The State of RF and Microwave Switches

RF and microwave switches are used extensively in wireless systems for signal routing, finding wide use in switching signals from antennas to the transmit and receive chains. They are one of the highest volume RF devices in use today, as several devices are typically contained in a block diagram. RF and microwave switches fall into the two main categories of electromechanical and solid-state switches. While electromechanical switches have not found wide use in RF and microwave applications since the PIN diode was developed, they are making some new in-roads in certain applications in the form of micro-electromes in certain applications in the form of micro-electromechanical systems (MEMS) devices. Solid-state switches are typically more reliable and exhibit longer lifetime than electromechanical switches, plus offer faster switching times. However, solid-state switches typically have higher intrinsic ON resistance and more harmonic distortion than mechanical switches.

Today’s CMOS silicon-on-insulator (SOI) and silicon-on-sapphire (SOS) switches are starting to challenge GaAs MMIC switches in many applications as their cut-off frequencies and breakdown voltages improve. Each technology has it advantages and disadvantages, which can be leveraged for various applications for an optimal solution. Up until a few years ago, MEMS switches were considered an emerging technology that had reliability and reproducibility issues. However, recent generations have solved many of those problems making them very competitive in several applications.

Oldie but Goodie
Starting in the 1950s, PIN diodes were the first widely used solid-state switching technology and are still in wide use today. They excel in very high power and high frequency applications with low insertion loss and better power handling capabilities when compared to most IC FET switches. A PIN diode operates as a variable resistor at RF and microwave frequencies. Its ON resistance varies from less than 1 ohm (ON) to more than 10 kohms (OFF) depending on the bias.¹ One limitation of a PIN diode switch is its lower frequency limit of a few kHz to about a MHz depending on the thickness of the intrinsic or I (intrinsic) region. Therefore, they do not operate all the way down close to DC like most IC switches such as GaAs MMICs or CMOS RFICs. They also require more current to operate compared to IC switches, which means they are typically not a good fit for mobile applications. However, they can be more desirable for higher power levels used in military, Satcom or base station applications.

Figure 1 Series PIN SPDT switch (a), shunt PIN SPDT switch (b) and series-shunt PIN SPDT switch (c).¹

As explained in an Agilent application note,¹ PIN diode switches can consist of a mixture of series and/or shunt diodes depending on the circuit requirements. Series PIN diodes can function within a wide bandwidth limited by the biasing inductors and DC blocking capacitors, while shunt diodes feature high isolation relatively independent of frequency (see Figure 1). Circuit designers often use transmission lines to create series lumped inductance to achieve a low pass filter effect, which enables the switch to work up to the desired frequency. Shunt diode switches have limited bandwidth arising from the use of quarter wavelength transmission lines between the common junction and each shunt diode. A combination of both shunt and series diodes are typically used to achieve optimal insertion loss and isolation performance in a diode switch, but there is a trade off between them. As seen in these various configurations, PIN diodes can require larger circuit areas to realize because of the passive components and multiple diodes needed for the switch design compared to IC switches. The larger footprint is an issue in compact and mobile designs.

Some examples of high power PIN diode products come from Aeroflex/Metelics, a company that offers surface-mount PIN diodes with 100 W CW power handling and 650 W pulsed power handling with insertion loss less than 0.2 dB and isolation of 53 dB. Skyworks offers a series of QFN packaged PIN diodes for high power applications of 50 W CW power handling and 500 W pulsed that operate to 6 GHz with better than 0.45 dB insertion loss and isolation of better than 37 dB. M/A-COM has a special KILOVOLT™ series of PIN diodes (ceramic packaged) that can handle multi-kilo watts of pulsed power for very high power applications in addition to many other offerings. These three companies, along with several others, have been manufacturing PIN diodes for many years, so they are well characterized and have proven reliability in many demanding applications. They are available in many form factors such as chip, beamlead, ceramic packaged and surface-mount packaged in addition to chip scale form factors.

PIN diode switches are not just offered as discrete devices; many manufacturers offer integrated diode MMICs. M/A-COM Tech pioneered the diode MMIC in the late 1980s with the Glass Microwave Integrated Circuit (GMIC), which used a glass process to isolate the devices by fusing the GaAs with a glass wafer.² This was the predecessor to the Heterolithic Microwave Integrated Circuit (HMIC) process, which is glass and Si fused together. M/A-COM Tech has recently developed multi-octave, high power switches with its AlGaAs PIN diode HMIC devices including multi-throw switches with power handling of 50 W CW (over 100 W pulsed) with less than 2 dB insertion loss and over 30 dB isolation at 40 GHz. This family of switches utilizes a patented AlGaAs heterojunction anode to reduce insertion loss and increase IP3 without compromising isolation; off-state capacitance remains unchanged. This family of AlGaAs/GaAs PIN-based switches probably has the highest frequency, broadest bandwidth response of any of the other technology currently available as these switches are fully functional from 1 MHz to 75 GHz. In addition, switching speeds of less than 500 picoseconds have been measured with this technology. Figure 2 shows a SP8T broadband AlGaAs HMIC PIN diode and its performance.

Another example comes from TriQuint, who offers a vertical PIN process as a foundry service. An example device is a GaAs monolithic PIN diode SP4T switch that operates from DC to 20 GHz. At a bias current of 10 mA per output arm, typical mid-band performance is 0.6 dB insertion loss with 40 dB isolation in the off-arms. Isolation and insertion loss can be adjusted by varying the output arm bias current of the switch.

The Current Workhorse
GaAs field-effect transistor (FET)-based switches have been the mainstay of RF/microwave switches since the 1980s when MMIC circuits became widely available at relatively low prices. Driven by DARPA funding for defense applications (MIMIC program) and a high demand for commercial wireless devices, GaAs MMIC reproducibility improved dramatically during this period and device costs were greatly reduced.³ FET switches are very stable and repeatable due to good control of the drain-to-source resistance. FET switches are voltage-controlled resistors so they provide low power operation, small size and relative design simplicity compared to PIN diodes. They are broadband (DC to 20 GHz devices are widely available) and have relatively high linearity.

Initially, GaAs MMIC MESFETs were widely used in the 1980s and ‘90s, but these have given way to PHEMT MMIC devices, which have better on resistance (Ron) characteristics and are now the most widely used GaAs MMIC switching devices. While MESFET devices were able to reach switching speeds down to tens of pico seconds, PHEMT devices suffer from gate lag as electrons can be trapped on the surface. PHEMTs typically have switching times in the hundreds of micro seconds as they can switch in tens of nano seconds (10 to 90 percent), but have gate lag times of several hundred microseconds (90 to 98 percent).

However, new developments such as M/A-COM’s nanosecond designs achieve about ten nano second switching times including gate lag. M/A-COM made a number of changes to the PHEMT process and device structure to overcome this problem.⁴ The number of surface states and interface traps were reduced at the ungated GaAs surface using cleaning techniques and the deposition of a special passivating dielectric. The formation of the Schottky diode gate was modified to both reduce gate resistance with no additional gate capacitance in order to minimize the RC charging time associated with device turn-on and turn-off. And a proprietary III-V layer was added to the PHEMT structure to further reduce the channel resistance and enable enhanced movement of charge through the device especially from the ungated recess region. This process optimization for low gate lag not only resulted in an improvement in the 90 to 98 percent switch settling time, but also exhibited reduction in the 10 to 90 percent switching speed. While CMOS Si-based switches do not suffer from gate lag, they typically switch in the range of micro seconds. This is because CMOS SOI devices are typically designed for low-frequency operation. Similarly, SOS switches can achieve nanosecond switching speeds without gate lag; however, most devices have been designed to optimize the tradeoff between speed and low frequency operation. One of the drawbacks of MEMS switches is switching speed as they typically exhibit speeds of tens to hundreds of micro seconds for electrostatically operated devices.

Low channel resistance allows GaAs MMIC switches to operate at low frequencies (very near DC) and reverse biasing completely depletes the channel in the OFF state providing excellent isolation at low frequencies.1 However, to operate down near DC, GaAs switches do need a charge pump circuit because the traditional capacitors prevent low frequency operation. The isolation degrades at higher frequencies due the effect of the drain-to-source capacitance. Figure 3 shows a simplified schematic diagram of a SPDT FET switch. The biasing path is not connected to the RF path simplifying the DC biasing path and eliminating the expensive RF choke. The chokes are used to reduce the insertion loss that results from the biasing path being connected to the RF port in PIN diode switches. The ON resistance of a GaAs FET is still typically higher than a PIN diode, so the insertion loss performance of FET switches is not as good as PIN switches. FET switches are voltage controlled so they consume far less current than current controlled PIN switches.

While GaAs MMIC switches were originally only available as depletion mode devices, they required a negative control voltage for operation that was not desirable compared to CMOS switches. An alternative to using a negative control voltage is to elevate or float the DC voltage at the source of the FET to +5 V and use a 0 to +5 V control voltage. Floating the DC voltage requires blocking capacitors that complicates the design and requires more circuitry. However, there are now a number of suppliers offering enhancement mode PHEMT (E-PHEMT) devices that do not require a negative gate voltage to operate (these devices are typically only offered in PA MMICs). They are normally OFF and use a positive voltage to turn the FETs on. This also allows integration of limited logic on the same chip that has always been an advantage with Si-based FET switches, but it is at the cost of current consumption. Many companies and foundries also offer E/D-PHEMT processes that can incorporate both FET modes on a MMIC so that each device type within the circuit can use the process that best fits its needs for performance. Companies such as TriQuint, RFMD, Skyworks, Hittite and M/A-COM Tech, along with others, use this technology in their switches for appropriate applications where the increased complication in processing is worth the added benefits. Therefore, the disadvantages for negative voltage operation and logic integration for GaAs MMICs has been decreased over the past few years, although Si CMOS still offers better integration opportunities with logic and memory circuits.

The GaAs MMIC switch market is still very large and they are used in many applications from commercial to military. Many circuit designs and switch types are available and optimized for almost any application. The design demands for compact, low current multi-throw handset devices are far different from high power base station or military radar applications. Most of the widely known component manufacturers such as Skyworks, RFMD, TriQuint, Hittite, CEL/NEC, Mini-Circuits and M/A-COM Tech offer a wide variety of devices depending on the application. GaAs MMIC devices are still progressing as they shrink die sizes, develop chip scale packaging and optimize the FET design, but the technology is relatively mature so the improvements are not revolutionary at this point. Over the last 20 years or so, GaAs PHEMT switches have offered the best overall performance for most high frequency (over a few GHz) and broadband applications that require low to medium power levels, but this is changing.

Gaining Ground
Standard Si CMOS-based FET switches have previously not proven to be good RF switches as they suffered from significant insertion loss and low isolation because the substrate is not insulating and breakdown voltages are low. One way to overcome the low breakdown voltage is to stack the FETs, but it is difficult to accomplish spreading the voltage evenly across the FETs so this has not worked very well in standard CMOS. However, Si-on-sapphire (SOS) and more recently Si-on-insulator (SOI) FET switches have been gaining market share in many applications as their insulating substrate quality, cut-off frequencies and breakdown voltages have improved. They are competing with GaAs switches in some high volume applications such as handset switches and even demanding military applications.

These Si technologies accomplish FET stacking with even voltage distribution across the FETs, low insertion loss, high isolation and better linearity than standard CMOS. An example of higher voltage operation is Peregrine Semiconductor SOS switches that have recently achieved 50 W CW power handling and greater than 80 dBm IIP3 on switch designs that are much higher than GaAs MMIC switches and similar to PIN diode MMICs (see Figure 4). This is allowing them to penetrate even the high power territory where PIN diodes have dominated. Peregrine maintains that SOS provides the highest linearity due to its fully insulating substrate and GaAs linearity has limitations due to the diode junction formed between the gates in the channel.

Peregrine Semiconductor and IBM recently teamed up to develop and manufacture future generations of Peregrine’s patented UltraCMOS™ silicon-on-sapphire (SOS) process technology, which is unique for its thin insulating layer. It provides the needed isolation, but is thin enough so that it minimizes the negative effects of a thicker Si layer that does not provide ideal high resistivity (see Figure 5). When fully qualified, the next-generation UltraCMOS RF ICs will be manufactured by IBM for Peregrine in the jointly-developed 180-nanometer RF CMOS process at IBM’s 200 mm semiconductor manufacturing facility in Burlington, VT. This development marks the first commercial use of 200 mm (8-inch) wafer processing for a silicon-on-sapphire process. An example of a high performance switch recently developed by Peregrine is a monolithic symmetric SP8T switch (manufactured on its STeP5 process) that covers from near DC to 4 GHz with IIP3 of +70 dBm, IIP2 of +130 dBm and insertion loss of 0.35 dB (900 MHz). The switch handles +35 dBm operating input power (across the range) with high ESD tolerance of up to 4 kV (HBM).

While GaAs MMIC switches offer good linearity and isolation with low ON resistance and low OFF capacitance (Coff), they do have some disadvantages. GaAs technology is relatively mature and while it still is improving, most major advances have probably been achieved. As a representative from Peregrine said, “There are not many more dials to turn to improve GaAs device performance.” SOS and SOI CMOS-based devices are improving quickly and are now closing in on a lower Ron·Coff product, a good figure of merit for switches, allowing the design of switches with lower insertion loss and higher isolation. Peregrine believes that SOS could achieve a product of less than 200 fs as they progress to 0.18 micron technology and even lower with 0.13 micron technology. Today some high performance GaAs devices are already at this level.

Up until recently, SOI resistivity was not as ideal as GaAs or SOS, so devices made with SOI technology exhibited higher levels of harmonic and intermodulation distortion. However, recent advancements in SOI CMOS technology have been able to reduce these effects to make them competitive with SOS.⁵ RFMD has demonstrated a SP9T SOI switch with similar performance to current PHEMT switches with a Ron·Coff product of 250 fs, which is close to a high quality PHEMT value of 224 fs. This compares favorably to current SOS products of 400 fs with 0.25 micron technology, but with 0.18 micron technology SOS is expected to be below 200 fs.

RFMD has three new SOI switches that are being released. This lineup has been designed into two major handsets, so it seems to be gaining momentum. Earlier this year, Skyworks introduced a symmetrical SP4T SOI switch. The device is designed for 3GPP bands from 0.70 to 2.7 GHz with typical insertion loss as low as 0.6 dB and isolation as high as 30 dB and harmonic performance less than 75 dBc at 0.9 GHz. At the same time, the company also introduced a W-CDMA DP4T SOI switch with a decoder. Today’s best GaAs PHEMT switches, 0.18 micron SOS and 0.18 SOI appear to be nearly equal in this figure of merit measurement. Table 1 compares the Ron·Coff figure of merit for various technologies. It should be noted that this figure of merit is effective for comparing raw device performance, but is not the only metric important for a complete switch design. Therefore, processes with lower Ron·Coff values can outperform ones with lower values when other tradeoffs are made in a complete design.

GaAs MMIC devices can have higher contact resistance than these Si technologies, increasing losses, and cannot integrate logic circuits as well as Si-based technologies. Driving GaAs switches also frequently requires extra interface components, and GaAs has limited capability to integrate other functions such as logic control and memory. While CMOS switches have greater than 2000 V (HBM) ESD tolerance, which is relatively robust, most GaAs MMIC switches are only around 200 V, making them susceptible to ESD damage and typically requiring special handling procedures. The same is true with most RF MEMS switches.

Cost is a major advantage of any type of Si wafer processing for large volume applications compared to GaAs because Si has lower material costs and larger wafers to reduce the cost per unit area. While in the past SOS substrates were very expensive, this is changing as the LED market is fueling demand for lower cost substrates and driving high volumes. According to Peregrine Semiconductor, government funding for the LED market along with the demand for low energy lighting could make sapphire substrates the highest volume electronic devices in the near future (see Figure 6). LEDs have been running on 150 mm substrates, putting them on the same wafer sizes as high volume GaAs.

However, neither SOS nor GaAs wafer processing can probably come close to the low cost of SOI switches, which use standard Si processing and larger wafer sizes, although Peregrine maintains that its SOS process uses fewer masking steps and standard processes that can match SOI costs. As SOI performance improves to match GaAs and SOS switches, they will probably have a cost advantage for high volume applications. One way that GaAs devices compete on cost is their setup costs (lower cost mask sets) are typically much less for wafer runs than Si, so for lower volumes they can gain a cost advantage especially for IDMs that produce their own devices.³

While SOI technology has not been competitive in the past, it seems poised to compete in the lower frequency (under 3 GHz), high volume market such as handset and perhaps even WLAN markets. This is mostly enabled by the recent availability of low cost, high resistivity Si substrates (1 kohm-cm), which were not available several years ago. As one representative from RFMD put it, “While SOI was thought to have poor linearity, we are finding that through careful switch branch layout, charge pump optimization and an excellent collaboration with our foundry partners, we are now meeting or exceeding the best of SOS reported performance, and SOI substrate costs are a fraction of sapphire at 200 mm, not counting the fact that SOI uses standard technologies and libraries.” Peregrine conversely states that its SOS technology uses standard CMOS processes and libraries. The company also maintains that its Step5 process leads the market in all performance parameters and provides better design flexibility.

Starting to Compete
RF MEMS capacitive switches were first developed and used in the early 1990s and typically use an electrostatic means to actuate the switch. They offer very low loss and high linearity compared to FET switches, but their switching speed is typically much lower. There are basically two types, the ohmic contact and capacitive contact. With ohmic switches, two metal electrodes are brought together to create a low resistance contact, while in capacitive switches, a metal membrane is pulled down onto a dielectric layer to form a capacitive contact. Figure 7 shows the construction used by Omron for its MEMS switches. The electrode is a special metal composition that flexes down when voltage is applied to turn the switch on and returns back to its original position without the applied voltage. The use of capacitive coupling has reduced issues associated with older generation MEMS switches of dry contact, metal to metal ohmic switching. Issues with sticking contacts, wear, etc., have been mitigated using this newer technology as suppliers have optimized the metallic materials and design.

RF MEMS switches can have no compression point until +36 dBm, as shown in the comparison to GaAs switches in Figure 8. Omron has designed a SPDT switch that operates at 34 V with typical insertion loss of 1 dB, isolation of 30 dB and return loss of 10 dB at 10 GHz. Radant manufactures some high isolation, low loss MEMS switches such as a SP6T DC to 20 GHz device with 22 dB isolation and less than 0.8 dB loss at 18 GHz and near zero harmonic distortion. They also have a very high isolation DC-12 GHz MEMS switch with better than 70 dB isolation and less than 0.3 dB insertion loss at 2 GHz.

Over the past decade, processing improvements, materials refinements and design changes have enabled designs with less than 0.1 dB loss through 40 GHz, low power consumption of tens of nanojoules per cycle and high linearity of greater than 66 dB, according to Memtronics. Reliability is on the order of 100 million cycles, minimum. In fact, Radant Technology reports that its devices have been independently tested by the Department of Defense (DoD) laboratories under a DARPA program to 1.5 trillion cycles, which was conducted in 30 months of continuous testing. This has allowed them to compete in several applications such as test & measurement and switching arrays for antennas. The advantage of the mechanical switch is that when it is off, it is physically isolated so there is little leakage. Leakage current is about 100 fA at 100 VDC. MEMS offer lower off-state capacitance and better off-state RF isolation than either FETs or PIN diodes. Like GaAs FETs, they have low ESD tolerance of around 100 V HBM so they require special handling.

iSuppli recently reported that it anticipates RF MEMS revenue to rise to $8.1 M this year, $27.9 M in 2011 and then $223.2 M in 2014. Much of this is projected to be from cell phone front-end adoption of tuning using RF MEMS switches and varactors to help boost the performance of smart phones. iSuppli states that WiSpry and TDK-Epcos are offering RF MEMS for high volume cell phone applications, while Analog Devices, Radant Technologies and XCOM Wireless (in cooperation with Teledyne Technologies) as well as Omron are targeting high end applications for testing and instrumentation such as ATC and RF test. Also noted at EuMW 2010 this year was DelfMEMS, who is also manufacturing and supplying MEMS switches.

An example of the new MEMS tuning technology is the TDK-EPC and WiSpry tunable modules that quickly switch in and out various values of capacitance to dynamically tune handset antenna to maximize efficiency. Startups Radant and MEMtronics are also focusing on defense applications. Outside of cell phone and instrumentation, wireless infrastructure switches could be replaced by cheaper, higher performance RF MEMS devices. Another opportunity is in defense applications for radio systems and phased-array antennas.

Future Technologies
While most of the recent advances in switch technology have concentrated on the lower power applications driven mostly by handsets, the higher power applications are still dominated by PIN diode technology. GaN has been developed mostly for high power amplifier applications, but it also shows great promise as a future switch technology. A couple of GaN suppliers have started to release switch products. TriQuint has developed three broadband GaN on SiC MMIC switches to cover frequency ranges of DC-6, DC-12 and DC-18 GHz, as shown in Figure 9. These devices have maximum insertion loss of 0.7, 1.0 and 1.5 dB, and demonstrate 40, 15 and 10 W RF power handling, respectively, for 6, 12 and 18 GHz designs.⁶ Cree has advanced datasheet information available on a 25 W, 0.1 to 3 GHz SPDT GaN MMIC switch. It features less than 0.7 dB insertion loss, 15 ns switching speed, over 30 dB isolation and over 60 dBm TOI. With the need for high power switches with lower current consumption, GaN switches should eventually find their way into several applications, especially satellite and military designs.

There also has been some work done with microwave nano electromechanical systems (NEMS) switches that potentially could overcome the drawbacks of the current MEMS devices. Work has been done that shows graphene flakes that can operate as a switch up to 60 GHz with switching times of less than a nano second.⁷ This could enable all the benefits of MEMS switches while obtaining fast switching times comparable to the fastest solid-state switches. The simplified construction is shown in Figure 10. The device is a coplanar waveguide and an array of metallic graphene sheets suspended over it. The waveguide is made from gold strips deposited on a 500 micron thick semi-insulating Si substrate. The graphene flakes are suspended over the waveguide due to van der Waals forces, but could be attached via metallic contacts.

The State of Switches
Table 2 shows a summary of the key performance metrics for the various switch technologies covered. The “state” of RF and microwave switch technology today shows that the PIN diode is still very viable in high power, high frequency applications, but the most widely used technology is the GaAs MMIC, which still offers the best performance for most high frequency (over a few GHz) and broadband applications that require low to medium power levels. However, the market is changing as SOS and SOI CMOS switches are making significant in-roads in some high volume applications. Their performance is matching that of GaAs MMICs at frequencies up to a few GHz and they offer cost and integration advantages. SOI and SOS switches are also proving they can be viable in medium to high power applications as breakdown voltages have improved with FET stacking. In the future, GaN MMIC switches show great promise to take a foothold in higher power applications, probably replacing PIN diodes in some of these areas. MEMS switches are showing promise in the test & measurement, phased array and tunable module market, which promises to be significant in the large handset market.

References

1. Agilent Application Note, “Understanding RF/Microwave Solid State Switches and Their Applications,” No. 5989-7618EN, May 2010.
2. D. Gotch, “A Review of Technological Advances in Solid-state Switches,” Microwave Journal, Vol. 50, No. 11, November 2007, pp. 24-34.
3. P. Hindle, “2010 GaAs Foundry Services Outlook,” Microwave Journal, Vol. 53, No. 6, June 2010, pp. 20-28, 112.
4. T. Boles and A. Freeston, “New NanoSecond Switch Technology,” Microwave Journal, Vol. 53, No. 6, June 2010, pp. 56-60.
5. A. Tombak, C. Iversen, J.B. Pierres, D. Kerr, M. Carroll, P. Mason, E. Spears and T. Gillenwater, “Cellular Antenna Switches for Multimode Applications Based on a Silicon-on-Insulator Technology,” RFIC 2010 Conference, Anaheim, CA.
6. C.F. Campbell and D.C. Dumka, “Wideband High Power GaN on SiC SPDT Switch MMICs,” IMS 2010 Conference, Anaheim, CA.
7. M. Dragoman, D. Dragoman, F. Coccetti, R. Plana and A.A. Muller, “Microwave Switches Based on Graphene,” J. Appl. Phys., 105, 054309 (2009).