Academics have high hopes for ferroelectric materials. Adding a single layer of these materials, which have unusual electrical properties, to today’s transistors could radically decrease the power consumption of chips.
But as engineers presented the latest research on ferroelectrics at the IEEE International Electron Devices Meeting (IEDM), in San Francisco in December, the mood in the room fluctuated between excitement and doubt.
Many in industry are skeptical about the benefits of ferroelectrics. Still, the IEDM meeting made it clear that semiconductor companies are now paying attention. Researchers from GlobalFoundries presented data on the performance of ferroelectric-frosted transistors made using their 14-nanometer manufacturing technology.
The magic of ferroelectrics is their potential to free engineers from the “Boltzmann tyranny,” named for Ludwig Boltzmann, who did foundational work in thermodynamics, says Aaron Franklin, an electrical engineer at Duke University, in North Carolina. To boost the current through a traditional field-effect transistor by a factor of 10 at room temperature, engineers must apply at least 60 millivolts. This sets a lower limit on transistors’ power consumption, which engineers dream of limbo-ing under. Getting a strong signal at lower voltages would save power and enable longer battery lives.
Operating at lower voltages will be necessary for engineers to further shrink transistors. As they get smaller, they do a worse job of shedding heat. Shrink them too much and the overheating transistors will melt. Running transistors at lower voltages keeps temperatures in check.
Ferroelectric materials are defined by their tendency to experience profound electrical polarization in response to relatively puny electrical fields. Put a voltage across a ferroelectric film and charges—sometimes charged atoms—within it will quickly move from one side to the other. “You put half a volt on it, and because of the polarization it’s like applying a whole volt,” says Franklin.
Most of the ways around the Boltzmann tyranny require ditching traditional transistor designs altogether. Compared with those, the ferroelectric approach should be pretty straightforward. All the industry needs to do is add a ferroelectric layer. “It’s such a simple modification,” says Franklin, who cochaired the IEDM session.
This idea was first proposed in 2008. That year, Sayeef Salahuddin, now a professor at the University of California, Berkeley, and his Purdue University Ph.D. advisor Supriyo Datta published an influential paper showing that replacing a traditional insulator with a ferroelectric one should lead to power savings.
The idea didn’t gain much traction at the time. “It seemed crazy because we only knew of ferroelectrics that contained lead and other nasty materials, and the ferroelectric layer had to be very thick,” says Franklin. More recently, researchers have figured out how to encourage friendlier materials, such as hafnium dioxide, already used in chip components, to act as ferroelectrics. Instead of using these materials to replace insulators, as Datta had proposed, engineers typically layer them on top of existing insulators.
Even so, problems remain. The strange behavior of electrical charges in ferroelectric materials slows things down—it takes time for charges to relocate. Some researchers have predicted that transistors built with ferroelectrics will never exceed 100 megahertz. And some think that building these devices will require very thick layers of ferroelectrics—too thick to be practical.
At IEDM, after a presenter described how ferroelectrics could help engineers scale chips down to 2 nm, an audience member pointed out that the proposed designs did not leave enough physical space for a ferroelectric layer thick enough to provide the predicted benefits. The presenter, looking a bit flummoxed, replied that the work was theoretical.
Zoran Krivokapic, an electrical engineer who leads GlobalFoundries’ ferroelectrics project, says there are misunderstandings about what ferroelectrics can do. Data from experimental ferroelectric devices tend to be “all over the place,” he says. If researchers don’t take careful note of the buildup of charges in the ferroelectric and the semiconductor, making sure they are very closely attuned—a property called capacitance matching—
the devices will not work. Krivokapic says poorly fashioned devices have produced poor results, and caused engineers to underestimate the potential of ferroelectrics.
To overcome the speed problem, the GlobalFoundries team chose a ferroelectric material that does not require ions or atoms to relocate. In their experimental 14-nm transistors, Krivokapic says, clouds of electrons around silicon-doped hafnium dioxide experience the polarization. And electrons can move fast: Ring oscillators made with these transistors can switch at the same frequency as those made with the usual recipe, yet they require just 54 mV to achieve a tenfold increase in the current. Franklin says it’s difficult to pin down a theoretical minimum, since designs vary. However, ferroelectric devices typically don’t go below 30 mV—although some researchers have reported devices that switch at 5 mV.
GlobalFoundries’ devices require a 3- to 8-nm-thick layer of ferroelectric material, which is still relatively thick. But researchers are excited about this first practical demonstration. “This is not something from an academic lab, where you can argue that it’s not CMOS compatible,” says electrical engineer Deji Akinwande, of the University of Texas at Austin. “This field seems to be rapidly maturing to the point where even the big companies are working on it.”
These devices are not yet ready for production, says Michael Chudzik, a senior director at semiconductor equipment maker Applied Materials, but they do show that ferroelectrics are under serious consideration. In the semiconductor industry, he says, “you have to shoot ahead to actually hit it.”
This article appears in the March 2018 print magazine as “Ferroelectric Transistors: The Ultralow-Power Solution?”