Companies are working with different materials and approaches in different regions.
Demand is increasing for power amplifier chips and other RF devices for 5G base stations, setting the stage for a showdown among different companies and technologies.
The power amplifier device is a key component that boosts the RF power signals in base stations. It’s based on two competitive technologies, silicon-based LDMOS or RF gallium nitride (GaN). GaN, a III-V technology, outperforms LDMOS, making it ideal for the high-frequency requirements for 5G. But GaN is expensive with some challenges in the fab. And LDMOS (laterally-diffused metal-oxide semiconductor) has some limitations, but it isn’t going away.
Nonetheless, 5G is a fast-moving but complex market. In just one part of the supply chain, device makers manufacture RF chips like power amplifiers in fabs. From there the devices are shipped to base station vendors for integration. A so-called macro base station is a system located at a cell tower, which provides RF wireless coverage over a wide area.
Generally, the power amplifier device for previous-generation 3G base stations were based on LDMOS. LDMOS, a mature and inexpensive technology, took the early lead in the 4G base station market. Over time, GaN power amps made significant inroads in 4G, at the expense of LDMOS. Power amplifiers are small circuits that convert a low-power RF signal into a higher power signal in base stations and other systems. The power amplifier isn’t the only device in the base station. These other devices are based on various processes.
Nonetheless, GaN-based power amps also are gaining steam in 5G. As in 4G, China’s base station vendors are adopting GaN-based power amp devices for their initial deployments of 5G systems in China. Other base station vendors are following suit.
There are several reasons for that. 5G, a next-generation wireless technology that’s faster than today’s 4G, is being deployed in two different areas – sub-6GHz and mmWave (28GHz and above). Generally, at higher frequencies, LDMOS runs out of steam, prompting the need for GaN. Compared to LDMOS, GaN has higher power densities and operates over a much wider frequency range.
“The need for dense, small-scale antenna arrays in 5G infrastructure is resulting in key challenges around power and thermal management in RF systems. With their improved wideband performance, efficiency and power density, GaN devices offer the potential for more compact solutions that can address these challenges,” said David Haynes, managing director of strategic marketing at Lam Research.
LDMOS isn’t going away, though. Some mobile operators are deploying both low- and high-frequency bands for 5G. LDMOS is suited for the lower bands. So both GaN and LDMOS will find a place in 5G. “In macro stations, GaN has gradually been taking market share from LDMOS following its wide adoption in Huawei’s 4G LTE infrastructure equipment,” said Ezgi Dogmus, an analyst at Yole Développement. “In the sub-6GHz regime in 5G we see tough competition between LDMOS and GaN in lower-power active antenna systems. GaN is being adopted in bands where large bandwidth capacity is needed.”
Regardless, the numbers are staggering. The total GaN RF market will increase from $740 million to more than $2 billion by 2025, with a CAGR of 12%, according to Yole. Telecom infrastructure and military radar are the main drivers for RF GaN. In another example, China built 130,000 5G base stations in 2019, with plans to install 500,000 more in 2020, according to Handel Jones, chief executive of IBS. By 2024, China’s goal is to deploy 6 million systems, Jones said. Japan, Korea, the U.S. and others are also making a big push in 5G.
The numbers don’t tell the entire story. In RF GaN, there are other dynamics, including:
Evolving base stations
Today’s wireless networks revolve around the 4G LTE standard, which operates from the 450MHz to 3.7GHz frequency bands. 4G is fast but complicated. It consists of more than 40 frequency bands, plus the 2G and 3G bands.
The 4G LTE network consists of three parts – a core network, radio access network (RAN), and end-user devices like smartphones. Run by a mobile operator, the core network handles the overall functions in the network.
The RAN consists of giant cell towers, which is where the base stations are located. The RAN is basically a relay system with a multitude of cell towers in a given region.
A base station itself consists of two separate systems, the building baseband unit (BBU) and the remote radio head (RRH). The BBU, which is situated on the ground, handles the RF processing functions. It serves as the interface between the base station and the core network.
The RRH, which is on top of the cell tower, consists of three or so large rectangular boxes. An antenna unit resides on the top of the tower. The RRH handles the conversion of RF signals, while the antenna transmits and receives the signals.
Inside the RRH box, there is a set of chips, which consists of a transmit and receive chain. In simple terms, a digital signal is received in the unit. It is converted to analog, upconverted to an RF frequency, amplified, filtered and then sent out via an antenna, according to “everything RF,” a technology site.
“A relatively high-end LTE base station might have four transmitters. On every tower, there’s going to be four power amplifiers sending signals out to capture and send data to customers,” said Dan McNamara, an analyst at Mobile Experts, a research firm. “On each tower, there’s three of them. Think of it as a pie. Each one handles a certain circle in terms of the way the signals radiate out from the tower. So, there’s actually 12 (transmitters).”
Meanwhile, operators are now deploying 5G. Compared to 4G, 5G promises to deliver mobile network speeds with a 10X lower latency, a 10X higher throughput and a 3X spectrum efficiency improvement. “Mobile communication systems are migrating from 4G to 5G,” explained Sheng-Chi Hsieh, a researcher at ASE, in a recent paper at ECTC. “The new radio (NR) frequency bands are distributed in two defined frequency ranges (FR), which are FR1: 450MHz to 6GHz and FR2: 24.25GHz to 52.6GHz. There are three dimensions to improve the performance, which are massive IoT, low latency, and the enhanced mobile broadband (eMBB), for the usage of massive connectivity, ultra-high reliable and low latency, and capacity enhancement, respectively.”
Each nation has a different 5G strategy. For 5G, China uses 3.5GHz as the frequency. Then, a 5G base station resembles a 4G system, but it’s on a much larger scale. For sub-6GHz in 5G, let’s say you have a macro base station. The power levels at the antenna range from 40 watts, 80 watts or 100 watts.
On the RRH board, you have various devices such as power amps, low-noise amplifiers (LNAs), transceivers and others. The RF process is complex with several steps. “Think of the transceiver is the baseband digital side of things. Coming out of this transceiver, (a signal) goes into the RF. In general, you have some type of receive path. And for us, this is GaAs-based. It can also be silicon-based. It’s basically LNAs and there’s a switch,” explained James Nelson, director of 5G infrastructure accounts at Qorvo. “In this case, a lot of our modules that we make on the receive side are dual channel. That’s why you see effectively two power amplifier sections or transmit sections on the top and on the bottom. They would be identical because this is a dual channel. Where GaN plays is in these amplifier blocks. The amplification can be done a lot of different ways.”
Fig. 1: Evolution of macro base stations and antennas. Source: 5G Americas
5G is different in other ways. Instead of 12 transmit chains, as in 4G, there are 32 or 64 transmit chains in 5G. “The equivalent system in 5G is going to have 32 or 64 power amplifiers in each radio times 3. It’s a huge amount of material that is needed,” Mobile Experts’ McNamara said.
The next step is to integrate some or all of the RRH into the antenna. These integrated base stations make use of massive MIMO antenna systems. Incorporating tiny antennas, massive MIMO communicates with users via beamforming techniques.
In the U.S, meanwhile, 5G is fragmented. Some telecoms are deploying a faster version of 5G using mmWave frequencies at 28GHz. Today, mmWave is limited to fixed-wireless services. It’s a niche market with various challenges. The big 5G deployment in the U.S. will occur when carriers begin to deploy C-band technology at 3.7GHz. The timing of C-band is unclear.
GaN vs LDMOS
Generally, 5G base stations will incorporate GaN-based power amps for the higher frequencies. LDMOS is also in the mix for lower bands.
For years, base stations incorporated power amplifier chips based on LDMOS transistor technology. An LDMOS transistor is a lateral device that resembles a MOSFET. It has a source, gate and a drain.
LDMOS is slightly different than MOSFETs. “The source is connected with a P+ sinker to the backside of the wafer, which makes the backside of the die the source connection of the transistor,” according to Ampleon, a supplier of LDMOS technology. NXP and others also sell LDMOS products.
Based on silicon, LDMOS is processed in 200mm fabs down to 0.14μm geometries. LDMOS transistors are used to develop standard Doherty power amplifier chips for base stations. A Doherty power amp architecture has two amplifier sections, enabling high efficiencies in systems.
LDMOS continues to make improvements, but it arguably hits the wall at frequencies above 2GHz. “Historically, you had GSM in 900MHz, then at 1.8GHz and 2.1GHz. These were the traditional frequency bands, which were LDMOS dominated,” said Gerhard Wolf, vice president and general manager of RF products at Cree’s Wolfspeed unit. “Then, you also had 2.69GHz band 7 and 41 and going higher. This is when GaN came to play. The efficiency of GaN is better at higher frequencies compared to LDMOS. The efficiency of GaN at the 3.5GHz level is better.”
GaN is a wide bandgap technology, which refers to the amount of energy required for an electron to break free from its orbit. GaN has a bandgap of 3.4 eV, while silicon is 1.1 eV.
GaN devices handle more power with better characteristics than other technologies. GaN also enables higher instantaneous bandwidths. This means fewer amplifiers are needed in systems.
But RF GaN is more expensive than LDMOS. Linearity is also an issue with RF GaN. This involves the ability of a power amp to amplify a signal without distortion.
Nonetheless, GaN is used to make high-electron mobility transistors (HEMTs). GaN is the material, while HEMT is the device structure. A GaN HEMT is a lateral device with a source, gate and drain. Current flows from the source to the drain and is controlled by the gate.
Like LDMOS, RF GaN is used to develop power amp chips. For example, in a recent paper, Sumitomo described the development of a GaN-based wideband Doherty amplifier. The two-stage amplifier consists of one GaN transistor for the carrier portion and two transistors for the peaking section. Each transistor has a pair of 180-watt GaN dies.
GaN isn’t new. It can be traced back to the 1970s, when RCA devised a GaN-based LEDs. Two decades ago, the U.S. funded the development of GaN for military/aerospace applications. GaN is also used for CATV amplifiers, LEDs and power semiconductors.
The RF GaN market took off in 2014, when Huawei incorporated GaN-based power amps in its 4G base stations. At the time, LDMOS dominated the landscape, but that soon changed. “For the initial 4G rollout and deployments through the years, LDMOS technology was the main technology and did indeed dominate the market,” said Gavin Smith, RF product launch and global distribution manager at NXP. “Flash forward a few years. GaN technology began to be tested and tried for the next generation of cellular infrastructure as 4G started to ramp down. We saw this shift in technology need and demand, and began to change gears to be ready for 5G deployments with both LDMOS and GaN solutions.”
Meanwhile, Huawei and others have been installing 5G base stations in China. Like 4G, China’s OEMs are embracing GaN-based power amps. Other base station OEMs are following suit.
“LDMOS ran out of steam in the high band of 5G FR1. GaN-on-SiC is now the choice,” said Barry Lin, CTO of Wavetek, a III-V foundry that is part of UMC. “Due to its wide bandgap, high mobility and good thermal conductivity, RF GaN devices have an advantage of wideband applications, which is one of the keys for 5G communications. GaN-on-SiC RF is suitable for 48V Doherty amplifiers to achieve high efficiency, high ruggedness for high-power amplifiers in 5G base stations.”
LDMOS won’t disappear. Some operators in China are deploying low-frequency 5G bands. LDMOS may play a role here.
Then, if or when the industry migrates to a full-blown mmWave 5G network, operators also may deploy a series of small cell base stations. There are several technologies at play for small cells. “GaN-on-silicon RF has been demonstrated to be a very suitable candidate for 28V or 48V small cell power amplifiers,” Lin said. “GaN devices can provide a very wide band, high efficiency, and low noise performance for future MMIC TRX and power amplifier in mmWave bands in 5G FR2 applications.”
Making GaN
The first wave of 5G base stations have been deployed. Now device makers are developing new GaN-based power amp chips, hoping to capture the next wave of 5G base station deployments. Cree, Fujitsu, Mitsubishi, NXP, Qorvo, Sumitomo and others compete in the RF GaN device market. “Moreover, following the US-China trade war, numerous Chinese companies are trying to develop internally GaN RF for 5G infrastructure, while some U.S. companies have lost market share,” said Ahmed Ben Slimane, an analyst at Yole.
At the recent IMS2020 conference, various entities presented papers on what’s next in RF GaN. Among them:
RF GaN continues to improve, but it’s relatively expensive. Boosting the efficiencies is another challenge. And at times, GaN suffers from so-called dynamic on-resistance.
In response, RF GaN vendors are driving down the costs by migrating to larger wafer sizes, improving the process flow in the fab, among other steps.
As stated, a GaN HEMT is a lateral device with a source, gate and drain. The gate length determines the speed of the device, according to Qorvo. A smaller gate translates into a faster device. “The voltage scales with the gate length. As you go to smaller gate geometries, then you can’t swing as much voltage, which then limits your power capability,” Qorvo’s Nelson said.
In RF GaN, the most advanced gate length is 90nm. Vendors are mainly shipping RF GaN chips with gate lengths at 0.15µm to 0.5µm.
Each technology has its place. “0.15µm is one of the state-of-the-art processes. We also have higher frequency processes,” Nelson said. “You wouldn’t use a 0.15µm GaN process for 3.5GHz base stations. You don’t need that type of geometry for the power levels and the frequency. We have a 0.5µm process, which would be 65-volt capable. The radar guys like it. Not everybody is moving to 65 volts. Then, we have another process that’s targeted for 48 volts, which is common for base stations. Then, you have the 0.15µm versions, which can be between 28 and 20 volts.”
Nonetheless, in the fab, the RF GaN process starts with the development of a substrate. The main substrate for RF GAN is SiC (GaN-on-SiC). SiC substrates for RF GaN are based on 100mm wafers with 150mm in the works.
GaN-on-SiC has its pros and cons. It has high thermal conductivities, but the SiC substrates are prone to defects during the production phase. The substrates are expensive.
Others are working on silicon substrates or GaN-on-silicon, which can be produced in 200mm fabs. 200mm enables more dies per wafer, which lowers the manufacturing cost.
“I would conservatively say that 95% of the market is GaN on silicon carbide,” said John Palmour, CTO at Cree/Wolfspeed. “The idea behind GaN-on-silicon is that the substrate is cheap, but the thermal conductivity of silicon is one-third that of silicon carbide. It’s a lot harder to get rid of the heat. To compensate for that, you have to make the devices larger in GaN-on-silicon. You don’t really win on cost.”
Eventually, each technology will have its place. “GaN-on-SiC will focus on the highest power and performance applications, while GaN-on-silicon will address more cost-sensitive applications,” Lam’s Haynes said. “This is because GaN-on-silicon offers the promise of CMOS compatibility, the ability to leverage larger wafer sizes and more advanced wafer fabrication technologies, and the integration of GaN technologies with other solutions in multi-chip modules.”
Regardless of the substrate type, the next step is to grow epitaxial layers on the substrate using a metal organic chemical vapor deposition (MOCVD) system.
First, a buffer layer is grown on the substrate, followed by a channel layer and then a barrier. The channel, which transports electrons from the source to the drain, is based on GaN.
The buffer layer, which prevents electrons from moving into the substrate, is based on a GaN material doped with carbon or iron, according to Qorvo. Based on aluminum-gallium-nitride (AlGaN), the barrier isolates the gate and the channel.
“The top layer is typically a thin AlGaN layer, capping a few micron-thick GaN layer underneath to form a 2D electron gas necessary for a high-speed electrical conduction channel,” said Ronald Arif, senior manager of product marketing at Veeco. “Growing GaN-on-SiC by MOCVD is a mature process. The industry prefers to grow GaN material on a silicon substrate for cost and integration reasons. But this presents a significant challenge in terms of materials quality, uniformity and defectivity.”
Nonetheless, the next step is to form a source and drain electrode on top of the device. Then, a layer of silicon nitride is deposited over the structure.
Forming the gate is the next step. On the device, an etch system etches out a small opening. Metal is deposited in the opening, forming the gate.
The gate etch process works. But at times, the process can cause damage on the bottom and the sidewalls of the GaN surface.
So vendors are exploring the use of atomic layer etch (ALE) for GaN. ALE removes materials at the atomic scale, but it’s a slow process. So ALE might be used in combination with traditional etch processes for GaN.
“It may require a suite of etch processes that addresses the unique challenges of GaN HEMT and MIMIC fabrication,” Lam’s Haynes said. “These include the use of ALE to achieve atomically precise, ultra-low damage and highly selective etching of GaN/AlGaN structures. Using this approach, we have demonstrated a 2X reduction of post-etch GaN sheet resistance, compared to conventional steady-state etch processes and surface roughness that is equivalent to deposited epitaxial films. Such improvements have a direct impact on improved device performance and reliability.”
Finally, the substrate is thinned, and bottom part is metallized. Vias are formed between the top and bottom of the substrate, which reduces the inductance, according to Qorvo.
Conclusion
For years, meanwhile, vendors have been talking about using GaN as the power amp in smartphones. Today’s phones use gallium arsenide (GaAs) processes for the power amp.
GaN is too expensive for smartphones. On the other hand, GaN is gaining traction in several other markets, making it one of many technologies to watch.
Related Stories
 |
 |
 |
 |
 |
 |
 |  |
Excellent Article on 5G challenges,
Very comprehensive and easy to follow technical read. Thank you.