By BURAK OZPINECI, LEON TOLBERT
Some technological revolutions are flashy, and some are almost invisible. We’re quite familiar with the flashy ones; they’ve given us powerful computers we can hold in the palms of our hands, devices that can pinpoint our locations by way of orbiting satellites, and the ability to bank and shop without leaving our homes.
But none of these innovations would have occurred without the technology that delivers power to them. Over the last half century, a more subtle revolution in power electronics has provided us with compact and efficient semiconductor devices that can manipulate, regulate, and convert electricity from one form to another.
Silicon has long been the semiconductor of choice for such power electronics. But soon this ubiquitous substance will have to share the spotlight. Devices made from silicon carbide (SiC)—a faster, tougher, and more efficient alternative to straight silicon—are beginning to take off. Simple SiC diodes have already started to supplant silicon devices in some applications. And over the last few years, they’ve been joined by the first commercially available SiC transistors, enabling a new range of SiC-based power electronics. What’s more, SiC wafer manufacturers have steadily reduced the defects in the material while increasing the wafer size, thus driving down the prices of SiC devices. Last year, according to estimates made by wafer maker Cree, the global market for silicon carbide devices topped US $100 million for the first time.
Within five years, we should see this market balloon as SiC devices find their way into power electronics for hybrid and all-electric vehicles, creating simpler and more efficient power systems. SiC power devices will also become vital in solar and wind energy creation, by reducing the energy lost as electricity is converted to a form that can be used on the power grid. Eventually, silicon carbide could remake the grid itself by eliminating the need for bulky substation transformers, thereby saving an enormous amount of energy that is now wasted as electricity makes its way from power plants and other sources to its final destination. Although the field of SiC power electronics is still relatively immature, we expect it’s in for a big growth spurt.
Silicon-based devices are so mature and inexpensive to manufacture, it might be hard to believe that any material could shake silicon from its perch. But silicon carbide is quite special. Many of the material’s most attractive properties stem from a single physical feature: SiC’s bandgap, the energy needed to excite electrons from the material’s valence band into the conduction band. Silicon carbide electrons need about three times as much energy to reach the conduction band, a property that lets SiC-based devices withstand far higher voltages and temperatures than their silicon counterparts.
One of the biggest advantages this wide bandgap confers is in averting electrical breakdown. Silicon devices, for example, can’t withstand electric fields in excess of about 300 kilovolts per centimeter. Anything stronger will tug on flowing electrons with enough force to knock other electrons out of the valence band. These liberated electrons will in turn accelerate and collide with other electrons, creating an avalanche that can cause the current to swell and eventually destroy the material.
Because electrons in SiC require more energy to be pushed into the conduction band, the material can withstand much stronger electric fields, up to about 10 times the maximum for silicon. As a result, a SiC-based device can have the same dimensions as a silicon device but withstand 10 times the voltage. What’s more, a SiC device can be less than a tenth the thickness of a silicon device but carry the same voltage rating, because the voltage difference does not have to be spread across as much material. These thinner devices are faster and boast less resistance, which means less energy is lost to heat when a silicon carbide diode or transistor is conducting electricity.
Because of these features, silicon carbide could be used to replace slow silicon switches with alternative designs that are faster and more energy efficient. To sustain voltages beyond about 200 volts, a silicon transistor has to be quite thick. This added thickness boosts resistance, which in turn demands impractically large devices in order to maximize current-carrying capacity. To mitigate this problem, high-voltage silicon switches tend to be bipolar transistors: They use both holes and electrons. The design carries more current, but it takes time for all the charge carriers to fully exit the device. When the transistor is being switched from its “on,” current-carrying state to its “off,” voltage-blocking state, there is a period of overlap where the remaining charge carriers are exposed to high voltage and dragged through the device, dissipating heat.
Using silicon carbide instead of silicon in high-voltage devices will let manufacturers replace slow silicon bipolar transistors with single-carrier, or unipolar, devices such as metal-oxide-semiconductor field-effect transistors, or MOSFETs. Very few charge carriers are left behind in such devices, so the transistors can be switched quickly and far more efficiently. The faster devices also have the added benefit of more-compact and less-expensive packaging because they require smaller control circuitry.
For all its fine qualities, silicon carbide has been a difficult material to master. One of the biggest hurdles to its widespread use in power electronics has been in wafer manufacturing. When engineers first started working with the material in the 1970s, they struggled to grow large single crystals of the stuff—the silicon and carbon atoms had a habit of combining with one another to form a hodgepodge of different crystalline structures.
Over the years, researchers succeeded in creating larger and larger single-crystal wafers. And in 1991, a few years after the company was founded, Cree released the first commercially available SiC wafers. They were just an inch across and used mostly for research, but it was a start. Since then Cree and other manufacturers, including Dow Corning,SiCrystal, TankeBlue, and II-VI, have made steady progress in boosting the size of the wafers; these days 4-inch SiC wafers are common, and 6-inch wafers are on the horizon. A larger wafer size means that more devices can be built on each wafer, which drives down device costs.
At the same time, companies have been working to overcome another early stumbling block: a high number of defects in SiC crystals. Unlike silicon, SiC doesn’t have a liquid phase. As a result, SiC crystals are grown layer by layer from vapor at roughly 2500 °C. This process is difficult to control and can easily create tiny, tornado-like tunnels called micropipes (shown at right), which arise from dislocations in the crystal early in the wafer formation process.
Devices built atop these micropipes don’t perform as designed. Even a few micropipes per square centimeter is enough to erode device yield and thus boost costs. But as wafer producers fine-tuned manufacturing processes, they also made steady strides in eliminating such defects. In 2005, the U.S. firm Intrinsic Semiconductor Corp., later acquired by Cree, debuted 3-inch SiC wafers with no micropipes, and 4-inch micropipe-free wafers are now available.
Of course, wafers would be nothing if there weren’t devices to build on top of them. In 2001, more than 50 years after the first silicon power electronic devices emerged,Infineon Technologies, based in Neubiberg, Germany, released the first commercial SiC device. It was a Schottky diode, a simple junction made from metal and semiconducting material. Schottky diodes rectify alternating currents in much the same way that a standard p-n junction does, but the devices exhibit much faster response times. Although they cost more than silicon diodes, SiC Schottky diodes offer a range of benefits, including better energy efficiency and reliability and cooler operation. They also eliminate the need for devices like snubbers, which would otherwise be used to protect silicon circuitry from current spikes. In less than 10 years, SiC Schottky diodes have all but replaced the silicon p-n diodes in switched-mode power supplies for computers, particularly those in large data centers. Manufacturers now offer SiC Schottky diodes that can withstand voltages as high as 1700 V, more than five times the maximum voltage of comparable silicon devices.
But to truly revolutionize power electronics, you need a second component: transistors. These more sophisticated devices have taken longer to realize in silicon carbide. It wasn’t until 2008 that the first SiC transistors—junction field-effect transistors (JFETs) manufactured by Mississippi-basedSemiSouth Laboratories—finally hit the marketplace. The number of transistor offerings has since boomed. SiC transistors with a range of architectures are now offered by the likes of Cree,Infineon, Rohm, and TranSiC. Each design has its advantages, and the jury’s still out on which one will get the biggest share of the market, but the competition is clearly heating up.
At Oak Ridge National Laboratory, in Tennessee, we’ve been exploring how well SiC diodes and transistors work as the power electronic devices for all-electric and hybrid electric vehicles. After the battery, electronics are the key added cost to these vehicles. Electronics are needed to convert wall power into battery power, to recharge the battery from the engine or from the brakes, and most important, to operate the traction drive, which transforms battery power into electricity that can run the motors that propel the vehicle. Of all the electronics in an electric vehicle, the traction drive draws the most power.
The drive has two main parts: a boost converter that increases the DC voltage from the battery and an inverter that converts this electricity into the three-phase AC needed by the motor. The three-phase inverter in turn consists of six diodes and six transistors. In computer and laboratory simulations at Oak Ridge, we’ve shown that simply swapping silicon diodes with SiC Schottky diodes cuts the inverter’s energy loss by 33 percent, consistent with other estimates. The reduction doubles if you also replace the silicon transistors with SiC transistors. This boost in efficiency results mainly from SiC’s lower resistance—which means it loses less power to heat—and from faster, more efficient switching.
But SiC’s advantages don’t end there. Because it takes extra energy to kick electrons into the conduction band, SiC’s wide bandgap also makes the material much more heat resistant than silicon. Excess heat can excite so many electrons that it can interfere with a device’s operation. For silicon, this thermal failure occurs at around 150 °C, but SiC devices can withstand considerably more than twice that temperature. This thermal resistance makes SiC attractive for a range of rugged applications, including military systems and electronics for oil wells, geothermal plants, and robotic spacecraft.
In hybrid and all-electric vehicles, SiC’s operating temperature is high enough to obviate the need for one of the bulkiest engine components: the liquid cooling system. Hybrid vehicles need two cooling loops—one for the gasoline engine, which runs at 105 °C, and another to cool the power electronics and traction motor. Because silicon-based electronics stop performing above roughly 150 °C, and because some physical space separates the electronics from the coolant, this second loop needs to run even colder than the engine loop—at roughly 65 to 70 °C.
Liquid cooling adds significantly to the overall size of the engine, and if the liquid leaks out, it can destroy the electronics. Our simulations suggest that SiC inverters, because they can operate at higher temperatures, could reduce the size of the cooling system by 60 percent. If we combine these inverters with other high-temperature components like high-temperature capacitors, we might be able to eliminate the second loop altogether and simply cool the electronics with air. First, though, the packaging and peripheral components—the capacitors, control circuits, and drivers that turn transistor gates on and off—must also be made to withstand high temperatures. We’ve slowly been making progress on this front and have built drivers from scratch that work at up to 200 °C.
How efficient could SiC ultimately make electric vehicles? Electric traction drives already convert more than 85 percent of their power into usable mechanical energy, more than double the raw efficiency of a gasoline engine. But the U.S. Department of Energy (DOE) has set some ambitious goals (pdf) for boosting the efficiency even further. By 2015, the agency says that drives should convert 93 percent of their power into mechanical work and by 2020, more than 94 percent. In other words, it wants future drives to lose half as much energy as present-day drives. These efficiency targets wouldn’t be hard to reach by themselves, but the DOE also expects that electric traction drives in 2020 will be half the size and less than a fifth of the cost. These ambitious targets will be all but impossible to hit with silicon alone, but we think SiC has the potential to get us at least most of the way there.
One area where SiC devices are already making inroads is solar power. Photovoltaic panels, whether they’re mounted on a roof or spread across hectares of land, need inverters to convert the DC electricity made by the panels into AC electricity that can be fed into the power grid. This conversion process is already quite efficient: Silicon-based inverters lose just 2 to 3 percent of the energy they process. But inverters containing SiC diodes and transistors can easily cut that loss in half. Over the 20-year lifetime of a 10-megawatt solar plant, that could add up to hundreds of thousands of dollars in savings.
That’s just for starters. Infineon has estimated that improvements in power electronics could eventually reduce electricity consumption by as much as 30 percent. To get there, the U.S. National Science Foundation funded the creation of the FREEDM Systems Center in 2008, a corporate and academic partnership that is researching ways to build a smart, flexible power grid using wide-bandgap devices. Last year, the DOE’s Advanced Research Projects Agency–Energy also put money toward revamping power grid electronics. Two grants went to teams led by Cree and GeneSiC Semiconductor that are exploring ways to make SiC devices that can operate at more than 10 000 V, up to 15 000 V—well beyond the capabilities of silicon devices.
Remaking the power grid calls for SiC components that don’t yet exist, including high-voltage bipolar transistors and p-n diodes. But if the research succeeds, it will pave the way for new devices that can connect distribution lines to higher-voltage transmission lines. At present, that job is performed by massive, multiton transformers, which dominate power substations. Someday, though, utility companies could replace these behemoths with far more efficient solid-state transformers, each the size of a suitcase.
Of course, that’s still a long way off. One key technical hurdle will be continuing to improve the quality of SiC channels. Today’s SiC transistor channels carry charges a factor of 10 slower than their theoretical limits, but modifications, such as better surface quality, should help.
Right now, silicon carbide is experiencing the same sorts of growing pains that silicon did in the 1950s and 1960s, when physicists and engineers saw it as a replacement for germanium. Despite the fact that SiC devices are still relatively new and more expensive than their silicon counterparts, the material has already demonstrated clear advantages over the alternatives. As more and more such devices come to market and their capabilities expand, they could start a revolution of their own.
This article originally appeared in print as “Smaller, faster, tougher.”
About the Authors
Burak Ozpineci and Leon Tolbert are both IEEE senior members. Ozpineci heads the power electronics and electric machinery group at Oak Ridge National Laboratory, in Tennessee. Tolbert is the Min Kao professor of electrical engineering and computer science at the University of Tennessee, in Knoxville. The two began working on silicon carbide devices in 2001, when a friend of a friend of a friend sent along some of the first SiC Schottky diodes. “We were hooked,” Ozpineci says.