Features:

SALTED AWAY SILICON
by J. William Bell, Access magazine editor

Ask the stereotypical packrat--just because you replace something doesn't mean you get rid of it. The new television sits on top of the old one, a TV Guide mountain abuts the recliner, and someone might use that old bicycle someday. Room for a life amongst the clutter constantly diminishes.

As a silicon wafer is etched to become a microchip, it suffers a similar fate. The silicon is bombarded with boron, and the boron shoves its way into the bulk silicon's crystal structure. This boron, properly implanted, forms transistors and the other minute electronic devices on the chip.

But for every boron atom that is introduced, a silicon atom is knocked loose. These silicon atoms can sit relatively inert in the open spaces of the crystal structure. Unfortunately, loose silicon atoms can also combine with one another and form large "self-interstitial defects."

"Clever reactions take place at low energies in the structure, and the silicons have no reason to leave," explains NCSA material scientist Jeongnim Kim. Instead, the interstitial defects twist the bonds between surrounding silicon atoms. This influence can alter the properties of the structure--leaving you with electronic switches that don't switch or resistors that don't resist. In other words, a microchip that doesn't work as it should.

During implantation, the silicon is commonly heated, or annealed, to allow most defects to relax and improve chip performance. However, this process can introduce new complications of its own. As an alternative to implantation, the chip can also be built from the ground up, one atomic strata at a time. The cost of this approach is usually prohibitive.

The fabrication of microchips and other electronic devices goes on, and continues to improve, in the face of these difficulties. But a group of NCSA users isn't satisfied, recognizing that a clearer vision of how defects form will ultimately yield better products. Researchers from the Ohio State University, High Performance Technologies, Inc. in Aberdeen, MD, and NCSA are identifying a series of previously unknown silicon defect structures and modeling the defects' evolution from singleton silicons to large-scale havoc wreakers. The simulations are led by David Richie of HPTi, Richard Hennig and John Wilkins of Ohio State, and NCSA's Kim. They rely on NCSA's Titan Linux cluster and IBM p690 supercomputer.

Cheap but reliable

The team's models rely on a multi-level approach. Accelerated molecular dynamics simulations--which track the interaction of individual pieces of the structures at the atomic level--are completed first. They provide a catalog of possible defect configurations and the energy pathways that the pieces follow as they take on those configurations.

These calculations are the workhorse, representing a computationally "cheap but reliable way to find structures," according to NCSA's Kim. To begin, a relatively small number of silicon and boron structures are established, and the initial conditions for the surrounding environment, such as temperature, are altered slightly in order to produce a variety of defect structures.

In an effort reported in a January 2004 issue of Physical Review Letters, for example, a series of 20 structures was modeled at four different temperatures (800, 900, 1,000, and 1,100 degrees Kelvin). Twenty nanoseconds of time, in 0.5 nanosecond chunks, were modeled for each structure at each temperature. Though this may seem like a vanishingly short period, the calculation moved on a two-femtosecond timescale. A femtosecond is one quadrillionth of a second. Thus, 20 nanoseconds are represented by about 10 million individual snapshots of the forming defect structure.

"A nanosecond is a very short time period for an experiment, but it is a long time to see anything interesting during microscopic simulations. During a few pico- to nanoseconds, you get lots of boring stuff and interesting events occur very rarely," Kim says.

To cut through the boring stuff, the team's code, called OHMMS, uses a real-time data compression and feature detection method developed by David Richie. The code automatically detects transition events, in which the defects begin to form, and stable structures, or the defects in their final forms. This detection system effectively flags the beginning and end of a defect's evolution. OHMMS also incorporates a defect recognition code that sorts the structures into like kinds, further simplifying the analysis process.

"Defect configurations are often surprisingly complex," the team wrote in a 2003 report, "[and] molecular dynamics simulations performed with OHMMS may lead to thousands of stable configurations. Conventional tools require highly trained individuals to categorize these structures one at a time. [Automated], immediate, and thorough analysis of the massive data amount generated by OHMMS is the key tool for studies of highly complex structures."

OHMMS was originally created by NCSA's Kim in the late 1990s while she was a post-doctoral researcher working for John Wilkins at Ohio State. Development continues in the hands of Richie and the Wilkins research group. In 2003 alone, the team used more than 200,000 hours of computing time at NCSA. They also ran models at the Ohio Supercomputing Center and the U.S. Department of Energy's National Energy Research Scientific Computing Center.

More than molecular dynamics

Once interesting structures are identified, their geometries are confirmed and refined using density functional theory, which is more precise than the molecular dynamics simulations, and a code called VASP. These calculations also determine characteristics of the structures, such as energies and rates at which the structures form. These features are key to chip designers and other applied researchers.

"Device design isn't at the atomistic level where molecular dynamics simulations take place," Kim says.

The team's work, especially that discussed in the recent Physical Review Letters, reveals not only the higher-order features that researchers need but also some of the flaws in the assumptions that researchers frequently make.

"The complexity revealed in our study is striking compared to the picture uncovered in previous efforts. A good example is the extended chainlike structure of the tri-interstitial ground state, which defies the conventional assumptions of compactness and simple symmetry previously used to investigate these systems," HPTi's Richie says.

In fact, the team has uncovered four structures that were previously unknown. They suspect that these large, complex interstitial structures capture other interstitials at their terminal ends. The daisy-chaining of interstitial silicon atoms that results might explain the observed extended, rod-shaped defects that cause problems for chip designers.

"The practical aim of these and most simulation modeling is to speed the selection of the most likely approaches for detailed experiments that lead to development and manufacture of better chips," says Wilkins, a physics professor at Ohio State. "Of course, we hope improved atomic understanding will also speed up future modeling efforts on new problems."