Millions of gallium nitride transistors (GaN) are already in use in chargers for our mobile devices. But are these power semiconductors also suitable for applications beyond consumer goods, i.e., in industrial and automotive electronics?
At electronica 2022, Markt&Technik editor Ralf Higgelke discussed this with prominent panelists from power electronics. Two arguments emerged in favor of gallium nitride: reliability and the capability to integrate measurement and protection circuitry on the chip.
For Dr Denis Marcon, General Manager of Innoscience Europe, as well as for all the other panelists, the question of whether GaN is reliable is settled: GaN is reliable. "Many people are just afraid because there is not enough field experience," he summarizes the psychological issue users face. "And no matter whatever tests you do in the lab, that data always has to be confirmed by field experience. That's the final piece of the puzzle to convince industrial and automotive customers to shift to GaN," adds Marcon, who obtained his Ph.D. in 2011 on the topic of GaN reliability. For this reason, gallium nitride is evolving from consumer applications to those with high reliability requirements, such as industrial and automotive.
Professor Florin Udrea, CTO and co-founder of Cambridge GaN Devices, explaines that gallium nitride could even become the most reliable semiconductor material stating: "GaN is very efficient, so the junction temperatures can be lower than silicon. Consequently, GaN devices are also much more reliable because, according to Arrhenius, reliability decreases exponentially with increasing temperature."
Another reason why gallium nitride is so reliable inherently stems from its lateral device structure. "We can integrate protection functions monolithically on the chip, which is simply impossible with vertical power devices, i.e., neither silicon nor silicon carbide," stresses Gene Sheridan, CEO and co-founder of Navitas Semiconductor. Protection against overvoltage, overcurrent, overtemperature, short circuit, as well as gate protection and ESD protection for all pins are fundamental features that enhance the reliability of gallium nitride, he said.
Sheridan also highlights another aspect: "Reliability is always inversely proportional to the number of interconnects. By reducing manual tasks, reducing the number of components, and reducing the number of interconnects at system level, reliability is substantially improved beyond just the transistor."
That includes heat sinks, which often become obsolete with GaN, he said, as well as passive components that become smaller and can thus be automatically picked and placed surface-mounted or even integrated into the GaN IC. However, not all components must be integrated monolithically, the experts unanimously agree - especially in the case of half-bridges. This approach is highly desirable because 70 to 80 percent of all power electronic systems are based on half-bridge concepts.
"At Infineon, we always co-package a silicon chip because more complex functions are easier to deploy in a CMOS technology in silicon," clarifies Dr Gerald Deboy, Distinguished Engineer Power Discretes. "In gallium nitride, we only integrate what is absolutely necessary there." According to Deboy, this includes pull-down transistors in addition to the measurement and protection functions already mentioned.
Such a hybrid integration is very effective, according to Professor Udrea, because two lateral component structures - i.e., gallium nitride on the one hand and silicon CMOS on the other - can easily be tied together because you do not have to worry about how to contact the backside of the substrate. Nevertheless, Udrea identifies the monolithic integration of half-bridges as a mid-term goal.
As a new semiconductor material, however, gallium nitride also displays novel failure mechanisms. "For example, there are dielectric breakdown mechanisms in HTRB which do not exist in silicon. Also, dynamic on-resistance is an issue unique to gallium nitride," Dr Deboy points out. "That's why we at Infineon decided to do end-of-life testing, really killing components. Then we look at the population and create Weibull curves from that."
It is also important for Infineon to keep all manufacturing processes in-house. "We really want to control all the manufacturing steps along the entire supply chain," Deboy continued. "We only purchase the silicon substrates. That's not complex at all. But for everything else - the fabrication of the super-lattice, the gallium nitride stack and the transistor cell - we rely on our own processes and our own manufacturing facilities."
Prof. Florin Udrea is not as reluctant as Deboy about failure mechanisms, citing an example, "Ten years ago, many people were discussing vertical leakage currents between the drain and the substrate in MOSFETs and IGBTs. Today, no one talks about it anymore, because we know by now that this leakage current only occurs at extremely high voltages."
He transfers this to the so-called current collapse in gallium nitride HEMTs: "A few years ago, people were discussing current collapse. Today, you don't hear about it anymore. It is true that the on-resistance increases briefly when the device is switched on - called dynamic Ron - but we now have this well under control. Even in the most demanding tests to highlight this dynamic Ron, the value only increases by 20 to 30 percent. And in real world applications, the value is most likely well below ten percent."
Many people think that gallium nitride is less suitable than silicon carbide for blocking voltages above 650 or 900 V. Prof. Udrea vigorously disagrees: "There are already developments at 1.2 kV. And we ourselves have manufactured a multi-channel lateral GaN device in the lab that can handle as much as 10 kV! I'm not saying that this will be a market for lateral devices, but I cannot see that the shift to 1.2 kV and to high power applications will be difficult."
Denis Marcon agrees, pointing out that silicon carbide has been on the market for a longer time and is therefore more mature. "I wonder what would have happened if GaN had started at the same time and had the same technological maturity today as SiC. Maybe the situation would be completely different today at 1200 volts," the Innoscience manager says.
But gallium nitride will likely find a huge market at lower blocking voltages as well. Dr Gerald Deboy explains this as follows: "100- and 200-volt GaN is suitable for secondary-side synchronous rectification with no reverse recovery delay. Thus, the switching frequency on the primary side can be raised to hundreds of kilohertz, which in turn is only viable with high-voltage GaN." Prof. Udrea agrees but suggests that the advantages of GaN over silicon become smaller with lower blocking voltage. "The lower the voltage, the higher the frequencies have to be for GaN to fully leverage its advantages," says the CTO of Cambridge GaN Devices.
One final advantage offered by power semiconductors based on gallium nitride over those based on silicon carbide and silicon at the moment is availability and the supply chain. Silicon MOSFETs and IGBTs still have extremely long lead times, and with silicon carbide, raw wafers are the limiting factor. Hassane El-Khoury, CEO of onsemi, recently noted that onsemi’s entire output of SiC semiconductors for 2023 has already been sold.
The situation with GaN is quite different, as "there are more than three dozen 150- and 200-mm wafer fabs just in the U.S. that are likely to be mothballed," Gene Sheridan clarifies. "With GaN, we can bring new life to these fabs - for decades to come - in half the time it takes to build a new fab," says the Navitas CEO. Those fabs are fully depreciated, so the cost structure is close to zero, he adds. "We get a lot of capacity, so we can keep lead times very short for a long period of time."