OFF-state, and parameter extraction was carried out to determine the ON- and OFF-state equivalent lumped circuit component models. Power handling was also evaluated in both the ON- and OFF-state by measuring scattering parameters as the devices were gradually exposed to RF power levels from −20 to +10 dBm. The bending measurement was conducted using a 3D-printed curved structure. For a detailed diagram of the measurement setup, please refer to Supplementary Fig. 21. IMD measurement A two-tone test was conducted to assess the third-order intermodulation distortion. A Keysight P5024A VNA was used to both generate two input signals (f1 = 2.365 GHz, f2 = 2.415 GHz) and to measure the power levels of the output signals. These tones were combined by an RFCM APDS2-0580-01S power divider and delivered to the device through T-plus GSG probes on a probe station. The input power was swept across a range of –30 to +10 dBm. Accurate power levels were ensured by calibrating the VNA with a Keysight U8487A power sensor. The output signals at the fundamental and the 2.465 GHz (2f2–f1) intermodulation frequency were captured using the Keysight Network Analyzer software. For a detailed diagram of the measurement setup, please refer to Supplementary Fig. 21. Full EM-wave and high-frequency circuit simulations Electromagnetic simulations were conducted using Keysight ADS 2020, and COMOSL software.The model incorporated a 430 µm-thick sapphire substrate with properties given in ADS (dielectric permittivity εr = 9.3, loss tangent tan δ = 0.00002). The conductors were modeled as copper (Cu) and gold (Au) with a conductivity of σCu = 59.6 × 106 and σAu = 41.1 × 106 S m⁻¹, respectively. The thickness was equivalent to that of the electron-beam evaporated films to ensure consistency with the fabricated devices. Supplementary information Supplementary Information Transparent Peer Review file Supplementary Information Transparent Peer Review file