**5. Switch Calibration & Measurement**

In order to verify the switch design, the RF MEMS switch is fabricated and tested. An on-wafer measurement is set up as presented in Figure 16. A Keysight VNA (PNA-5245A) is used to conduct a two-port S-parameter measurement. Four sets of VNA extender pairs (VDI WR-5.1, WR3.4, WR2.2, and WR-1.5) are used to up-convert VNA test frequency sweeping to 140–220 GHz, 220–330 GHz, 330–500 GHz, and 500–750 GHz. To conduct the on-wafer measurement, DMPI T-wave probes (WR5.1, WR3.4, WR2.2, and WR1.5) are also used correspondingly [25]. An SEM image of the calibration kit and completed RF MEMS contact switch is provided in Figure 17. The switch's DC high voltage bias and ground are provided through a bias tee integrated into the DMPI probes. A series 10 MΩ resistance is placed in the DC bias circuit to limit the current below 10 μA in the potential breakdown condition.

Before measurement, on-wafer calibration is applied using Through-Reflection-Line (TRL) standards and WinCal XE calibration software. Standards include a 120 μm through, two reflection and three lines. The three different lines have additional lengths of 45 μm, 67 μm and 113 μm. Redundant reflections and lines are used for higher calibration accuracy [25]. Calibrated measurement shows the loss of 50 Ω CPW is around 3.5 dB/mm at 500 GHz and around 6.5 dB/mm at 750 GHz.

**Figure 16.** On wafter two-ports probing set up used for MEMS switch RF measurement is shown in this diagram.

**Figure 17.** SEM image of (**a**) on-wafer TRL calibration kit and (**b**) an example silicon switch.

The silicon switch measurement is compared with corresponding HFSS simulation results in Figure 18a,b. The ON state performance is shown in Figure 18a. The measurement shows the switch's impedance is well matched and has a small variation over frequency. The return loss is better than 10 dB in the whole band. The switch's insertion loss is thus dominated by the actuator contact resistance and metal loss. Previous discussion demonstrated the contact resistance will bring insertion loss merely on a scale of 0.1 dB across the whole band; meanwhile, the metal loss will cause switch insertion loss to increase

slightly over frequency. The measured insertion loss (S21 and S12) in Figure 18a fits the expectation. The return loss (S11 and S22) is mostly better than 20 dB as HFSS simulation suggested. Because the measurement is conducted in four different waveguide bands through four calibrations, there are inevitably certain minor measurement errors at the boundary frequency points. However, the return loss measurement trend still fits the simulated curve.

**Figure 18.** The THz MEMS switch HFSS simulation measurement comparison. (**a**) Silicon switch ON state measurement and simulation. (**b**) Silicon switch OFF state measurement and simulation. (**c**) Quartz switch ON state measurement and simulation. (**d**) Quartz switch OFF state measurement and simulation.

The silicon switch's OFF state measurement is presented in Figure 18b. Because the CPW sections are isolated by the air-gap capacitance, any incident RF energy is mostly reflected. In this measurement, the return loss (S11 and S22) curves are only slightly lower than 0 dB, which fits the simulated result well. The isolation plot (S12 and S21) also match the simulated curves.

A similar condition is also observed in the fused quartz switch's ON and OFF state measurements. Comparing Figure 13a,c, the silicon switch's impedance match is expected to be better than the fused quartz switch in the 200–500 GHz band, which is due to different bias pad-actuator coupling conditions and associated resonance. The measurement in Figure 18a,c matches such expectation. Meanwhile, the fused quartz switch and silicon switch have similar insertion loss conditions, which is very reasonable. In both switch designs, insertion loss is majorly determined by the actuator's contact resistance and gold metal loss. In both switches, the actuation bias voltage is 55 V, and both actuators have the same thickness. The actuation force is expected to be similar, which leads to the same expected contact resistance. Gold is used in both switch designs, which provides the same metal loss over frequency. Those factors determine both switches to have fairly similar and constant insertion loss over frequency; as frequency increases, the insertion loss increases by a similar magnitude in both silicon and fused quartz designs.

The quartz switch measurement in Figure 18d also follows the HFSS simulation result well. Similar to the silicon switch's OFF state measurement, the fused quartz switch has 0.3–0.4 fF air gap capacitance. The impedance of this capacitance is dominant compared with 50 Ω transmission line impedance. As frequency increases, the reduced capacitance impedance leads to reduced isolation and rising S21/S12 measurement curves over frequency.
