The September 2022 issue of IEEE Spectrum is here!

IEEE websites place cookies on your device to give you the best user experience. By using our websites, you agree to the placement of these cookies. To learn more, read our Privacy Policy.

N early 70 years ago, two physicists at Bell Telephone Laboratories—John Bardeen and Walter Brattain—pressed two thin gold contacts into a slab of germanium and made a third contact on the bottom of the slab. The flow of current through this configuration could be used to turn a small signal into a larger one. The result was the first transistor—the amplifier and switch that was, arguably, the greatest invention of the 20th century. Thanks to Moore’s Law, the transistor has delivered computers far beyond anything thought possible in the 1950s.

Despite germanium’s starring role in the transistor’s early history, it was soon supplanted by silicon. But now, remarkably, the material is poised for a comeback. The world’s leading-edge chipmakers are contemplating a change to the component at the very heart of the transistor—the current-carrying channel. The idea is to replace the silicon there with a material that can move current at greater rates. Building transistors with such channels could help engineers continue to make faster and more energy-efficient circuits, which would mean better computers, smartphones, and countless other gadgets for years to come.

For a long time, the excitement over alternative channels revolved around III-V materials, such as gallium arsenide, which are made from atoms that lie in the columns just to the left and right of silicon in the periodic table of elements. I was active in that research. In fact, eight years ago, I wrote a feature for this magazine heralding the progress that had been made in constructing transistors with III-V channels.

The two transistors in the FinFET-based inverter contain finlike channels, which stand out from the plane of the wafer (top, one set of fins (in pink) from a bird’s-eye view; bottom, an oblique view of another set). The distance between each fin in the top image is in the tens of nanometers.Images: Heng Wu/Purdue University

But as we eventually discovered, the III-V approach has some fundamental physical limitations. It’s also likely to be too expensive and difficult to integrate with existing silicon technology. So a few years ago, my team at Purdue University, in West Lafayette, Ind., began experimenting with a different kind of device: a transistor with a channel made of germanium. Since then, we’ve demonstrated the first complementary-metal-oxide-semiconductor (CMOS) circuits—the kind of logic inside today’s computers—made with germanium grown on ordinary silicon wafers. We have also constructed a range of different transistor architectures using the material. These include nanowire devices, which may be next in line when the present state-of-the-art transistor design, known as the FinFET, can’t be miniaturized any longer.

Best of all, it turns out that putting germanium back into the mix isn’t as big a challenge as it might seem. Transistors that use a combination of silicon and germanium in the channel can reportedly be found in some recent chips, and they made an appearance in a 2015 demonstration of future chip-manufacturing technology by IBM and partners. These developments could be the first steps in an industry trend to adopt the use of higher and higher proportions of germanium in the channel. In a few years’ time, we may find that the material that brought us the transistor has helped usher it into a new age of remarkable performance.

Germanium was first isolated and identified by the German chemist Clemens Winkler in the late 19th century. Named in honor of Winkler’s homeland, the material was long considered a poor conducting metal. That changed during World War II, when germanium’s semiconducting properties—that is, its ability to switch between permitting and blocking the flow of current—were discovered. Solid-state devices based on germanium boomed in the postwar years; U.S. production grew from a few hundred pounds in 1946 to meet a demand for over 45 metric tons of the stuff by 1960. But silicon ultimately won out; it became the material of choice for logic and memory chips.

There are some good reasons why silicon dominated. For one thing, silicon is far more abundant and thus a lot cheaper. Silicon also has a wider bandgap, the energy hurdle that must be overcome in order for a transistor to carry current. The larger the bandgap, the harder it is for current to leak across the device when it’s supposed to be off, draining power. As an added benefit, silicon also has better thermal conductivity, making it easier to draw away heat so that circuits don’t overheat.

Given all those advantages, it’s natural to wonder why we’d ever consider introducing germanium back into the channel. The answer is mobility. Electrons move nearly three times as readily in germanium as they do in silicon when these materials are close to room temperature. And holes—the electron voids in a material that are treated like positive charges—move about four times as easily.

Speedy Circuits: This nine-stage CMOS ring oscillator, presented in 2015, was built on a germanium-on-insulator wafer.Image: Heng Wu/Purdue University

The fact that both electrons and holes are so mobile in germanium makes the material a convenient candidate for constructing CMOS circuits. CMOS employs two different kinds of transistors: the p-channel FET (pFET), whose channel contains an excess of free-moving holes, and the n-channel FET (nFET), which has a similar excess of electrons. The faster these electrons and holes can move, the faster the resulting circuits can be. And because less voltage must be applied to draw those charge carriers along, circuits can also consume considerably less energy.

Of course, germanium isn’t the only such high-mobility material. The III-V compounds mentioned earlier, materials such as indium arsenide and gallium arsenide, also boast excellent electron mobility. In fact, electrons in indium arsenide are nearly 30 times as mobile as they are in silicon. So far, so good. The problem is that this amazing property does not extend to the holes in indium arsenide, which are not much more mobile than holes in silicon. That limitation makes it almost impossible to make a high-performance pFET, and the lack of a fast pFET rules out speedy CMOS circuitry, which is not designed to tolerate a very large difference in the speed between nFETs and pFETs.

One potential fix is to take the best of each material. Researchers at various institutions, such as the European semiconductor research organization Imec and IBM’s Zurich laboratory, have demonstrated ways to make circuits that build nFET channels from a III-V material and pFET channels from germanium. This technique could lead to very fast circuits, but it also complicates the manufacturing process.

For this and other reasons, we favor a straight-germanium approach. The germanium channels should significantly boost performance, and the manufacturing challenges are expected to be more manageable.

To make germanium—or any alternative channel material—work in mass manufacturing, we must find a way to incorporate the material on the dinner-plate-size silicon wafers that are used to make today’s chips. Fortunately, there are multiple ways to build a germanium layer on a silicon wafer that can then be fashioned into channels. Using a thin layer of the stuff significantly mitigates two key problems with germanium—the fact that the material is costlier than silicon, and that it is a relatively poor conductor of heat.

But replacing silicon in a transistor channel isn’t just a matter of slotting in a thin, high-quality layer of germanium. The channel has to work seamlessly with the other components of the transistor.

The transistor in today’s ubiquitous CMOS chips is the metal-oxide-semiconductor field-effect transistor, or MOSFET. It has four basic parts. There are the source and drain, which are the origin and destination point for the current; the channel that connects them; and the gate, which is essentially a valve that controls whether current flows through the channel.

In reality, there are several other ingredients inside a state-of-the-art transistor. One of the most critical is the gate insulator, which prevents the gate and channel from short-circuiting. The atoms in semiconductors such as silicon, germanium, and III-V compounds like gallium arsenide are arranged in three dimensions. There is no way to create a perfectly flat surface, so the atoms that sit on the top of the channel will have a few dangling bonds. So what you want is an insulating layer that links up with as many of those dangling bonds as possible, a process called passivation. If it isn’t done well, you’ll wind up with what could be described as an “electrically bumpy” channel, full of places where charge carriers can get temporarily trapped, lowering mobility and therefore the speed of the device.

Happily enough, nature has provided silicon with a high-quality “native” insulator that matches up well with its crystal structure: silicon dioxide (SiO2). Although today’s state-of-the-art transistors contain a more exotic insulator, they still include a thin layer of this native oxide in order to passivate the silicon channel. Because silicon and SiO2 are close structurally, a well-made layer of SiO2 can bind to 99,999 of every 100,000 dangling bonds, which is about how many there are in each square centimeter of silicon.

Gallium arsenide and other III-V materials do not have native oxides, but germanium does, which means it should, in theory, have an ideal material to passivate a germanium transistor channel. The problem is that germanium dioxide (GeO2) is weaker than SiO2, and it can absorb or even be dissolved by the water used to clean wafers during the chip manufacturing process. To make matters worse, it is hard to control the GeO2 growing process. A layer of GeO2 1 or 2 nanometers thick is needed for a state-of-the-art device, but it’s difficult to make layers thinner than about 20 nm.

Researchers have studied some alternatives. Stanford University professor Krishna Saraswat and colleagues, who spurred interest in the idea of using germanium as an alternate channel material way back in the early 2000s, first explored zirconium dioxide, which is a “high-k” insulator of the sort used in today’s high-performance transistors. Building on that team’s work, a group based at Imec, in Leuven, Belgium, examined what could be done with an ultrathin layer of silicon to improve the interface between germanium and such high-k materials.

But germanium passivation took a big step forward in 2011, when a team led by Professor Shinichi Takagi of the University of Tokyo demonstrated a way to control the growth of germanium insulator. The researchers first grew a nanometer-thick layer of another high-k insulator, aluminum oxide, on the germanium channel. Once this layer was grown, the ensemble was placed in an oxygen-filled chamber. A fraction of that oxygen passed through the aluminum oxide layer to the underlying germanium, mixing with the germanium to form a thin layer of oxide (a pairing of germanium and oxygen but technically not GeO2). In addition to helping control the growth process, the aluminum oxide acts as a protective cap for this weaker, less stable layer.

Bridge to Higher Performance: Chipmakers may one day turn to nanowire channels, like these germanium structures. Nanowires can be surrounded by a gate on all sides for added control.Photo: Heng Wu/Purdue University

S everal years ago, inspired by this finding, and facing the difficulties involved in creating pFETs with III-V channels, my group at Purdue began investigating ways to build germanium-channel transistors. We began by using germanium-on-insulator wafers, developed by the French wafer manufacturer Soitec. These wafers are standard silicon wafers topped with an electrically insulating layer underneath a 100-nm-thick layer of germanium.

With these wafers, we can build transistors in which all the standard silicon parts—the source, channel, and drain regions—are made of germanium. This is not necessarily the way a chipmaker would opt to make transistors, but it was an easy way for us to start studying the basic properties of germanium devices.

One of the first obstacles we faced was finding a way to handle resistance between the source and drain regions of the transistor and the metal electrodes that connect them to the outside world. This resistance arises from a natural electronic barrier called the Schottky barrier, which forms when a metal and a semiconductor come in contact with one another. Silicon transistors have been endlessly optimized to make this barrier as thin as possible, so that charge carriers have a very easy time tunneling across it. Getting similar behavior in a germanium device, however, requires some smart engineering. Thanks to the nuances of electronic structure, holes move from a metal into germanium quite readily, but electrons don’t. That means nFETs, which rely on the movement of electrons through the device, will have a very high resistance, wasting heat and drawing only a fraction of the current needed for fast circuits.

A standard way of thinning the barrier is to add more dopant atoms to the source and drain regions. The physics are complicated, but think of it this way: More dopant atoms mean more free charges. And with this profusion of free-ranging charge carriers, the electrical interaction between the metal electrodes and the semiconducting source and drain regions is stronger. That stronger coupling tends to promote the tunneling of charges across the barrier.

Unfortunately, this technique does not work as well in germanium as it does in silicon; the material can’t withstand the high concentration of electron-donating dopants that would be needed to thin the Schottky barrier. But what we can do is go where the dopant density is highest.

We can accomplish this by taking advantage of the fact that state-of-the-art semiconductors are doped by using ultrahigh electric fields to push ions into the material. Some of these dopant atoms stop fairly quickly; some go pretty far in. Ultimately, you wind up with a bell-curve-like distribution: The concentration of dopant atoms is highest at some depth and then tapers as you go shallower or deeper. If we recess the source and drain electrodes into the semiconducting material, we can put them in contact with the highest concentration of dopant atoms. That strategy dramatically reduces the contact-resistance problem.

Regardless of whether chipmakers ultimately use this strategy to thin the Schottky barrier in germanium, it is a useful demonstration of what the material is capable of. When we began our research, the best germanium nFETs produced currents of 100 microamperes for each micrometer of width. In 2014, at the Symposia on VLSI Technology and Circuits, in Hawaii, we reported on germanium nFETs with a record drain current of about 10 times that amount and more or less on a par with silicon—not bad for a preliminary demonstration. Some six months later, we reported the first circuits containing both germanium nFETs and pFETs, a prerequisite for making modern logic chips.

Since then, we have used germanium to build more advanced transistor designs, such as FinFETs—the current state of the art. We’ve even made germanium-based nanowire transistors, which could well replace the FinFET in coming years.

These advanced transistor designs will likely be needed for germanium to be adopted in mass manufacturing, because they offer better control over the transistor channel. Thanks to germanium’s small bandgap, a transistor with a germanium channel could require as little as a quarter of the energy that a silicon-channeled transistor needs to switch to a conducting state. This offers the potential for lower-power operation, but it also makes it easier for current to leak through the switch when it’s supposed to be off. A device with better channel control will let chipmakers take advantage of the small bandgap without compromising switching performance.

We’ve made a good start, but we’ve got more work to do. For one thing, there is a need for additional wafer-scale experiments that can demonstrate transistors with high-quality germanium channels. We also need to make refinements to the device design in order to boost the speed.

Of course, germanium is not the only option for tomorrow’s transistor channels. Researchers continue to explore III-V materials, which could be used in addition to germanium or on their own. And there is a dizzying array of other potential improvements to transistors—and the way they are wired together—on the horizon. That list includes carbon-nanotube transistors , vertically oriented switches, 3D circuits, channels made from a mix of germanium and tin, and transistors that operate by a process called quantum tunneling.

We may end up adopting several of these technologies in the coming years. But adding germanium to the channel—even initially mixed in with silicon—is a solution that will allow chipmakers to keep improving transistors in the near term. Germanium, the primordial material of the solid-state age, could be a powerful elixir for its next decade.

This article appears in the December 2016 print issue as “Switching Channels.”

Peide D. Ye is a professor of electrical and computer engineering at Purdue University in Indiana.

Your weekly selection of awesome robot videos

Evan Ackerman is a senior editor at IEEE Spectrum. Since 2007, he has written over 6,000 articles on robotics and technology. He has a degree in Martian geology and is excellent at playing bagpipes.

Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.

There's really nothing I can say to prepare you for this German music video, which features Spot for some reason.

I'm told that the music video is about how the idealized version of a forest is somewhat at odds with technology, and that bringing your fancy fleece jackets and robots along with you into nature can kind of ruin the experience. I get it. Also, that IR shot of Spot at night is suuuper creepy.

I'm going to assue that KIMLAB is not at all confused about which superhero has what equipment, and instead that Spot is cosplaying that one specific scene in Avengers: Endgame.

Dongwon Son, who's now at PhD student at Korea Advanced Institute of Science and Technology, wrote in to share this work he did at Samsung Research. Somehow, they know exactly what my desk looks like most of the time.

Rethink Robotics and Sawyer: still a thing!

The designer of one of the most destructive combat robots ever built gives some tips on how to hit harder.

Some satisfying precision syringe filling.

When you put the Lockheed Martin Missiles and Fire Control Operations Team together with Boston Dynamics, you get something not nearly as exciting as you were probably expecting.

The FIA World Rallycross series goes all-electric, paving the way for more top-tier races to switch over

Lawrence Ulrich is an award-winning auto writer and former chief auto critic at The New York Times and The Detroit Free Press.

The 2022 FIA World Rallycross Championship is a series of races in Western and Central Europe carrying on through November—in which all competing cars are now EVs. The heat pictured here took place in Hell, Norway on 14 August 2022.

With superpowered cars like the Model S Plaid, Tesla upended the stereotype of electric cars as being slow, frumpy and boring.

Now electric cars are making scorching inroads on the racing scene, faster than many folks dared to believe.

Formula E, the FIA’s offshoot of the global spectacle that is Formula 1, has gone from awkward baby steps to sprint-level speed in just eight years. Its own stepchild, the inaugural Extreme E series, is staging electric off-road battles in such exotic locales as Greenland, Senegal, and Sardinia, with five teams led by racing superstars including Chip Ganassi, Nico Rosberg and Michael Andretti. The latest is World Rallycross, a wild-and-wooly mix of off-road rallying and tarmac sections. That series unexpectedly swore off gasoline and staged its first all-electric race on 13-14 August in Hell, Norway.

Behind the spent carbon fumes, a skeptic might detect a whiff off electric opportunism on the part of some lesser racing organizations, desperate to grab a foothold, headlines or the overtaxed attention spans of sports and racing fans. (Extreme E’s events are being held with no spectators, created entirely for Web and TV broadcast). But commercial interests or hucksterism aside, few would argue that the technology itself isn’t making impressive strides.

“For each race minute, we will get 1 kilowatt-hour back from each axle.” —Björn Föerster, Porsche

A remorseless stopwatch comparison of Formula E—the current state-of-the-art in racing EVs—vs. F1 doesn’t flatter the battery brigade. At the Monaco Grand Prix circuit, where Formula E recently ran the full course for the first time, an F1 car driven by the likes of Lewis Hamilton is faster by at least 10 seconds per lap, an embarrassing eternity in racing terms. With about 1,050 horsepower, the F1 car has more than three times the power, vastly more aerodynamic downforce, and it’s lighter as well, at roughly 795 kilograms. But the picture is skewed at the top of the heap: Top F1 teams spend nearly US $500 million to run two cars for a single race season, though that includes superstar driver salaries that can top US $50 million. Still, Ferrari, Red Bull, Mercedes, and McLaren spend hundreds of millions of dollars each year on race engineering and R&D alone. The FIA caps the cost of a single F1 engine at around US $15 million, but some teams try to hide true costs to avoid penalties.

So the performance gap remains huge, but Formula E is trying to close it. In Formula E’s early years beginning in 2014, drivers actually had to jump into a second car in the pits around the event’s midpoint, because a single car couldn’t complete a full race on a single charge. Due in part to those stamina demands, these original electric, open-wheel racers brought a puny 177 kilowatts (240 horsepower), less than many family sedans, and maxed out around 225 kilometers per hour. Today’s second-generation car has a 250 kW (335-hp) electric motor on its rear axle, and can reach a solid 280 kph. But it’s the latest Formula E car that’s set to turn heads, including at the annual race on the piers of Brooklyn, with the Manhattan skyline as dramatic backdrop—which happens to be a short walk from my apartment in the Red Hook neighborhood.

The third-generation Formula E car, set to make its competitive debut in Saudi Arabia in early 2023, is smaller, lighter, faster and more environmentally friendly than any electric racer in history.

The series’ first dual-powertrain car brings a robust 600 kilowatts, (805 hp), boosted by a new 250-kW front motor. Theoretically, the cars can top 320 kph, though the series’ shorter street courses won’t leave enough room to actually reach those speeds. The car sucks up so much energy through regenerative braking—about 40 percent of its total power—that no hydraulic rear brakes are required, a racing first regardless of powertrain type. Eliminating those brakes helps trim 60 kilograms of weight, dropping the total to 840 kilograms; or 1,848 pounds, only about 100 more than an F1 monster. Ultra-fast charging at 600 kW nearly doubles the most powerful public chargers for civilian drivers. The cars feature recyclable batteries and bodies made from linen and recycled carbon fiber from last year’s cars. Hankook tires use natural rubber and recycled fiber, and will be recycled after every race.

The GT4 e Performance has 10 handling settings to take advantage of an EV’s extreme torque sensitivity, in a way that makes internal combustion engine (ICE) cars’ “torque vectoring” controls seem primitive.

With all race series increasingly focused on such sustainability, reduced emissions and solid citizenship, a potential changing of the guard was on full display at this year’s Goodwood Festival of Speed in the U.K. At Goodwood’s notorious, hay-bale-lined Hillclimb, even showroom EVs such as the Lucid Air posted times that—to use a British term—gobsmacked old-timers and the younger set alike. A car called the McMurtry Spierling—an oddball, single-seat track car—set the Hillclimb record by more than one second over the Volkswagen ID.R. That Volkswagen itself had merely set a stunning Pikes Peak hillclimb record; and clocked the second-fastest time of any car in history on the Nürburgring Nordschleife circuit (at about six minutes and five seconds). That’s second only to Porsche’s 919 Evo Hybrid, a monster built from the bones of its LeMans-winning racecar, specifically to set a record on the ‘Ring.

In realms such as the 24 Hours of LeMans, Porsche’s legendary sports cars and racing prototypes have posted more wins and championships than any manufacturer. And Porsche also chose Goodwood for the debut of its incredible Porsche Cayman GT4 ePerformance. That car, based on its showroom Cayman sports car, previews a coming generation of Porsche’s customer racing teams, proving that, as one executive told IEEE Spectrum, “The electric future can be fun.”

The Cayman e-Performance clearly shows Porsche on the cusp of building all-electric cars that can match the performance of their top production-based 911 GT racers—while using a fraction of the energy, and with zero tailpipe emissions.

The all-electric GT4 can already match the Monaco lap times of the brand’s fearsome 911 GT3 Cup racer. The curvaceous prototype now embarks on a two-year world tour to prove its viability for racing customers and series honchos alike. Björn Föerster, the GT4 ePerformance’s head of technical development, told me that “With this laboratory of ePerformance, we have just started to step into our new playground.”

The 718 Cayman GT4 ePerformance.Porsche AG

Even the engineers of Porsche, born with gasoline in their veins, are learning new tricks in this electric lab. Here’s one: Adding all-wheel drive actually makes the GT4 “lighter,” Föerster says.

“Why would you put on another powertrain to make a car lighter?” Foerster asks. “But that was the big learning we had in the beginning of development.”

This counterintuitive claim doesn’t apply to the street. But on track—where cars swing from full throttle to full, neck-straining braking at every corner entry and exit—it’s a different story. Because of that constant, brutally front-loaded braking, Porsche learned a driven front axle more than doubles energy capture via regenerative braking. Enough energy, in fact, to carry 220 fewer kilograms of battery. The result is a “lighter” electric Porsche racer with more of the quicksilver agility for which they’re known.

“For each race minute, we will get 1 kilowatt-hour (kWh) back from each axle,” Foerster says.

Over a 25-minute race, that’s 50 kWh of energy capture, which nearly matches the total 60 kWh of usable battery power in a buffered, 80- kWh pack. Add the two numbers, and the GT4 ePerformance has the 110 kWh necessary to run a full Cup race and (ideally) take the checkered flag. This plug-in GT4 can also charge from 5-to-80 percent in 15 minutes.

Internal combustion is notoriously inefficient, turning at most 40 percent of burnt fossil-fuel energy into forward motion. On the street, EVs are beginning to top 90 percent efficiency, with the latest Formula E car targeting 95 percent, aided by recovering so much otherwise wasted braking energy.

To overcome racing’s daunting issue of running out of battery juice, the GT4 ePerformance can also flexibly adjust power at will to run at maximum performance for a set race time; with no components reaching a thermal overload that requires dialing down power to cool things off.

So the GT4 can blast through a full Cup race at the sweet spot of 603 horsepower. Flip a switch, and the Porsche can spool up 1,073 horsepower—more juice than an F1 car—for roughly 18 minutes. Dial back to 402 horsepower, and stamina grows to more than 45 minutes. As with street cars and their performance settings, that previews how racers can easily optimize cars for various race series and applications.

Experts add that, because of their physical nature of electric motors and their magnetic fields, EVs are opening new worlds of system feedback and perception, with handling advantages that ICE cars can only dream of. General Motors’ street EVs can already measure electric torque every 10 microseconds, and adjust output before tires even begin to slip—a potential game-changer in terms of control strategies. The GT4 e Performance has 10 handling settings to take advantage of this newfound sensitivity, in a way that makes ICE cars’ “torque vectoring” controls seem primitive.

Of course, ICEs are notoriously inefficient, turning (at most) 40 percent of their burnt-fossil energy into forward motion. On the street, EVs are beginning to top 90 percent efficiency, with the latest Formula E car targeting 95 percent, greatly aided by recovering so much braking energy that would otherwise be wasted.

Street or track, try this for game-changer, Föerster says: The Porsche’s battery stores the equivalent of nine liters (2.4 gallons) of fuel.

“We have an electric race car with a “tank” of nine liters, that can run a half-hour on track,” Foerster says. Its traditional GT Cup race could run for three minutes, or just two laps, on the same amount of fuel. That means efficiency rises by a factor of 10. For now, he says, the only reason ICE can compete is because of the unmatched energy density of its polluting, carbon-based fuel.

“The efficiency of an electric drivetrain is already so good that you don’t have to develop anything more,” he says. “If you’re attracted by lap time, you have to go for an electric car.”

The extensive use of MIM/MOM capacitors in analog/RF designs presents parasitic extraction challenges to designers. Understanding best practices and recommended tools for extracting the complex geometries of capacitor devices, as well as the in-context coupling effects for those devices in sensitive analog/RF blocks, enables designers to accurately apply the appropriate extraction process to different parts of the design.