"The goal of our project," says Bernholc, "is to understand in atomic detail the growth modes of semiconductors and to identify the conditions for high quality layer-by-layer growth." Step-flow growth is the most common growth process in molecular beam epitaxy (MBE), a relatively new method of producing chip-quality silicon crystals. In MBE, a gaseous beam deposits a one or two atom layer of semiconductor material on a base layer or "substrate," with the inter-atomic spaces of the substrate forming an egg-carton-like template for the adatoms.
To simulate these processes, Bernholc uses a powerful computational approach called quantum molecular dynamics (QMD). His code, adapted from a method developed by European physicists Roberto Car and Michele Parinello, is a truly quantum approach that allows the electrons and atoms to move freely with time, so that the atom-atom interactions are computed ab initio ("from the beginning"), directly from the electronic forces. Bernholc' C90 adaptation of QMD runs at 693 million calculations a second (Mflops) on a single C90 processor, one of the fastest codes running on this machine.
Taking advantage of "friendly user" testing on Pittsburgh's new C90 in early 1993, Bernholc carried out a series of four-processor runs (2.3 Gflops) that resulted in the first ab initio calculations giving the structure of steps. "We compared the calculated atomic and electronic structure," says Bernholc, "to atomic resolution images from scanning tunneling microscopy (STM). We verified that we did indeed reproduce what was observed experimentally, and our results provide a better interpretation of the experiments than others that had been proposed. So we now know what the step structures are with some precision."
go back to the main page