07.15.10 - Permalink
by J. William Bell and John Toon
Electrons flowing through semiconductors are our electronic world's lifeblood. They are pumping through your cell phone, TV, and computer, the airplane you took last week and the car you drove this morning, and, of course, supercomputers. As a result of that importance, researchers are on the lookout for new materials that might make that flow of electrons more efficient or allow for novel designs and features in electronics. They're in search of a "post-silicon" world where not all electronics are based on this now-ubiquitous material.
A team from Georgia Tech uses NCSA's Cobalt shared-memory supercomputer to study a possible candidategraphene, a thin lattice of carbon atoms just one atom thickand the metal contacts that would connect it to other parts of electronic devices. They've also worked closely with NCSA staff to improve their code and run better across multiple processors.
"Lots of people are very excited to use graphene in electronics because electron mobility is very high," says Salvador Barraza-Lopez, a postdoc on the Georgia Tech team.
Recent results were published in February 2010 in Physical Review Letters.
With large-scale, first-principles calculations at NCSA and the Texas Advanced Computing Center, the Georgia Tech research team conducted detailed atomic-level calculations of aluminum contacts grown on graphene.
"Graphene devices will have to communicate with the external world, and that means we will have to fabricate contacts to transport current and data," says Mei-Yin Chou, a professor and department chair in the School of Physics at Georgia Tech. "When they put metal contacts onto graphene to measure transport properties, researchers and device designers need to know that they may not be measuring the intrinsic properties of pristine graphene. Coupling between the contacts and the material must be taken into account."
The calculations studied two contacts up to 14 nanometers apart, with graphene suspended between them. In their calculations, the researchers allowed the aluminum to grow as it would in the real world, then studied how electron transfer was induced in the area surrounding the contacts.
"We built atomistically resolved contacts, and by doing that, we solved this problem at the atomic level and tried to do everything consistent with quantum mechanics," Chou says.
Because metals typically have excess electrons, physically attaching the contacts to graphene causes a charge transfer from the metal. Charge begins to be transferred as soon as the contracts are constructed, but ultimately the two materials reach equilibrium.
The study showed that charge transfer at the leads and into the freestanding section of the material creates an electron-hole asymmetry in the conductance. For leads that are sufficiently long, the effect creates two conductance minima for the suspended and clamped regions of the graphene, according to Barraza-Lopez. This means that the energy range where there is high electron mobility is larger.
The researchers modeled aluminum, but believe their results will apply to other metals such as copper and gold that do not form chemical bonds with graphene.
The team's 2010 publication is, according to Barraza-Lopez, the culmination of work that started two years ago as part of NCSA's Fellows program. While a postdoc at Virginia Tech, Barraza-Lopez and his advisor Kyungwha Park spent the summer at NCSA. They were using Smeagol, a community simulation code, to model the transport of electrons between two gold contacts.
"We had problems with memory allocation, so the size of the systems that we could model had constraints," says Barraza- Lopez. Collaborating with NCSA's Greg Bauer and Jing Li, "we were successful in parallelizing the code further, making memory not so much of an issue anymore."
At the time, they were modeling a single molecule connecting the two contacts. Thanks to the parallelization work with NCSA and a shared-memory computing system like Cobalt, however, Barraza-Lopez and the rest of the team at Georgia Tech are now looking at much larger systems. The model in Physical Review Letters, for example, considered devices with as many as 464 atoms and 5,600 electronic orbitals. The simulation was able to capture the effect of device widths as large as 100 nanometers.
"We've reduced the memory requirements tremendously, but we still need the large memory pool that a shared memory computer offers," Barraza-Lopez says. With Cobalt and hopefully with the new Ember shared memory machine when it comes online, "We're not looking at a tiny flake of graphene. We're at graphene with dimensions and real-world properties that are relevant to people doing experiments."
Parts of this story were first published by the Georgia Tech Research News & Publications Office.
This project was funded by the Department of Energy.