State of the Art Sub-Terahertz Switching Solutions

In this paper the state of the art in RF switches for mm-wave frequency range is summarized and evaluated. Several leading technologies is presented from typical semiconductor devices based on transistors and diodes on Si, SiGe or III-V semiconductor substrates to more unconventional solutions such as microelectromechanical or phase-change material switches. The most important parameters and characteristics for those technologies are gathered and compiled for comparison. Besides different technologies, also various switch topologies of mm-wave switches are presented, assessed and compared. Furthermore, new emerging technological solutions approaching mm-wave range involving two-dimensional materials are also presented. Their evaluation is focused on proposed designs and current results for experimentally evaluated prototypes. Although the performance of these devices are currently not competitive with more traditional approaches, some reported results near the mm-wave range makes them a promising solution for future mm-wave switches and an interesting topic for further research and development.


I. INTRODUCTION
Efficient millimeter wave switches are important components for development of new communication, sensing, imaging, testing and instrumentation systems [1]- [3]. Applications of such switches include switched -beam reconfigurable antennas, polarization switching, multiband receivers, transceivers, time division duplexing systems, radiometers, test circuits with multiple signal paths. Moreover, integration of the switches allows building these systems as system-on-chip millimeter wave monolithic integrated circuits (MMIC) or system-in-package solutions for greater miniaturization and increased efficiency. However, to allow on-chip integration, the switch technology has to be compatible with other parts of the integrated system such as amplifiers, digital signal processing circuits and in some cases also antennas.
There are a number of technologies available for millimeter wave switches. Most of them employs semiconductor devices produced with typical manufacturing technologies The associate editor coordinating the review of this manuscript and approving it for publication was Teerachot Siriburanon . using silicon or III -V semiconductors. Among them we can distinguish transistor-based and diode-based solutions. Both bipolar (mostly heterojunction bipolar transistors -HBT) and field-effect transistors (using CMOS or high electron mobility transistor -HEMT technologies) are used in mm-wave applications. Besides them, there are less conventional technologies available such as micro electro mechanical systems (MEMS) and phase changing materials (PCM). An emerging new technology that can potentially bring significant improvements in performance of the future switching components is application of two-dimensional (2D) materials. As different technologies and types of switches excel in different parameters, it is beneficial to gather together and compare these devices in order to select the most suitable solution for particular application.
In this paper, the extensive review of state-of-the-art in mm-wave (30 to 300 GHz frequency range) switching devices is presented comparing different technologies as well as switch construction topologies. Besides the devices of highest performance, also some new technologies currently bringing considerable research interest are highlighted. These new devices at their current state of development show some VOLUME 10, 2022 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ performance deficiencies, mostly in terms of operating frequency range, switching speed or power consumption. On the other hand, they offer low insertion loss and high isolation exceeding capabilities of state-of the art devices in more common and mature technologies.

II. TRANSISTOR -BASED SWITCHES A. CMOS
The simplest and lowest cost semiconductor technology is a silicon complementary metal oxide semiconductor (CMOS) process. It is a well-established technology widely used in the industry using field-effect transistors. Another advantage of CMOS switches is possibility of their integration on the same semiconductor die with other components of the electronic devices.
The most typical topology used in CMOS switches is a λ/4 shunt [4]- [9]. It uses a quarter wavelength sections of transmission lines as impedance transformers ( fig. 1). In this topology input terminal of the switch is connected to two branches with quarter-wave transmission lines. Between the ends of these lines and switch outputs the transistors are connected between line and ground. When the transistor is in the on-state, it lowers the impedance at the end of the λ/4 line which is transformed to high impedance at the input. That isolates this branch of the switch. On the opposite branch the transistor stays in off-state and it is matched to line's characteristic impedance with parallel stub made from another section of transmission lines. This topology is widely used in most of the available RF switches. However there are significant drawbacks, especially for high frequency applications. The transmission lines used in this type of construction manufactured in semiconductor typically have high losses due to lossy substrate and also occupy large area on the chip. Another problem is a tradeoff between isolation and bandwidth. As the isolation is dependent on the low impedance of the transistor in the on-state, wider transistors are required. On the other hand increasing the transistor width increases it's off-state equivalent capacitance which decreases the available bandwidth. Shunt switches using 90nm process presented in [4] allows to achieve insertion loss level of below 1.9 dB and isolation better than 25 dB in 50-70 GHz. The 65 nm process allowed to achieve slightly better performance as FIGURE 1. Schematic diagram of an SPDT switch with shunt topology [4].
shown in [5] and [6] with insertion loss less than 1.6 dB in 0-94 GHz band and less than 3.6 dB in 110-170 GHz band. Isolation of 19-32 dB and over 19.5 dB was achieved in the respective bands.
The Silicon-On-Insulator CMOS (CMOS SOI; fig. 2) implementation for mm-wave shunt switches was also reported [8] which can be used to achieve better performance of CMOS switches at higher frequencies with insertion loss less than 1.5 dB and isolation around 20 dB in 140 -220 GHz band. Another COMS SOI shunt switch implementation presented in [9] allowed to achieve better isolation (26.5 -32 dB in 110-170 GHz band) but at the expense of higher insertion loss from 4 to 5 dB. A distributed switch with 4 shunt stages was also reported in SOI technology [10]. It exhibits high operation bandwidth from DC to 220 GHz with high isolation (30 -50 dB) and moderate insertion loss (< 3.1 dB). Another proposed topology for the mm-wave switches is a travelling-wave switch [11]- [13]. This topology ( fig. 3) is similar to shunt switches and also uses sections of transmission lines. These highly inductive sections of transmission line are loaded by transistors. Several stages of this structure are used to obtain periodically loaded inductive transmission line. In the off-state of the transistors, their capacitance loads the inductive line to provide impedance matching which allow transmission (''artificial transmission line''). When transistors are in the on-state, the signal path is shorted to ground through their on-state resistance. The quarter-wave impedance transformer is added to the input side of the switch to transform short into open.
Application of travelling-wave switch concept in CMOS devices allowed to achieve higher isolation (over 27 dB in 50-94 GHz band [11] and over 40 dB in 70-110 GHz band [12]; fig. 4) compared to λ/4 shunt technique. These switches also offer higher power handling capabilities compared to other CMOS devices (IP1dB > 15 dBm). However their insertion loss is higher with below 3.3 dB in [11] and below 3.4 dB in [12]. Similar results for this type of structure can be also found in [13].
Further development of travelling-wave distributed switch with coupled inductors instead of transmission line sections was proposed in [14] which allowed to increase the isolation to over 33 dB in 54-84 GHz range and simultaneously decrease the insertion loss to below 2.5 dB also offering smaller occupied wafer area. This type of distributed switch was also reported in CMOS SOI [10] with very high operation bandwidth (DC -220 GHz) and good insertion loss and  isolation characteristics (<3.1 dB and >37 dB, respectively in 60-220 GHz band).
Another two topologies presented for CMOS switches employ artificial resonator [15] and hybrid coupler [16]. First of them uses three coupled transmission lines ( fig. 5). The one in the center is connected to input port and grounded on the opposite side. Two lines on the sides are connected to output ports and shunt transistors. When one of the transistors is in on-state, the line is shorted and port connected to it is isolated. At the same time the second transistor is in the off-state and it's parasitic components with coupled lines form artificial resonant network with bandpass transmission response from input to coupled output. This type of switch exhibits similar performance in terms of insertion loss (3.5-4 dB in 130-180 GHz band) and isolation (22-23 dB) to λ/4 shut switches, but offers significantly smaller occupied wafer area. It is also one of the fastest published CMOS switches in this frequency range with switch time around 580 ps.
Another example of a CMOS switch using coupled transmission lines can be found in [17]. This device operates in higher frequency range (290-320 GHz) with relatively high 5-7 dB insertion loss and high isolation around 40 dB. Results for both examples shows that the main drawback of coupled line based designs is narrow operation bandwidth. In [16] the concept of using 90 • hybrid coupler with direct and coupled ports loaded with shunt transistors to achieve a SPST switch ( fig. 6). It offers small occupied area, but its performance is inferior compared to other published CMOS mm-wave switches. Among CMOS-based devices, there are also commercially available switches for mm-wave region. However their operating frequency is currently limited to about 60 GHz. An example of such device is PE42525 SPDT switch with operating frequency range from 9 kHz to 60 GHz [18]. It offers insertion loss below 3dB and isolation better than 35 dB in whole operating range.
One of the important limitations of all silicon CMOS devices is the low voltage operation which limits their power handling capabilities (maximum reported IP1dB around 15 dBm). Increasing transmitted power requires higher bandgap III -V semiconductors.
which heterojunction bipolar transistors (HBT) can be fabricated. This type of transistors allowed for decrease of the parasitic components and increase the transition frequency compared to silicon CMOS technology. It also offers better power handling capabilities still allowing high level of integration and low manufacturing costs.
Similarly to CMOS devices most SiGe mm-wave switches employs λ/4 shunt topology. SiGe HBT transistors ( fig. 7) with less parasitic components allow decreasing the insertion loss compared to CMOS switches. Shunt switches shown in [19]- [21] offer lowest insertion loss values in mm-wave range among SiGe devices and noticeably better than CMOS devices at less than 1.25 dB in lower frequency band   [19] and between 1.1 dB and 1.5 dB in 75-100 GHz band [20], [21]. Isolation of these switches is comparable with CMOS devices (18-22 dB). The highest isolation in SiGe shunt switches was achieved in [22] and [23] (21-29 dB [22] and 26-31 dB [23] in 80-170 GHz and 110-170 GHz bands, respectively), however this values are still comparable with CMOS, and these devices exhibit higher insertion losses in the 2.6-4.5 dB range.
It should also be noted that SiGe shunt switches can achieve significantly faster operation (75 ps in [19]) and also higher power handling capabilities compared to CMOS switches (IP1dB = 35 dBm in [20]).
Better performance with simultaneously higher isolation, lower insertion loss as well as operation from DC can be achieved in SiGe using series-shunt topology ( fig. 8) proposed in [24]. This type of switch utilizes three transistors. Two of them, in anti-parallel configuration form a series switching element which is followed by single shunt transistor. In this configuration insertion loss of 1.3-3.2 dB and isolation of 20-40 dB in 0-110 GHz frequency range was achieved. Presented device also exhibits fast switching time of under 250 ps.
Other notable SiGe device topology is a through-load switch ( fig. 9, 10) proposed in [25]. This two-port device can switch between two states: a through connection or a 50 load for both of its ports. It uses a 3 dB directional coupler composed with coupled lines and a two transistors connecting coupled ports to ground. In the off-state transistors   represent high impedance at coupled ports which causes low loss transmission between remaining ports. In the on-state of the transistors represent impedance close to 50 obtained by optimization of transistor size. In such case the signal is dissipated in the transistors and ports are isolated. This configuration allowed to achieve good performance (1.4-2.2 dB insertion loss and 20-30 dB isolation) and also a low occupied wafer area of 0.019 mm 2 .
SiGe HBT switches operating at higher frequencies (220-280 GHz) based on resonator topology was also reported [26]. These devices are constructed similarly to CMOS devices presented in [15] and achieve isolation comparable with SiGe switches for lower bands, however the insertion loss is significantly higher (4.2-5.3 dB insertion loss and 14-29 dB isolation).

C. InP DHBT
Next semiconductor technology for mm-wave switching is a InP double heterojunction bipolar transistor (DHBT) [27]- [31]. This type of transistor has heterojunctions at both emitter-base and base-collector junctions. It allows to obtain high frequency operation, low base-collector capacitances and fast switching with higher breakdown voltage than single heterojunction HBT's.
The InP DHBT for the 90-170 GHz reported in [27] uses double λ/4 shunt topology. It achieves significantly better isolation (>45 dB) compared to CMOS and HBT devices, but at the cost of slightly higher insertion loss than CMOS. The power handling capabilities are similar to both previously mentioned technology. This technology is suitable for integration and offers good performance with low power consumption and moderate costs.
Devices operating at higher frequencies were also presented in InP DHBT [28]- [31]. First of them [28] is a typical λ/4 shunt construction operating in 220-325 GHz frequency range. It offers insertion loss under 4.1 and high isolation of around 35 dB. It also exhibits higher power handling capability compared to most CMOS and HBT devices with IP1dB >20 dBm. Similar insertion loss (3-6 dB) with higher maximal isolation figures (up to 50 dB in 90-325 GHz range) was reported in [29]. Lower insertion loss shunt devices were also reported (1.2 -2.9 dB), however with significantly lower isolation (13.1-15.7 dB) [30]. Another notable device [31] uses amplifier topology and offers slightly lower insertion loss under 4 dB and also lower isolation over 20 dB. However it is optimized for high speed operation with switching times around 40 ps which shows significant improvement over CMOS and HBT technologies.
It should also be noted that published InP DHBT transistor switches occupy significantly higher wafer areas (0.95 and 0.536 mm 2 for [27] and [28], respectively).

D. GaN HEMT
Another semiconductor material used for mm-wave switching is a GaN [32]- [34]. It offers high bandgap and high thermal conductivity which makes it suitable for high power density devices. GaN high electron mobility field effect transistors (HEMT) achieve transition frequencies and noise figures comparable to other semiconductors such as SiGe, InP or GaAs while at the same time their breakdown voltages are around five times higher.
GaN λ/4 shunt switch proposed in [32] offers lower insertion loss (0.9 -1.4 dB in 60-110 GHz frequency range) compared to all previously considered technologies with high IP1dB over 27 dBm. However its isolation is very low, over 9 dB in whole band. In [33] isolation was improved to values comparable with SiGe. Semiconductor die image of this λ/4 shunt switch is shown in fig. 11.
In [34] slightly modified λ/4 shunt topology was used to achieve improved performance compared to CMOS, HBT and DHBT transistors in wide frequency range   Similarly to InP DHBT's, presented GaN HEMT transistors occupy large wafer area from 0.3 to 1.5 mm 2 .

E. GaAs HEMT
Another semiconductor suitable for construction of HEMT transistors is GaAs. This semiconductor material exhibits high resistivity which leads to lower both parasitics and losses. At the same time it is suitable for high transition frequency transistors. VOLUME 10, 2022 Among the highest performance switches in GaAs travelling wave and shunt topologies are present [35]- [37]. Travelling wave switch shown in [35] exhibit low insertion loss under 2 dB and isolation over 30 dB which is comparable with state of the art CMOS and HBT devices in the same frequency range . In [36] another travelling wave switch is presented with wider operation frequency range (0-135 GHz), insertion loss (<1-5 dB) comparable to CMOS and HBT devices and improved isolation in 16-45 dB range.
In [37] device exhibiting lower insertion loss than in previously considered technologies (under 1.1 dB in 59-77 GHz range compared to [19] and [21]) with higher isolation over 25 dB. This device uses modified travelling-wave topology with transistors integrated with CPW transmission line ( fig. 14, 15).

F. InGaAs mHEMT
To avoid creation of deep level traps in heterojunction HEMT devices, buffer layer in the junction is added creating a metamorphic HEMT (mHEMT). In InGaAs metamorphic transistors are constructed with GaAs substrate, AlInAs buffer layer and InGaAs channel. This construction also allows to optimize performance of the transistors by changing indium concentration in the channel [38].
There are multiple examples of the InGaAs mHEMT based switches for mm-wave frequency range [39]- [44]. Most of them are based on typical λ/4 shunt topology ( fig. 16). In [39] a shunt switch operating in 53-150 GHz frequency range is presented. It achieves insertion loss under 3 dB which is only slightly higher than for state of the art SiGe HBT and GaN HEMT devices, but at the same time the isolation is  [41]. This device also exhibits higher isolation than HBT and GaN HEMT devices of comparable performance in terms of insertion loss. The power handling capabilities of these devices is also comparable to HBT switches (IP1dB around 19 dBm; fig. 17). InGaAs mHEMT switches for higher operating frequencies were presented in [40] and [42]. In [40] the insertion loss of 1.5-4.5 dB and 13.5-22.8 dB isolation was reported in 122-330 GHz range. Switch presented in [42] operates in 200-330 GHz band with insertion loss in 1.7-5 dB range and isolation over 14 dB. As can be noticed, performance of these devices is only slightly worse compared to aforementioned InP DHBT devices working in similar frequency range reported in [28] an [31].
Other noticeable topologies for InGaAs mHEMT switches were reported in [43] and [44]. Switch presented in [43]  It should be noted that transistors used in switching applications can also be used as active devices to amplify the switched signal [45]. Such devices can be used to compensate significant losses at high frequencies. The drawbacks of this approach are unidirectional operation of active switches and additional noise introduced by active circuit.
An interesting concept is presented in [44]. In this switch, the transistors are configured as diodes and used in distributed shunt configuration. The diodes created form HEMT transistors exhibit low Schottky capacitance and resistance lower than that of the transistor. Its resistance in still higher than PIN diodes, but it can be easily integrated into MMICs unlike PIN devices. The performance of this device is better than devices in other technologies mentioned before working in similar frequency range. Diode based switch exhibits 1.3-1.8 dB insertion loss and 32-41 dB isolation in 50-70 GHz with IP1dB of 20 dBm (fig. 19).

III. PIN DIODE SWITCHES
Besides transistors, there are other classes of devices available for mm-wave switching applications. First of them is a PIN diode. PIN diodes are widely used in RF switches due to their low parasitic capacitance, low series resistance and high power handling capabilities. There are also a significant drawbacks. First are different manufacturing requirements compared to transistors which are a major obstacle to integration of the PIN diodes into MMICs. Another weak sides of PIN diode switches are high control current requirements and high switching times.
PIN diode based switches reported in literature or commercially available typically use shunt topology and GaAs semiconductors [46]- [49]. In terms of insertion loss this type of switches can be comparable or slightly better than switches in other technologies presented earlier (1.1-1.6 dB in 75-100GHz range [46] and 0.75-2 dB in 85-105 GHz range [47]; fig. 20). However their isolation is moderate (over 21dB), and better overall performance can be achieved with InGaAs mHEMTs [41] with very similar insertion loss, but significantly better isolation. The IP1dB of the mentioned PIN switches was not reported, so possible advantages of PIN in power handling cannot be evaluated.
There are commercially available mm-wave switches based on GaAs PIN diodes. An Analog Devices HMC-SDD112 [48] is a 55-90 GHz SPDT shunt switch available as bare die. It offers insertion loss in 1.0-2.5 dB range and high isolation in 25-45 dB range ( fig. 21). The power handling and switching speed specifications are not provided by the manufacturer. It should be noted that this device requires high control current of 22 mA at 5V.
Another examples of commercially available mm-wave PIN switches are HSW1001 and HSW1003 waveguide switches manufactured by HXI [49] (fig. 22). These devices working in 75-110 GHz range are optimized for low insertion loss (HSW1001) or high isolation (HSW1003). First offers under 2.2 dB insertion loss with over 18 dB isolation while second exhibits under 3.5 dB insertion loss with over 40 dB isolation (fig. 23). These switches are characterized by switch-on times (150 and 175 ns, respectively) significantly slower than switch-off times (25 and 30 ns, respectively). These switching times are one to two orders of magnitude larger than reported for state of the art transistorbased devices.  Typically, PIN diodes are not compatible with MMIC manufacturing technological processes which hinders their integration. However, as reported in [50] there are techniques that allow PIN diode construction using typical SiGe BiC-MOS technology ( fig. 24) widely used for MMICs. Presented device, with 77-133 GHz operating frequency range is based on typical λ/4 shunt topology. It offers performance comparable to SiGe HBT devices (1.4-2 dB insertion loss and 19-22 dB isolation), but with higher power handling capabilities (IP1dB over 24 dBm, with measurement limited by available source power).
Another special type of PIN diode which can be used in mm-wave switching applications is a surface PIN diode (S-PIN). Contrary to typical PIN diodes, the S-PIN structure is lateral. It creates an area, which conductivity can be controlled by external bias. So far most reported S-PIN devices are waveguide based [51], [52] but the construction of the S-PIN diodes allows to use them also in planar transmission lines. As shown in [51], [52] this type of switching element can be integrated into mm-wave antenna arrays to obtain beam steering. The reported devices use S-PIN diodes manufactured using SOI technology to create reconfigurable radiating slots in the waveguide. Such antenna array can achieve ± 45 • beam steering range [51]. As the S-PIN switches are integrated with antennas, their insertion loss and isolation are not provided. In [52] switching speeds of 10 -20 µs are reported, which allowed for time-modulated  antenna array construction. An important drawbacks of the S-PIN are high power consumption and requirements for bipolar bias voltage.

IV. MEMS DEVICES
Another class of mm-wave switches are microelectromechanical system (MEMS) based devices. The construction of most of the high frequency MEMS switches is based on the shunt topology [53]- [60]. In the off-state of the switch, the transmission line (typically coplanar weveguide) is shorted by a conductive membrane. Therefore the reflective switches are typical for MEMS. The membrane of the MEMS switch is actuated electrostatically thus the required control voltage  is significanly higher (45-60 V [53], [58]) than in semiconductor based devices. It should be noted that control voltage for the MEMS can be significanly reduced with apropriate construction of the device. MEMS RF switches operating up to 60 GHz with actuatiuon voltage of 6 V were reported in [60]. Their RF performance is comparable with other MEMS devices with under 0.7 dB insertion loss and isolation in 20 -50 dB range in 30 -60 GHz band. Such devices can be a suitable for 5G applications. An interesting obsevation shown in [60] is that actuation voltage increases with operating temperature of the switch.
Membranes are typically made of metal but other conductive materials such as graphene are also possible [61]. MEMS devices used to be very difficult and expensive to integrate within MMICs due to complicated and incompatibole manufacturing processes. Nowadays MEMS -based switches were reported using typical IV group semiconductor materials [53]- [59]. They can be produced with processes compatible with other MMIC building blocks [54], [56] ( fig. 25). It is also possible to use MEMS devices in waveguide switches [58], [59] (fig. 27).
Due to low parasitics, these devices can achieve very low insertion losses, significantly better than transistor or PIN diode based switches in the similar frequency range (under 0.8 dB in 0-67 GHz range [53]; 0.8-1 dB in 70-110 GHz range [55] and 1.23-1.68 dB in 110-170 GHz range [56]; fig. 26). At the same time isolation is comparable (over 20 dB  in [62] and 20-30 dB in [55]) or better (18.25-54.5 dB in [56]). Devices operating at higher bands (220-325 GHz in [57]) were also reported with slightly higher insertion loss and lower isolation (under 1.1 dB and under 24 dB, respectively). MEMS devices can also work in high temperatures and provide high power handling capabilities. The main drawback of the MEMS switches is slow switching speed (in order of 10 µs [58]). In some applications such as time-modulated systems slow switching speed is main obstacle for using MEMS switch.

V. PHASE-CHANGE MATERIALS
An interesting class of mm-wave switches are devices using phase-change material (PCM). These devices typically utilize two materials: germanium telluride (GeTe) and vanadium dioxide (VO2) [62]- [68]. Aforementioned PCM materials are thermo-chromic, i.e. change their states between conductor and insulator under influence of heat. GeTe is a material widely used in the industry for manufacturing of optical storage media (CD, DVD) which leads to low cost and high availability. It has unique properties which allows creation of latching switches which require power only during switching time. As reported in [62] these times can be as low as 2 µs which is almost an order of magnitude faster than comparable performance MEMS devices ( fig. 28, 29).   Till now, GeTe switches with operating frequency range up to 67 GHz were reported [62]- [65]. As shown in [55] and [56] this type of device offers very low insertion loss (under 1.1 dB in [55] and under 0.6 dB in [56]) as well as high isolation (over 39 dB in [55] and over 24 dB in [65]). GeTe switches can also achieve high power handling capabilities (i.e. 35.5 dBm in [65]).
Reported PCM based switches with higher operating frequencies utilize VO2 [66], [67]. This material does not have latching properties and requires constant 20 mW of heat in on-state. In [58] VO2 PCM based switch operating in 1-110 GHz frequency range was presented ( fig. 30). It offers very low insertion loss under 0.6 dB, only comparable with MEMS devices and also good isolation between 15 and 45 dB ( fig. 31). The switching time around 2 µs is comparable with GeTe based device.
Even higher operating frequency (210-290 GHz) for VO2 based device was reported in [67]. This device also offers very low insertion loss (0.5-1.5 dB), but with moderate isolation (11-20 dB).The switching time for this device equals 10 µs. An important advantages of PCM switches are also high power handling capabilities (IP1dB over 30 dBm in [67]) and small size. The main drawbacks are significant power consumption (in case of VO2 material) and non-standard process which is not directly compatible with typical MMIC technologies. However some integration possibilities have been reported [68].

VI. 2D MATERIALS
A special class of materials which is of great interest for many diverse applications in electronics is two-dimensional (2D) materials. 2D material is a special atomically-thin form of matter which properties differs from bulk material of the same composition. Some of these materials exhibit features which can be utilized in construction of high frequency devices. The development of the 2D material based RF devices is still at the early stage, thus not many mm-wave devices were reported and their operation frequency is mostly below 100 GHz.

A. MoS 2 MONOLAYERS
One of the 2D materials proposed for mm-wave switches is molybdenum disulfide (MoS 2 ). This material, commonly known as lubricating agent is a semiconductor which can be  deposited in atomically thin monolayers. MoS 2 monolayers exhibit high direct bandgap which makes them suitable for transistor construction ( fig. 32).
An monolayer MoS 2 RF switch with 0-50 GHz operating frequency range was reported in [69]. It offers very low insertion loss (under 0.5 dB) compared to other transistorbased devices as well as PIN diodes ( fig. 33). The isolation figures are moderate in 15-50 dB range. These figures show that even on the early stage of development MoS 2 switches exhibit competitive performance to more mature semiconductor devices. Therefore further research is required to investigate the capabilities of MoS 2 and improve the construction of the switches.

B. GRAPHENE
Graphene is a novel form of carbon which can form 2D layers. Despite its low thickness graphene retains good thermal and electrical conductivity of carbon which makes is suitable for electronic devices. An interesting property of graphene nanolayers is the ability to change their conductivity with external voltage. Due to this phenomenon graphene is considered for switching applications. There are several publications that analyses the possibilities of graphene-based RF switches for millimeter-wave band [61], [70]- [77]. However,  most of these papers include only theoretical background and simulation results of such structures. Additionally operating frequencies of presented switches are below 100 GHz. Among aforementioned publications several different switch designs is shown. In [70] a design of SPDT switch based on slotted substrate integrated waveguide is presented ( fig. 34). The simulated results that the proposed device exhibits better than 3 dB insertion loss and 17 dB isolation from 59.5 GHz to 60.8 GHz.
Another design developed for 60 GHz band and based on coplanar waveguide (CPW) with graphene layers between conductors ( fig. 35) is presented in [71]. Simulation results shows insertion loss better than 5 dB and isolation between 32 dB and 45 dB in 20-70 GHz range ( fig. 36). Similar construction ( fig. 37) can also be found in [72].
Among very few experimentally examined graphene-based devices variable attenuators operating at the low end of the mm-wave range [73], [74] should be noted. The variable attenuators can be considered as a special type of the switch where there are multiple states between on-state and offstate. The device presented in [65] utilizes similar topology to the switches reported in [70]- [72] where voltage dependence of graphene resistance is used to change the state of the switch. This three-stage construction based on CPW operates   in 3.5-28 GHz frequency range with 2.5 dB insertion loss and 14 dB isolation. It should also be noted that this device exhibits high power consumption (390 mW) as the DC current flows through graphene elements.
In [74] two types of the variable attenuators for 10-40 GHz frequency range are examined. Both of them use variable resistance of grapheme element, first in CPW and second in slotline configuration. Contrary to construction reported in [73] one large two layer graphene element is used with additional insulating films. The insertion loss and isolation figures for both CPW and slotline attenuators are similar to those in [73] with 3 dB insertion loss and 15 dB isolation. However, application of insulating layers allowed for much better return loss characteristics of both deices (in the −20 to −40 dB range).
It should be noted that in these constructions the graphene layer itself is a switching element instead of being a part of semiconductor device. Another important characteristic of these devices is relatively high bias voltage (from 4 V in [74] to around 30 V [69], [71]) required to change state of the graphene layer.  Graphene layers can also be used to construct MEMS devices. This concept is investigated in [61], [75] and [76]. These structures are based on graphene nanolayer membrane suspended over specially constructed coplanar waveguide ( fig. 38).
When bias voltage is applied the membrane bends and creates a short circuit in transmission line. The membranes can be constructed from either monolayer or multilayer grapheme. The simulations shows that with this design and multilayer graphene membranes it is possible to obtain insertion loss below 0.3 dB and isolation over 20 dB in 1-60 GHz frequency range ( fig. 39). It should be noted that shown graphene-based MEMS devices require lower voltage to change state compared to typical metal-based MEMS switches.
Another concept for application of graphene in high frequency switching devices is a semiconductor device with grapheme monolayer used as a gate [77]. The reported device is based on AlGaN/GaN semiconductor structure similar to HEMT transistor embedded into a CPW with graphene layer used as a gate. The switch is constructed in series topology with switching device placed between two sections of CPW. In the on-state, without gate bias, the transmission occurs through two-dimensional electron gas channel at the AlGaN /GaN interface (fig. 40). When the graphene gate is polarized, it depletes the channel interrupting the transmission in CPW.
In contrast to typical HEMT transistors, bias voltage controls not only the conductivity of the channel but also the conductivity of the graphene gate. This property of graphene can be used to decrease the conductivity of the gate in the offstate. Therefore the parasitic coupling between two sides of the switch can be greatly reduced improving isolation. The switch reported in [77] operating in DC-84 GHz frequency range exhibits high isolation (17.5-77 dB), especially at low frequencies, but also very high insertion loss (7.4-19.4 dB). It also allows fast switching with rise and fall times of 25 ns and 17 ns, respectively. Considering the early stage of development of this type of switch and suboptimal design (i.e. series topology which typically introduces considerable losses) there is a room for improvement in high frequency performance. Fast operation and high low frequency isolation makes it a good candidate for further research on graphenebased switches.
If we consider a modulator as a switching device there is a considerable interest in application of graphene in optical modulators [78]. A significant number of publications on high-speed and high bandwidth devices operating in near infrared (around 1.5 µm/200 THz) can be found. However there are also reports of devices operating at frequencies as low as 200 GHz [79], [80]. As the devices are intended for operation in optical systems, they are configured for freespace incident beam modulation. Thus, they are impractical for application for typical mm-wave circuits. However it seems that straightforward adaptation for waveguide circuits is possible. Devices proposed in [79] and [80] are optically controlled which can be beneficial for some applications due to a fact that no electrical bias connections are required and they can be easily interfaced with optical communication systems. The principle of operation is similar for both reported devices. The active element of the modulator is a metamaterial created by deposition of graphene layer on a semiconductor plate. Device reported in [79] utilizes graphene on silicon (GOS) structure while in [80] graphene on germanium (GOG) is used. The metamaterial structure placed in the path of the mm-wave beam introduces very low attenuation without external excitation. In order to change the state of the modulator, the structure is illuminated with laser beam (40mW at 0.488 µm wavelength and 800 mW at 1.55 µm wavelength in [79] and [80], respectively). The carriers induced in the illuminated structure increase the conductivity and therefore attenuation of the mm-wave beam. The reported modulation depth can be as high as 99% for the GOS device operating in 0.2-2 THz range [79] and 94% for GOG device operating in 0.25-1 THz range. However, it should also be noted that modulation speed of these devices is very low at around 200 kHz [80].

VII. CONCLUSION
An overview of state-of-the-art mm-wave switches has been presented. Different manufacturing technologies as well as device topologies were considered in order to summarize their properties and compare performance. Additionally, novel unconventional solutions for mm-wave switching devices were highlighted.
The summary and comparison of the most important parameters of switches in different technologies as well as commercially available mm-wave devices is presented in Table 1. Considering transistor-based devices III-V semiconductor devices allow to achieve the highest operation frequency. However, advances in well-established CMOS technology enables operation of these devices up to 300 GHz. An important advantage of the CMOS devices is very straightforward integration with other circuits due to its prevalence in the industry. Therefore it can be used to create efficient, highly integrated system-on-chip solutions. The main drawbacks of CMOS are low power handling capability and slightly higher losses compared to other types of transistors. For high power applications switches based on GaN field effect or SiGe bipolar transistors can be used. In comparison with GaN transistors SiGe HBT's can achieve higher power handling, high speed switching, lower die size and cost, but at the expense of lower isolation figures. The highest switching speeds can be achieved in InP based solutions.
The PIN diodes, very common in microwave switches, can achieve similar loss, isolation and power handling performance to transistors but their operation frequency does not exceed 130 GHz. Moreover their low switching speed, large size and high power consumption combined with manufacturing process incompatible with MMICs makes them less attractive for mm-wave applications.
Among mm-wave switches there are also less conventional solutions compared with typical semiconductor devices. First of them is a MEMS, which allows to achieve significantly lower insertion losses than transistor based devices with comparable or higher isolation. Also the power handling capabilities of MEMS can be very high, although it is not reported for mm-wave devices. The main drawbacks of MEMS are very low switching speed, high control voltage requirements and complicated manufacturing process which hinders their integration. Another solution which allows to achieve very VOLUME 10, 2022 low insertion loss similar to MEMS is application of phasechange materials. An important advantage of PCM is almost an order of magnitude faster switching compared to MEMS. On the other hand, the PCM switches exhibit significantly smaller isolation figures and high power consumption due to actuation by heat.
A solution for future mm-wave switches which brings considerable research interest is application of 2D materials such as MoS 2 or graphene. Published results for MoS 2 devices operating at low end of mm-wave range shows performance competitive with typical semiconductor solutions, especially in terms of insertion loss. Therefore it is a promising technology for further research.
Graphene is a 2D material which diverse applications are a subject of great research effort in the last decade. However, mm-wave switching devices with graphene are on the early stage of development. Moreover, among investigated devices, very few were manufactured and experimentally evaluated. Most of the proposed solutions utilizes the phenomenon of graphene conductivity changes with applied voltage for controlling the attenuation in transmission lines. Graphene can be also used as a membrane in MEMS devices. Another proposed application of graphene in mm-switching is semiconductor device with graphene used as a gate. Reported performance of the first prototypes of such device shows very good isolation at low frequencies and low switching times. Therefore, it is a promising solution for further development and optimization.