Structural models of amorphous silcon nitride, a-Si3N4. consisting of 112-448 atoms were studied using density functional methods. We used continuous random alterating networks with well-defined topology for the respersentation of chemical order in the material as theoretical precursors. The models were optimized within the DFT framework and compared them to one “ab inito derived” model obtained from quenching a hypothetical melt. The strong chemical order is maintained in the network models even after Car-Parrinello molecular dynamic (CPMD) simulations at elevated temperatures for several pico-seconds, In contrast, the “ab initio derived” model exhibits n-N bonds.
The optimized strutures of Si3N4 have between 2.6 and 3.2 g/vm3 and comprise few topological defects only. The dominant defects are ever over-coordinated Si and N atoms and the 2-connected is averaged over a dozen modles is, averaged over a dozen models, about 1%. Some models are even free of three-connected Si. The calculated bulk moduli decrease with decreasing density of the a-Si3N4 model. We furthermore investigated the properties of the material ater alloying elements such as H and O, espically their capacity to reduces interal strain.