RF Device And Process Biz Heats Up

The RF device and process technology markets are heating up, especially for two critical components used in smartphones—RF switch devices and antenna tuners.

RF device makers and their foundry partners continue to ramp up traditional RF switch chips and tuners based on RF SOI process technologies for today’s 4G wireless networks. And recently, GlobalFoundries introduced a 45nm RF SOI process for future 5G networks. RF SOI, the RF version of silicon-on-insulator (SOI) technology, makes use of high-resistivity substrates with built-in isolation.

Then, seeking to disrupt the landscape, Cavendish Kinetics, a fabless IC design house, is ramping up a new and competitive generation of RF switches and antenna tuners based on an alternative technology, RF MEMS.

RF switches and tuners are among the critical components inside a phone’s RF front-end module. The RF front-end integrates the transmit/receive functions in a system, and the RF switch routes the signals. Tuners help the antenna to adjust to any frequency band.

Regardless of the device and technology type, the challenges are daunting in the RF market today. “Several years ago, RF was a fairly simple thing to design,” said Paul Dal Santo, president and chief executive of Cavendish Kinetics. “But things have drastically changed now. First, you have an incredibly broad frequency range that the RF front-end has to handle, extending from 600 MHz to 3 GHz. With advanced signaling technologies and 5G coming up, it will be 5 GHz to 60 GHz. That puts some incredible challenges on front-end RF designers.”

With those challenges in mind, cellphone OEMs have to weigh tradeoffs for some new component choices. Specifically, for RF switches and antenna tuners it boils down to two technologies—RF SOI-based devices and RF MEMS.

RF SOI is the incumbent technology. Devices based on RF SOI are capable, but they are beginning to encounter some technical issues. On top of that there are price pressures in the market, and there are issues as the devices migrate from 200mm to 300mm fabs.

In comparison, RF MEMS have some interesting properties and are making inroads in select areas. In fact, Cavendish Kinetics says that its RF MEMS-based antenna tuners are being incorporated by Samsung and other OEMs.

“RF contact MEMS provide very low on-resistance, and hence low insertion loss,” said Chris Taylor, an analyst with Strategy Analytics. “But RF MEMS also do not have a production track record, and high-volume wireless system OEMs will not commit in a big way to a new technology and small suppliers. Of course, RF MEMS have to come in at competitive price points to alternatives, but mainly OEMs want proven product reliability and reliable supply sources.”

RF front-end
Still, the segment is worth watching amid a mixed business environment for smartphones, the big market for RF switches, tuners and other components. In total, smartphone shipments are expected to grow a mere 1% in 2017, compared to 1.3% growth in 2016, according to Pacific Crest Securities.

On the other hand, the RF front-end module/component market for cellphones is projected to jump from $10.1 billion in 2016 to $22.7 billion by 2022, according to Yole Développement. In total, the RF switch device market was a $1.7 billion business in 2016, according to Strategy Analytics.

The RF market is growing, as OEMs continue to add more RF content in smartphones. “Multi-band LTE is also working its way to lower-tier devices,” Strategy Analytics’ Taylor said. “Switch content is growing.”

The number of RF switch devices per phone jumped amid the shift to 4G, or long-term evolution (LTE). “We are talking about a huge number of units per year,” Taylor said. “Most, but not all (RF switches), go into cellphones, and the vast majority of these are now SOI. RF MEMS is still a nascent market and is tiny relative to RF SOI switches.”

Despite the shipment numbers for RF switches, there is stiff competition and price pressure in the market. The average selling prices (ASPs) for these devices are between 10 to 20 cents, he said.

Meanwhile, in a simple system, the RF front-end consists of several components—power amplifier; low-noise amplifier (LNA); filter; and RF switch.

Fig. 1: A simple front end module. Source: Globalfoundries, “Designing Next-Gen Cellular and Wi-Fi Switches Using RF SOI,” Technology, May 2016.

“The main purpose (of the power amp) is to make sure there is enough power for your signal or message to reach its destination,” said Randy Wolf, a member of the technical staff at GlobalFoundries, in a recent presentation.

An LNA amplifies a small signal from the antenna. An RF switch routes signals from one component to another. “(A filter) prevents any unwanted signals from coming in,” Wolf said.

In a cellphone, the RF functions were simple in 2G and 3G wireless networks. There were four frequency bands in 2G, and five for 3G. But for 4G, there are more than 40 frequency bands. 4G not only incorporates the 2G and 3G bands, but also a number of 4G bands.

On top of that, mobile operators have deployed a technology called carrier aggregation. Carrier aggregation combines several channels, or component carriers, into one big data pipe, enabling more bandwidth and faster data rates in a wireless network.

To deal with the frequency bands and carrier aggregation, OEMs require complex RF front-end modules. Today’s modules might integrate two or more multi-mode, multi-band power amplifiers, as well as several switches and filters. “It depends on the RF architecture that is adopted. The number of PAs are driven by the regional frequency bands that are being addressed by the handset. In a typical ‘global’ phone, which addresses multiple/all-cellular markets globally through a single SKU, the frequency bands coverage is comprehensive,” said Abhiroop Dutta, strategic marketing manager for mobile at Qorvo. “In a typical integrated module RF front-end implementation for such a phone, one option is to use an RF front-end with split band modules that address high, mid and low frequency bands.”

In contrast, smartphone OEMs may design regional phones for specific markets. “An example is handsets that are targeted at the China domestic market. In this case, the RF front-end needs to support bands specific to the region,” Dutta said.

Fig. 2: A 4G front end. Source: GlobalFoundries, “Designing Next-Gen Cellular and Wi-Fi Switches Using RF SOI,” Technology, May 2016.

LTE phones also have two antennas, main and diversity, according to Cavendish Kinetics. The main antenna is used for transmit/receive functions. Basically, a diversity antenna boosts the downlink data rates in phones.

In operation, a signal reaches the main antenna. It then moves to an antenna tuner, which allows the system to adjust to any frequency band.

Then, the signal moves into a series of RF switches. “It switches into the appropriate bands you want to use as far as GSM, 3G or 4G,” GlobalFoundries’ Wolf said. From there, the signal moves into a filter, followed by the power amp. Finally, it reaches the receiver.

With this complexity in mind, cellphone OEMs face some challenges. Power consumption and size are critical. “Because of this complexity, you are getting more loss in the front-end, which is having a negative impact on your LNA total noise figure of your receiver,” Wolf said.

Clearly, the RF switch plays a key role to help address the problem. In total, a smartphone might incorporate more than 10 RF switch devices. A basic RF switch comes in a single pole, single throw (SPST) configuration. This is a simple on-off switch.

Today, OEMs use more complex switch configurations. Ron*Coff is a key metric for RF switches. “Ron*Coff is a ratio of how much loss occurs when a radio signal goes through a switch in its ‘on’ state (Ron, or on-resistance) and how much the radio signal leaks through the capacitor in its ‘off’ state (Coff, or off capacitance),” according to Peregrine Semiconductor.

All told, OEMs want RF switches with no insertion loss and good isolation. Insertion loss involves the loss of signal power. If the switch doesn’t have good isolation, the system could encounter interference. “Overall, the challenge for RF front ends is to support the increasing performance needs, aligned to evolving standards and increasing frequency band coverage. All of this while looking to reduce the size of the solution given the shrinking RF footprint, as phones get thinner. The key metrics like insertion loss, power at the antenna, and isolation are still drivers of the solutions within our RF portfolio,” Qorvo’s Dutta said.

The solutions
Today, for the power amplifier in cellphones, mobile OEMs mostly use gallium arsenide (GaAs) technology. Then, several years ago, OEMs migrated from GaAs and silicon-on-sapphire (SoS) to RF SOI for RF switches. GaAs and SoS, a variant of SOI, simply became too expensive.

RF SOI is different than fully-depleted SOI (FD-SOI), which is geared for digital applications. Like FD-SOI, RF SOI has a thin insulating layer in the substrate, enabling high breakdown voltages and low leakage for RF.

“The mobile market continues to favor RF SOI, as it delivers low insertion loss and harmonics and high linearity over a wide frequency range, good performance, and cost-effectiveness,” said Peter Rabbeni, director of the RF business unit at GlobalFoundries.

Today, Qorvo, Peregrine, Skyworks and others provide RF switches based on RF SOI. Typically, RF switch makers use foundries to make these products. GlobalFoundries, STMicroelectronics, TowerJazz and UMC are among the players in the RF SOI foundry business.

So, OEMs have several choices in terms of component vendors and foundry offerings. Generally, the foundries offer RF SOI processes anywhere from the 180nm to 45nm nodes and at different wafer sizes.

The decision to go with one node or another depends on the application. “With regards to technology scaling in RF SOI, everything is about fitting technology solutions to the appropriate end application from a technology performance, cost and power standpoint,” said Walter Ng, vice president of business management at UMC.

Even with the choices, RF switch makers face some challenges. The RF switch itself incorporates a field-effect transistor (FET). As with most devices, FETs suffer from unwanted channel resistance and capacitance.

In an RF switch, the FETs are stacked. Typically, some 10 to 14 FETs are stacked in today’s RF switches. As the number of stacked FETs increases, the device may encounter insertion loss and resistance, according to experts.

Another issue is capacitance. In an RF switch, 30% or more of the unwanted capacitance is attributed to the interconnects in the device, according to Skyworks in a paper in 2014, entitled: “Recent Advances and Future Trends in SOI for RF Applications.” The interconnects are the metal layers or tiny wiring schemes in chips, including RF SOI-based switches.

Generally, in 4G phones, the mainstream processes for RF switches are the 180nm and 130nm nodes in 200mm fabs. Many but not all of the interconnect layers are based on aluminum. Used in the IC industry for years, aluminum interconnects are inexpensive, but they also have a higher capacitance.

As a result, copper is used for select layers in RF devices. Copper is a better conductor and is less resistive than aluminum. “The traditional metal stack for 130nm RF CMOS process offerings involves a combination of aluminum interconnect layers for cost, and copper interconnect layers for performance,” Ng said. “In this way, it is a more optimal solution to balance cost and performance. RF SOI solutions are typically a certain number of aluminum metal layers and one or more copper layers.”

Typically, copper is used on the top layers as ultra-thick metal options for improving passive device performance. “A thick top layer metal, preferably copper, improves performance by minimizing ohmic losses,” he said.

More recently, RF device makers have been migrating from 200mm to 300mm fabs, where the processes range from 130nm to 45nm. Generally, the 300mm foundry fabs process wafers using only copper interconnects.

By using only copper interconnects, switch makers can lower the capacitance. But 300mm also translates to higher wafer costs, creating some friction in the market. On one hand, OEMs demand lower prices in the cost-sensitive cellphone market. On the other hand, device makers and foundries want to maintain their margins.

“Today, only a very small percentage of RF SOI is being produced on 300mm,” Ng said. “There are many reasons for it, including cost/availability of 300mm RF SOI substrates, infrastructure support post silicon processing and other factors. However, we expect that these challenges will be addressed to a large part over the next couple of years, and then most of the higher volume RF SOI applications will migrate.”

Until then, the industry may face some supply/demand issues on 300mm. “We believe that demand will continue to challenge supply until more of the production migrates to 300mm. Then, it will be a question of how quickly that capacity comes on line and how well it will match demand at the time,” he said.

Generally, today’s RF SOI processes are geared for 4G phones. Seeking to get a jump in the 5G arena, GlobalFoundries recently introduced a 45nm RF SOI process for 5G applications. The process makes use of high-resistivity trap-rich SOI substrates.

5G is the follow-on to 4G. Today’s LTE networks operate from 700 MHz to 3.5 GHz. In comparison, 5G will not only co-exist with LTE, but will also operate in the millimeter-wave bands between 30 GHz and 300 GHz. 5G will enable data transmission rates of more than 10Gbps, or 100 times the throughput of LTE. But the mass deployment of 5G isn’t expected to occur until 2020 and beyond.

Regardless, 5G will require a new class of components. “(45nm RF SOI) is focused primarily on enabling 5G millimeter wave front-ends, which integrate the PA, LNA, switch, phase shifters to create an integrated mmwave steerable beamformer for 5G systems,” GlobalFoundries’ Rabbeni said.

For 5G, there are other solutions as well. RF MEMS is one possibility. In another possible solution, TowerJazz and the University of California at San Diego recently demonstrated a 12Gbps, 5G phased-array chipset. The chipset uses TowerJazz’s SiGe BiCMOS technology.

The winner? Time will tell. “It isn’t clear whether RF MEMS will have advantages for 5G,” Strategy Analytics’ Taylor said. “Monolithic integration could win the day for SOI at least above 6-GHz.”

What is RF MEMS?
RF SOI-based switches will continue to dominate the landscape, but there might be room for a new technology—RF MEMS. “SOI has made incredible improvements over time. The resistances have dropped and the linearity has become better,” Cavendish Kinetics’ Dal Santo said. “But an SOI switch is really just a transistor turned on or off. It’s not very good when it’s on and it’s not very good when it’s off.”

RF MEMS has been in the works for years. Today, Cavendish, Menlo Micro and WiSpry (AAC Technologies) are developing RF MEMS for mobile apps.

RF MEMS are different than sensor-based , such as gyros and accelerometers. Sensor MEMS are transducers that turn mechanical energy into an electrical signal. In contrast, RF MEMS conducts the signal.

Initially, Cavendish and others have targeted RF MEMS for the antenna tuner market, where RF SOI-based switches and other technologies are used.

Fig. 3: Antenna tuner with switch. Source: Cavendish Kinetics

“You have antennas. If they are fixed, they can’t possibility make the range required for the frequency bands that need to be supported. So they need to be tuned,” Dal Santo said. “Right now, the predominate method is to take a switch and either switch in different capacitors, fixed capacitors or different fixed inductors. The problem with that is that antennas are high-Q devices. You have to be careful that you don’t load them or you will see radiated performance losses.”

In comparison, Cavendish’s tuners have 32 different capacitance ranges. “They are fully programmable and have a very high-Q performance. So they are very low loss. You can use these to tune your antenna to the range of frequencies you need to support,” he said.

Going forward, Cavendish plans to take on RF SOI-based devices in the larger RF switch arena. “If you replace those with a real switch, which is a MEMS switch, then you can see the accumulated insertion loss benefit you get across your receiver or transmitter,” he said.

But will RF MEMS devices displace those based on RF SOI? One company, TowerJazz, can perhaps provide some insights. TowerJazz provides traditional RF SOI processes, and is also the foundry provider for Cavendish’s RF MEMS devices.

“There may be some small overlap where RF MEMS and RF SOI will compete for the same application. In general, they will complement each other (with) RF MEMS winning in the most demanding applications and RF SOI the rest,” said Marco Racanelli, senior vice president and general manager of the RF/High Performance Analog Business unit at TowerJazz.

“RF SOI continues to evolve and thus remains viable and preferred for RF switching applications and some portion of the low-noise amplifier market,” Racanelli said. “However, there are some applications where alternative technologies, such as SiGe for low-noise amplifiers and MEMS for switches, can offer improved linearity or lower losses. And while RF SOI will continue to serve an expanding market, other technologies will have their place as well.”

RF MEMS is making a dent in the antenna tuner market. Time will tell if the technology can crack the switch business. “In the future, RF MEMS can help increase data rates in handsets by providing a more linear and lower loss switch relative to one built in RF SOI,” he said. “This can be understood if one considers that in RF MEMS, metal plates can be made to come in direct contact in the ‘on’ state and form a metal, low-loss, linear connection. Higher levels of linearity allow more bands and more intricate modulation schemes that can increase data rates in handsets.”

Related Stories
RF SOI Foundry Biz Heats Up
Waiting For 5G Technology