247 RESULTS FROM FIRST PRINCIPLES MOLECULAR DYNAMICS SIMULATIONS ON a-Si Peter A. Fedders* and David D. Drabold** *Washington University, Department of Physics, Campus Box 1105, St. Louis, MO 63130, USA. "**Department of Physics, University of Notre Dame, Notre Dame, IN 46556, USA ABSTRACT We report on some recent results of first principles molecular dynamics simulations on a- Si. These simulations yield interesting results that challenge the standard beliefs about what constitutes a defect, how light induced defects arise, and about the origins of conduction band tailing. 1. INTRODUCTION We wish to report on the results of some molecular dynamics simulations performed over the the past year. These simulations were performed with the first principles ab initio code developed by Sankey, Niklewski, Drabold and others [1]. The simulations were performed on periodic supercells that contained 63 and 216 atoms. The details of these cells are contained in the literature [2,3]. In performing a molecular dynamics simulation, there is always a trade-off between the accuracy of the forces and the size of the supercell that one can use. In our simulations we have used a computer code that gives very accurate forces, but the size of our cells was a rather modest 63 or 216 atoms. We wish to note, that at least for studying defects, the accuracy of the forces is extremely important. In earlier work we showed that the forces generated by various angle-dependent potentials (or classical forces) are inaccurate to the point of yielding only very questionable results [3]. The situation is illustrated in Fig 1. Typically the size of the force of one Si atom on a neighbor is 5 eV/A while the change in the force on a Si atom when one changes the number of electrons on a dangling bond, is only about 0.1-0.2 eV/A. Since the charge state of a defect is of paramount importance, it is vital the the forces be at least this accurate. In fact, in Fig 1, we have assigned an error of about 0.5eV/A as the maximum acceptible error. As shown in other work [3], typical angular dependent force potentials give errors of between 1 and 10 eV/A. This includes the forces derived by Stillinger and Weber [4], by Biswas and Hammond [5], and by Carlsson, Fedders, and Myles [6]. Furthermore, no such scheme is able to assign different forces to defect states depending on their charge configuration. Thus classical codes based on angular-dependent forces are too inaccurate to simulate the behavior of defects reliably and cannot include vital quantum mechanical effects such as doping, changes in the Fermi level, and the charge state of a defect. Fig 1 also shows that the errors in forces from a first princi- ples calculation using only 32 atoms and one k-point in the Brillouin Zone is about 0.3 eV/A. Finally, a cell with 216 atoms and using one k-point or a cell with 63 atoms and using 4 k- points gives errors of about 0.03 ev/A. These are the supercells used for the simulations described in this paper. Mat. Res. Soc. Symp. Proc. Vol. 219. ©1991 Materials Research Society