Nanotech and supercomputing
team for advanced transistors
Posted January 31, 2012
A view inside a nanowire. Click image to enlarge and for more information.
Intel released a novel transistor last year, the Tri-Gate, which replaced a two-dimensional flat planar gate with a 3-D silicon fin. This “reinvention of the transistor,” Intel announced, would increase performance up to 37 percent at low voltage. At 22 nanometers, it was 10 nm smaller than the current state-of-the-art transistor.
Such advancements keep Intel on the pace of improvement Moore’s law predicted and promises a host of applications, from computational devices to arrays of environmental and medical sensors.
This latest advance was made possible, in part, by the Network for Computational Nanotechnology (NCN). Started almost 10 years ago at Purdue University under the direction of Mark Lundstrom and Gerhard Klimeck, NCN is home to nanoHUB (nanohub.org). (See sidebar, “Government-industry-university collaboration: a recipe for success.”)
The Klimeck group has been building a suite of nanoelectronic modeling tools called NEMO/OMEN. They’ve used the tools to model various devices, including finFETs, the fin-featuring transistors. (FET: field effect transistor.) The OMEN code has demonstrated almost perfect parallel scaling while exceeding performance of a petaflops, or quadrillion computations a second, on a high-performance computer.
Klimeck models quantum mechanical electron flow in nanoscale devices’ transistors and semiconductors. The models have grown from portraying a few hundred atoms when he started his research at Texas Instruments in 1994 to 100 million atoms today.
The Department of Energy’s INCITE (Innovative and Novel Computational Impact on Theory and Experiment) program has granted Klimeck 15 million processor hours on Jaguar, the Cray XT5 at Oak Ridge National Laboratory. Jaguar is one of only a few computers that operate at the petascale. OMEN, for example, has scaled to a whopping 221,400 of Jaguar’s 224,000 processor cores with a sustained performance above 1 petaflops.