Silicon still wins in sub-6 GHz, but after that GaN looks increasingly attractive.
The global race to launch 5G mmWave frequencies could provide a long-anticipated market opportunity for gallium nitride (GaN) as an alternative to silicon.
GaN is more power-efficient than silicon for 5G RF. In fact, GaN has been the heir apparent to silicon in 5G power amplifiers for years, especially when it comes to mmWave 5G networks. What makes it so attractive is its ability to efficiently handle higher voltage in a much smaller area than comparable laterally diffused MOSFETs (LDMOS) devices. In addition, it can power a much wider range of mmWave frequencies than standard silicon.
“Most of the difference is related to the operating voltage of the transistors, where GaN can be 50V or higher and 28nm CMOS is perhaps 1.8V,” said Keith Benson, director of RF/microwave amplifier and phased array IC products at Analog Devices. “In the end, power is voltage times current, and a higher operating voltage makes high power easier. In regard to comparing GaN versus silicon, in general, that’s a complicated answer as they are very different. GaN is still expensive from a wafer cost, but the mask cost is far less than cutting-edge CMOS.”
Fig. 1: Comparing power and frequency of different materials in the microwave range, which includes mmWaves. Source: Analog Devices
That’s a start, but 5G has a host of problems to contend with as it scales from the sub-6GHz range into millimeter wave technology.
“From a technology standpoint, 5G suffers from attenuation issues, requiring multiple antennae to improve signal quality using spatial multiplexing techniques. Each antenna requires dedicated RF front-end chipset [and power amplification],” said Ajit Paranjpe, CTO at Veeco. “Today, GaN is slowly replacing silicon in specific applications, such as the RF amplifier front-end of 4G/LTE base stations.”
Next-generation designs will open the door to GaN in smaller devices—micro-cells, femto-cells and even smaller access points, although for different reasons than for higher-powered devices.
“For devices with a lower power load the benefit is in the footprint, not just in board space, as well as the layout of the antenna,” Paranjpe said. “That’s where GaN offers the best fit because it operates at higher voltages.”
With the big boxes, however, the issue is how much power is wasted, not only how much is used, according to Earl Lum, president of EJL Wireless Research. “Those [power amplifiers] are only maybe 30% or 35% efficient, so if you put 100 watts into it, you only get to transmit maybe 35 and the other 65 turns into heat.”
At a recent conference in Shanghai, Huawei demonstrated base stations with self-contained liquid-cooling systems. That may seem like overkill, but the density of chips in base stations generates a significant amount of heat. Most other OEMs stuck to traditional methods, but there were a lot of abnormally long aluminum fins and other heat-sink mechanisms on display, Lum said.
Most chipmakers responded by assuming they would be supporting both materials, though a few are pushing hard on one side or the other.
Qorvo, Wolfspeed, NXP, Sumitomo and other chipmakers — especially those with experience in the microwave communications market — have promoted GaN for years as a likely successor to LDMOS in 5G base station power amplifiers (PA) and other applications.
“The data demands driven by 5G and the onset of IoT will require capacity and speeds that, for example, mmWave technology can deliver,” said Gerhard Wolf, vice president and general manager for the RF product line at Wolfspeed. “GaN on silicon carbide is the optimal material for mmWave technology because of its high power density and ability to operate at high frequencies.”
Cree/Wolfspeed is one of the companies making a big bet on the growth of demand for GaN-on-silicon carbide. In fact, in May it announced plans to invest $1 billion to expand its GaN-on-SiC capacity 30-fold using a redesigned, 253,000 square-foot facility currently producing 150mm wafers near its Durham, N.C., headquarters. “This is in direct answer to the demand for this next-generation technology,” Wolf said. “Today, communications infrastructure customers are rapidly pre-investing significantly in 5G ramp-up and we’re proud to be leading the charge on this movement.”
Growth in sales to the defense industry, and to both the 4G and 5G mobile telco market, is forecast to drive sales of GaN components from $380 million during 2017 to $2 billion by 2024, according to a June report from Yole Développement. However, the firm noted that the vast majority of PAs in the civilian wireless market will be silicon.
Sub-6GHz
The same report also predicted “remarkable progress in cost-efficient LDMOS technology,” allowing silicon to continue to challenge GaN in sub-6GHz, active-antenna and massive-MIMO implementations.
That progress almost certainly would include developing standard-silicon components for mmWave networks, as longtime RF systems designer Anokiwave has done—while adding functions to reduce the effort of calibrating phased-array antennas and to reduce power use by as much as a third.
“GaN is fine in isolated uses and a few high-energy applications where it is already popular — LiDAR and radar in particular,” according to Alastair Upton, Anokiwave’s chief strategy officer, who uses his company’s success at building phased-array antenna components and controllers from standard LDMOS as evidence GaN is unnecessary. “We put out our first chips at 28GHz in 2016, and with each generation we’ve gotten orders of magnitude greater efficiency in size and weight. We continue to drive the cost down at a very rapid pace.”
But Upton noted that, at least for the sub-6GHz version of 5G, GaN chips will have to compete with silicon’s economies of scale with only marginal additional benefits.
Good designers can get silicon to do astonishing things, but a large amplifier will dissipate heat more slowly than a small one. So GaN or anything else that can handle the same voltage as silicon in a much smaller space makes the whole process more power-efficient, said Alex Lidow, co-founder and CEO of GaN power-supply provider Efficient Power Conversion (EPC).
“Gallium is a better semiconductor than silicon,” Lidow said. “That’s been well known for quite a while.”
GaN traditionally has been more expensive than silicon, and when compared weight-for-weight, there is still a big difference. But the process of laying down GaN and its components epitaxially on silicon or silicon carbide brings GaN effectively up to par with silicon, and sometimes can cost slightly less, Lidow said.
Analog Devices’ Benson points to a related trend. “The process technology has finally progressed to a point where it’s reliable enough to be fielded into systems,” he said. “It took more than 10 years for the fabs to eliminate many of the issues before it was ready to put into real systems.”
Unavoidable heat issues
Thermal issues are endemic to power amplifiers and RF front-ends due to the huge difference between peak and minimum power requirements, and GaN is particularly good for this.
“You have to supply that same amount of power regardless if you’re transmitting at very high power, or sitting at very low power, which is much of the time, and that extra power is dissipated as heat,” said Andrew Zai, senior principal engineer at Raytheon and chair of the 5G Summit at IEEE’s recent International Microwave Symposium. “There is a technique the iPhone uses, called envelope tracking, that lets you adjust power at the level required at any instant in time rather than biasing for the worst case. But if you’re working with the amplitude modified wave form, you’re always going to have the problem of worst case vs. average.”
Envelope tracking is a key functional advantage for GaN, because silicon PAs can’t switch power levels up and down quickly enough to make the technique effective, said Lidow. “Envelope tracking isn’t easy for a PA to do, because you’re adjusting to track power in real time. Silicon can’t do it quickly enough, which is why it wasn’t implemented for a long time. The big savings is not on the electric bill, though. Size is more than just a thermal issue. It can be the difference between putting a 500-pound antenna on the side of a building compared to a 50-pound antenna. That’s a big selling point.”
Fig. 2: Wide bandgap chips (green) versus other materials (gray). Source: U.S. Dept. of Energy.
Systems design, not materials, as an alternative
Still, while GaN is likely to eat into silicon’s domain, it’s unlikely to actually replace silicon.
“Gallium, and some of the other III-V materials have legitimate advantages in some performance metrics over silicon, and they have played an important role—especially in defense and radiation-hardened-type devices,” said Subodh Kulkarni, CEO of CyberOptics. “Those cases are intrinsically harder to control at a process level, however. Silicon is a lot more forgiving and controllable. There are some interesting things going on with those materials to improve that, but I don’t see silicon going away right away.”
Others agree. “As the technology evolves, the sizes will shrink and we’ll see more of the ultra-lean design aspect of 5G that was supposed to help manage power and thermal issues by putting parts of the base station to sleep when they’re not needed,” said Rajeev Rajan, who runs global marketing, partnerships and CXO communications at ANSYS. “Samsung has said they’ll see a 25% reduction in some components based on the density of their ASICs for a digital front end. We’ll see more optimization from Qualcomm and Nokia and some of the other companies, which is when we’ll get better handsets and more consumer uptake.”
And while liquid cooling is many times more efficient than air cooling, it’s probably not going to replace air cooling in mobile network base stations, noted Sudhir Sharma, director of high-tech industry strategy and marketing at ANSYS.
Initial implementations are unlikely to be significantly different than previous versions, but that will get increasingly complicated as the technology progresses from sub-6 GHz to mmWave.
“These very high frequencies means you got to really a big increase in sensitivity and when you’re trying to move modules and RF ICs together,” said Ian Dennison, senior group director for customized IC and PCB group at Cadence. “The cross-dependency is very high. And so you now need people to be wearing multiple caps at once and be designing the package substrate and the ICs all at the same time.”
Benson agrees. “Generally distributed amplifiers tend to be larger, less efficient, with lower gain than a standard cascade amplifier,” he said. “But the distributed amplifiers are able to accomplish very wide bandwidths easily, which makes them attractive.”
Rather than relying on just heat sinks and wide baseband materials, it’s likely carriers will change the network topologies and hardware configurations to spread the RF transmission load out over many mid-sized base stations. They also can run fiber to microcells at various levels to avoid having just a few base stations with very high power and transmission levels.
Research from the Small Cell Forum suggests it may be possible to reduce power requirements by limiting the number of connections per device and using base stations more like fixed-wireless access point for small cells. In addition, components or network nodes can be selectively powered down at times of low activity. But the core issues are still the fundamental design of chips and chipsets in a way that makes power savings easier rather than more difficult.
“There may be system design issues I don’t see, but I wonder why there needs to be so many antenna elements in the first place,” said EJL’s Lum said. “For example, rule of thumb, you might need one chip for every four dipole elements that form a 2 x 2 MIMO block. So, if you’ve got a 512-element array, divide by four and you would need 128 chips for one radio sector. Multiply by three, and you need a lot of chips for that array. If I’m a chip guy I’m happy to be selling all those chips, but you’re adding heat for every one of those chips and don’t know what the heat dissipation is going to be like. So, why so many? Wouldn’t you want the lowest number of antenna elements possible? On the other hand, from the examples I see, LDMOS is also passing all the tests just fine at reaching the power levels needed at mmWave. So, unless there’s a good reason to switch, it’s probably better—cheaper or more stable—to stay put for now.”
The fundamental question is how much of that complex functionality it is possible to build into a single, homogeneous chip or SoC, Paranjpe said.
“It’s not an easy answer, but you have to decide if you can do high speed for the baseband processing, as well as needing the RF switches and the power amplifiers, monolithically, on silicon,” Paranjpe asked. “I’m not sure it really is possible. The requirements are so different from doing baseband and millimeter wave. Reconciling them will be very difficult.”
—Ed Sperling contributed to this report.
Related Stories
https://semiengineering.com/knowledge_centers/materials/gallium-nitride/
https://semiengineering.com/knowledge_centers/materials/silicon-carbide/
 |
 |
 |
 |
 |
 |
 |  |
Leave a Reply