Researchers rely on NCSA's Mercury to overcome a major barrier in shrinking capacitors for microelectronics use.

A supercomputer in your pocket? Probably not. But thanks to today's high-performance computing, the electronic devices you'll carry in your pocket in future will most likely be smaller and have larger, more powerful memories than the ones you're toting around now.

That's because advances in supercomputing combined with advances in science enabled a California research team to identify a solution to a key challenge faced in miniaturizing devices—shrinking the size of the capacitors. A capacitor in its simplest form consists of two conducting plates separated by an insulating layer called a dielectric. In electronic circuits they have many uses, such as providing barriers to prevent current leakage, storing charge, or storing data in memory applications.

Capacitors are often the largest component in the circuits found in microelectronics. But reducing capacitor size for use in ever-smaller electronic devices is hampered by depolarization effects arising from the electrode-film interfaces. This results in reduced interface capacitance—a "dead layer" at the surface that reduces overall performance.

Experimental colleagues brought the dead layer problem to the attention of computational scientists Massimiliano Stengel and Nicola Spaldin in the Materials Department at the University of California, Santa Barbara, and David Vanderbilt in the Department of Physics and Astronomy at Rutgers, The State University of New Jersey. With increasingly sophisticated methods for growing films and interfaces, "new experimental data was causing researchers to revisit long-held assumptions about the nature of this dead layer problem," Spaldin says. Her team "got involved because it seemed like an intriguing problem that we could directly contribute to—what are the fundamental limitations associated with such interfaces, and how much of the experimental limitations are due to defects or imperfect growth? In a calculation we can make a perfect interface without any defects and see what the intrinsic limit is for the behavior."

The team used first principles density functional theory calculations to prove that an intrinsic dead layer actually exists, then to study its origin. They calculated specifically the response of all of the atoms in a model capacitor to an applied electric field, then analyzed the dielectric response—the amount of polarization per applied electric field—locally across the structure. "In late 2006 we saw our first computer images of a 'dead layer': A huge suppression of the dielectric response at an ideal interface," says Spaldin. "While we could see what was causing it, and that it really was intrinsic to the system, it's taken us since then to figure out how to fix it."

Computer chemistry

What the team did to discover the fix, says Spaldin, was "almost like a chemistry experiment" except it was conducted via supercomputer. The project used over 100,000 hours of computing time, on NCSA's Mercury and also at the San Diego Supercomputer Center. Without current compute resources, she says, they would not have been able to take on this research.

"The ability to do this is something that is fairly recent, both in terms of the availability of the tools and techniques and the availability of computer resources," says Spaldin. "Even with the most sophisticated algorithms, this was still a big number crunch."

The team made repeated calculations with varied combinations of different metals and different ferroelectrics (materials with a spontaneous electric polarization that can be reversed with an electric field). The calculations used a density-functional theory code that the team customized with some methods extensions in order to constrain the electrical boundary conditions, allowing them to apply an electric field to the system or to constrain the internal polarization. Analyzing the results, the first thing the team noticed was surprising: a strong dependence on the details of the chemistry of the metal/ferroelectric interface. They then spent nearly another year exploring why particular combinations of chemistry responded as they did.

The next step

The team determined computationally that using barium titanate for the ferroelectric and platinum for the metal essentially reversed the effect of the dead layer, increasing the overall capacitance. The next step is for an experimental group to validate the team's work in the laboratory.

"It used to be thought that there was really a fundamental size limit below which you couldn't go, because your ferroelectric would stop being ferroelectric because of this dead layer and would just give up," Spaldin says. "What we've found is that with the barium titanate/platinum combination—even for just one layer, just one unit cell—the barium titanate stays ferroelectric."

If the work holds up experimentally, the potential then exists to make thin-film ferroelectric memory chips that are smaller than anything currently on the market and capable of holding significantly more data. This could make ferroelectric memory more competitive with magnetic memory. The team hopes their results motivate experimental scientists to further explore controlling the structure and termination of perovskite/simple metal interfaces, yielding additional discoveries in engineering the electrical properties of thin-film devices.

"There's so much else going on in integrated circuits that may end up influencing whether our work gets adopted or not, rather than the basic physics of the interface," Spaldin says, noting that it often takes years for new discoveries to be successfully incorporated into products.

Even if their capacitor isn't adopted by industry, the team established some major theoretical groundwork, which was published in Nature earlier this year. In particular, they are very excited about the capability they were forced to develop in order to answer the dead layer question. They believe the capability of controlling the electric field or defining the electrostatic boundary conditions in a density-functional theory calculation is something that is going to be widely applicable to a variety of research problems.

"Any time you want to model a material in a realistic device situation, you need to include metallic electrodes, which means you will have an interface between the metal and your material. Then in order to be able to make a physically meaningful calculation you need to be able to apply electric fields and/or control the electrical boundary conditions within the calculation. Our development of the tools to do this could turn out to be the most significant outcome of this work. We now have a tool—the theory community now has a tool—that can be applied to other combinations of materials where there are open questions and problems to solve."

Moving forward

The next direction for the team is also on the theoretical front. They'll explore combining the application of magnetic fields and electric fields in their calculations, allowing them to explore a whole new capability of what is called magneto-electric response.

"We'll be looking at how electrical properties change with magnetic fields and vice versa, as well as whether we can get coupled properties by applying both types of fields simultaneously," says Spaldin. "Now that we have this capability, that's the next frontier we're playing with."