By Gina Roos, editor-in-chief
Silicon (Si)-based semiconductors have a decades-long head start over wide-bandgap (WBG) semiconductors, primarily silicon carbide (SiC) and gallium nitride (GaN), and still own about 90% to 98% of the market, according to chip vendors. Though far from a mature technology, WBG semiconductors are making inroads across industries thanks to their performance advantages over silicon, including higher efficiency, higher power density, smaller size, and reduced cooling.
Getting the optimal design using either SiC- or GaN-based power semiconductors requires a little more know-how and careful consideration in several areas, including switching topology, electromagnetic interference (EMI), layout, paralleling, and the selection of the gate driver. It’s also important to address reliability and cost issues.
In overlap applications where Si, SiC, and GaN can be used, the choice comes down to density, efficiency, and cost, and once the designers understand those three parameters, it will guide them to which switching technology to use. (Image: Infineon Technologies)
Why move to WBG?
It all starts with the decision to move from Si- to SiC- or GaN-based power devices based on design goals.
There are three things — cost, efficiency, and density — that designers need to look at, whether they are using silicon or switching over to SiC or GaN, said Bob Yee, technical marketing engineer, power discretes, Infineon Technologies AG. Infineon plays in both the SiC and GaN markets, with its CoolSiC and CoolGaN portfolios, and also offers Si MOSFETS and IGBTs.
Cost is measured in dollars per watt, efficiency in percentage of power in/power out, and density in watts per cubic inch, Yee said. “Once you determine those goals, that will dictate the type of technology and where that cost point is.”
Growing demand for power semiconductor devices is driving the wide-bandgap semiconductor market. Key players have been investing in the development and mass production of materials and wafers for SiC and GaN. Where is the WBG market headed? Who are the dominant players? How are they addressing the decades-old issues of high cost, limited volumes, and constrained supply chains? This EE Times Special Project will unpack the technology, applications, and dynamics of the WBG semiconductor market.
Size and weight matters in understanding whether you use silicon or WBG, said Yee, citing an example of an adapter design in a small form factor, which likely would use a GaN transistor (HEMT) over Si MOSFETs. The reason? GaN’s higher switching frequency allows the designer to shrink the size of the magnetics, which accounts for a big portion of the power supply size.
“Designers have to understand what their density needs are, which eventually will dictate the efficiency because there is less space to dissipate heat in a small form factor,” he added. “This means the efficiency needs to be higher, pushing the designer to use WBG.”
The magic line
Silicon-based solutions over the decades have achieved higher efficiency and smaller size, but there is a point at which WBG semiconductors offer better efficiency. Yee cites an example of a 100-W power supply — 100 W in and 94 W out, which translates into a 6% loss or 94% efficiency. “That is the magic line where you separate from silicon to using WBG technology,” he said. “If engineers are designing up to 94%, it’s well covered with silicon and there’s no reason to go to WBG and pay more. However, if you are trying to achieve an efficiency of 96%, there’s really no other choice but to use WBG, and that comes down to the parasitic losses on the properties of the switches itself, in addition to the topologies.
“If you want to achieve 96% efficiency, you need a new topology that takes advantage of GaN or SiC,” he added.
A good example is using a power-factor-correction (PFC) topology. If a designer looks at how a switch technology is optimized for a particular topology — for example, a totem pole PFC, which takes advantage of WBG — it increases performance, said Yee, “which is why bridgeless totem pole PFCs are really a slam dunk for WBG.”
Designers need to evaluate WBG devices by looking at how the switch technology is optimized for a particular topology to achieve the biggest performance improvements. (Image: Infineon Technologies)
Challenges
Designers can optimize their designs for higher frequency, higher power density, and higher efficiency. And that’s where some of the WBG technology challenges appear. When switching at a higher frequency, designers need to look out for EMI and higher switching losses.
The parasitics of WBG, less than the silicon equivalent, mean that EMI is easily elevated because it’s a much faster switch. When you’re optimizing for high frequency, you need to pay attention to EMI, and additional switching losses need to be accounted for, said Yee.
Anup Bhalla, VP of engineering at UnitedSiC, a manufacturer of SiC FETs, SiC JFETs, and SiC Schottky diodes, agreed. “EMI problems become more serious especially if you’re trying to get the system benefits of higher power density, which really means everything gets smaller, and the only way it gets smaller is if you switch a lot faster. This allows you to make the transformers, inductors, heat sinks, and other things a lot smaller.”
Switching faster also means you’re running at high rates of change of voltage and current, which can lead to big voltage overshoots and EMI problems, so layouts get more challenging, said Bhalla.
“These fast voltage changes on the power side of the circuit can easily affect the signal side of the circuit because it can send a little voltage spike here or there without you knowing it,” he said. “It could trigger the gate driver at the wrong time and blow everything up, so you have to be more careful on your layout. It usually involves a considerable engineering effort [by the customer] to get there, and a lot of them have made this leap over the last four or five years.”
Optimizing layout
Layout can be a challenge; Yee said the biggest impediment is between the driver and the gate. “There are three terminals designers need to look out for. It’s the driver output to the gate input, whether it’s SiC or GaN, and the ground connection of the driver’s source to the source of the WBG device.”
The No. 1 thing they need to minimize is loop inductance because WBG parts are switching so much faster, said Yee. “If they do not pay attention to that, they will create a radio that’s going to emit radiation.” So special care needs to be taken for those connections. To mitigate the challenge, Infineon recommends the use of WBG devices with Kelvin source capability.
Layouts also impact paralleling for higher-power applications. Paralleling is fairly straightforward, said Bhalla. “It’s the same general physics — you have to keep the layout symmetric and balanced. We have to keep our parameter distributions between parts relatively tight so that all the parts look about the same so they will parallel easily.
“Designers like to take these fast parts and parallel them just like they were paralleling IGBTs in the old days,” he added. “And that is tough because IGBTs are a hell of a lot slower, and so they are somewhat easier to parallel. When you try to parallel and switch 10× faster at the same time you have to do much more work on how you lay things out.
“You have to be careful about at least doing a halfway decent layout so that all the current paths look about the same between the parallel devices. You can’t have one device with one -fifth of the inductance of the other one and then expect them to parallel; that won’t work.”
Sometimes the easiest way to show engineers how to solve challenges around layout and paralleling is to give them a demo board, said Bhalla. “We’re very careful to make sure that when you’re using these devices in parallel, the loop used to drive the gate has to be kept decoupled from the loop that is routing all the power/current. The gate drive circuit is a little loop, and then there’s a big powerful loop driving all the power/current, and you want to minimize the coupling between these two things. If you do that, you know paralleling goes a lot better and a lot easier.”
Same is true when using GaN devices. “Engineers have to understand layout better than they used to because GAN is fast,” said Larry Spaziani, vice president of sales and marketing, GaN Systems, a specialist in GaN HEMT/E-HEMT devices. “If you don’t have a proper layout, then you can run into problems with performance or EMI or even failure mode.
“GaN doesn’t change the rules of layout, but everything is smaller, tighter, and more compact, so you have to make sure that you do it right,” he added.
Minor tweaks for SiC
SiC can be used as a performance replacement for Si IGBTs or Si MOSFETs, partly because the driving structure is very similar — it’s a normally off part and uses standard drivers, but there are slight differences, explained Yee.
With Si MOSFETs, the driving voltage is 10 V to 12 V; however, if you use SiC, it’s 0 V to 18 V, and the undervoltage lockout (UVLO) changes from 8 V for Si to 13 V for SiC, so there are minor tweaks that the designer would have to do when moving from Si to SiC, explained Yee.
However, with GaN, the driving structure is completely different; it’s not the same as an IGBT or a MOSFET, he added. ”You have to use a specific driver with specific turn-on and turn-off times. So the designer really needs to pay attention to the driving scheme, not only with the timing, but if they’re going to parallel GaN FETs, they have to have a perfect symmetrical layout between the driver and the GaN FET.”
It’s important to note that designers can use a standard driver for GaN as long as it supports the gate drive voltage and UVLO, but again, it requires a tweak in design. Most suppliers recommend using the newer generation of gate drivers to get the highest performance by being able to switch at the fastest switching speeds.
Using a standard gate driver to drive a GaN device requires the addition of a negative voltage supply for turning the device on and off safely, compared with dedicated GaN drivers. (Image: Infineon Technologies)
“Only for GaN do you have to supply a positive and negative voltage if you’re using a standard driver, which is why we prefer customers to use a dedicated driver,” said Yee. He recommends Infineon’s 1EDF56x3 series of GaN gate drivers.
Not all SiC devices are created equal
The majority of WBG devices are not drop-in replacements for Si MOSFETs or Si transistors. The exception is cascode-type devices that require little to no additional engineering effort. However, designers lose some of the benefits of WBG semiconductors.
One example is UnitedSiC’s SiC products, which are all housed in silicon-compatible packages. This means the devices can literally drop these into sockets that previously used IGBTs or Si superjunction MOSFETs.
Bhalla said that one of the unique aspects of its products is that it manufactures cascode-based devices that work like a MOSFET. These SiC FETs include a SiC fast JET co-packaged with a cascode-optimized Si MOSFET to offer a standard gate drive SiC device packaged in standard through-hole and surface-mount packages. “Our cascode-type device is a literal drop-in without any changes other than maybe a gate resistance change,” he said.
In addition, the devices don’t require a special driver; they are compatible with standard silicon gate driver ICs that have been on the market for a decade from all the key suppliers, including the older ones used with SiC MOSFETs and “old school” IGBTs, said Bhalla.
A lot of good gate drivers have been developed specifically for SiC over the past two years, he added. “They’re more expensive, but people have started using them, and our devices are compatible with those better drivers, too.”
But there are some drawbacks, including not getting the highest performance out of the WBG devices. “We’re selling our superfast devices in these packages that have a lot of inductance in them,” said Bhalla. “And when you put a high slew rate (di/dt) in your circuit through these packages, it just exacerbates all the problems of fast switching — bigger overshoots, more oscillation, etc.”
Bhalla said the transition to much better packaging is a work in progress. “This is the reality: People are using the partial benefit of SiC and still get some benefit in their end systems doing it the cheap and dirty way.
“A great percentage of the world is still in silicon, so for them to make the move from silicon to silicon carbide, we offer a really good stepping stone,” he said.
Bhalla thinks that by next year, there will be a lot of top-side–cooled surface-mount packages and even surface-mount–type modules that integrate the entire half-bridge in one package. “It’s got to be done, because you know without that, users can’t get all the benefit out of it and they can’t move to the next level,” he said.
For example, UnitedSiC recently launched a 7-mΩ RDS(ON) , 650-V device in a TO-247 package. (Low RDS(ON) makes it possible to achieve higher efficiencies.) The company’s closest competitor has 3× higher ON resistance, but a problem that UnitedSiC ran into is the package leads actually get hotter than the chip. “So we took a 200-A device and derated it to 120 A, because when we use this device in practice, we see that the leads get hotter than the chip itself,” said Bhalla.
UnitedSiC introduced the first SiC FETs, with RDS(on) of <10 mΩ with increased efficiency and lower losses, by combining a third-generation SiC JFET and a cascode-optimized Si MOSFET in a familiar TO-247 package that can be driven at the same gate voltages of Si IGBTs. (Image: UnitedSiC)
GaN benefits
OEM designers across segments from consumer electronics to automotive have a couple of design requirements in common: They want higher power density and smaller electronics.
At a higher frequency, just about all the components — capacitors, inductors, transformers, etc. — in the power system can be smaller, said Spaziani, and because GaN is so efficient and generates so little heat, it doesn’t require any heat sinks, so designers can save space and cost just by removing the heat sink. Or they might stay with the same frequency to achieve higher efficiency. Oftentimes, even with 1% higher efficiency, it’s enough to get customers in the server power supply world from a platinum to a titanium level [96% efficiency], he said.
This is nothing different from what an engineer would normally do, said Spaziani. They usually have to optimize their board whether they’re using silicon or another technology, but there is a difference in gate drive. With GaN and SiC, the gate drive behavior is different than silicon MOSFETs and silicon IGBTs, so the first thing that an engineer has to ask is, “How do I drive the gate?”
Over the past 30 years, MOSFETS have basically become 0- to 12-V gate drive circuits, whereas GaN is either –3 to 6 V or 0 to 10 V or 0 to 5 V; they’re all a little different, said Spaziani. “But the good news is that GaN Systems is now six years into our journey, and we have about a dozen major semiconductor companies that have created drivers to drive GaN, so now, it’s just a simple application decision.”
GaN Systems also offers a circuit called EZDrive, which eliminates the need for a discrete driver. It converts a 12-V MOSFET drive to a 6-V GaN drive with about six components. “It’s really cheap, and adapter designers love this circuit,” said Spaziani. “It’s easy to use, takes no power, and is small, and they don’t have to have a customized gate driver.”
Debunking GaN myths
GaN vendors believe there are still several myths about GaN technology that are either false or half-truths. Issues include EMI, paralleling, avalanche capability, reliability, and cost.
EMI is worse with GaN devices. GaN provides great switching edges, enabling higher efficiency and higher frequency, but it doesn’t mean EMI is worse. In fact, vendors said it’s often better than silicon with a good layout and can result in a smaller EMI filter, thus lowering cost.
Paralleling is a common concern. One myth is that GaN is only good at low power and high frequency. GaN Systems, as an example, has customers switching at 20 kHz to 20 MHz, and at high power, they are paralleling devices. GaN transistors can parallel quite well; just make sure that each transistor carries about the same amount of current. For example, if you’re paralleling two devices and one transistor carries 70% of the current, it’s going to wear out faster and the circuit will fail sooner. Caveat: Devices from different SiC and GaN vendors parallel a little differently.
No avalanche capability. MOSFETs go into avalanche mode to clamp down a voltage spike to protect the rest of the circuit from failure. The way that GaN device manufacturers get around this is by designing a lot of margin into the voltage rating. For example, GaN Systems’ 650-V–rated devices don’t fail until well over 1,000 V.
Reliability and cost is not equal to silicon. Reliability is measured by failures in time (FIT). Silicon has been around for decades and it’s proven to be reliable by most suppliers. But that’s not the case for WBG semiconductors. Like any new technology, reliability risks are raised and costs are higher. It’s a tough comparison between WBG devices and silicon devices simply because the reliability of silicon chips is well documented, and high-volume production over the years has driven down costs.
But some WBG vendors, like GaN Systems, say that reliability [FIT] is on par with silicon, with the price gap closing significantly over the past five years, dropping from 3× to 5× more expensive to 1.5× to 2× more.
GaN Systems’ devices show an FIT rate of <0.1. (Image: GaN Systems)
WBG suppliers provide design tools, demo boards, and guidance to help customers transition to SiC and GaN devices, but in the end, it’s the customer who has to put in the effort and do the R&D to make the leap to the new technology.
“All of the benefits are there, but they don’t come without engineering effort on the part of our customers with guidance,” said Bhalla.
Articles in this Special Project
GaN, SiC Offer a Power Electronics Alternative
By Maurizio Di Paolo Emilio
Wide bandgap materials may be posed to replace silicon for some low-power, high-frequency applications.
Where Is the Wide-Bandgap Market Going?
By Maurizio Di Paolo Emilio
While silicon still dominates the market, the emergence of GaN and SiC devices will soon direct technology toward new, more efficient solutions.
Wide-Bandgap Materials in Hybrid and Electric Vehicles
By Stefano Lovati
Electric vehicles are looking to wide-bandgap semiconductors, which offer greater power efficiency, smaller size, lighter weight, and lower overall cost.
Q&A: Wide Bandgap Semiconductors Poised to Make a Splash
By Maurizio Di Paolo Emilio
GaN Systems says wide bandgap chips will become ubiquitous across a range of industries for several applications.
How SiC Devices Have Changed the Face of Semiconductor Sector
By Anup Bhalla
SiC FETs are already opening up new applications at higher power and higher switching frequencies.
Wide Bandgap Technologies: New Norm for 21st century Power Electronic Applications
By Thomas Neyer and Mehrdad Baghaie
Researchers and universities have experimented with several wide bandgap materials, which showed high potential to replace incumbent silicon technologies.
GaN Enabling a Revolution in Charger Design
By Chris Lee
GaN switch technology is no longer nascent: engineers are leveraging it to innovate and provide better switching solutions.
Silicon Carbide Adoption Enters Next Phase
By Orlando Esparza
Demand continues to grow for silicon carbide (SiC) technology that maximizes the efficiency of today’s power systems while simultaneously reducing their size and cost.
Gallium Nitride: The Future of Grid Has Already Arrived
By Masoud Beheshti
For years, designers have described a future where gallium nitride (GaN) can help realize unprecedented levels of power density, system reliability, and cost in grid applications.
Silicon Is Dead…and Discrete Power Devices Are Dying
By Alex Lidow
Beyond performance and cost improvement, the most significant opportunity for GaN semiconductor technology to impact the power conversion market comes from its ability to integrate multiple devices a single substrate.
Solving the Challenges of Driving SiC MOSFETs
By Ming Su & Mitch Van Ochten
Silicon carbide (SiC) provides a number of advantages over silicon for making these power switching MOSFETs.
Advertisement
Learn more about Electronic Products Magazine
