[an error occurred while processing this directive]

NERSC 3 Greenbook

Next: First Principles Simulation of Up: Basic Energy Sciences Previous: Basic Energy Sciences

Multi-scale Materials Simulations on Massively Parallel Machines

Martina Bachlechner, Rajiv K. Kalia, Aiichiro Nakano, Andrey Omeltchenko, Kenji Tsuruta, and Priya Vashishta
Concurrent Computing Laboratory for Materials Simulations,
Louisiana State University.

In recent years we have witnessed rapid progress in large-scale atomistic simulations, highly efficient algorithms for massively parallel machines, and immersive and interactive virtual environments for analyzing and controlling simulations in real time. As a result of these advances, simulations can now reliably predict properties of materials in advance of fabrication and are thus capable of providing vital input to experimental efforts.

In the past few years, the LSU group has carried out multi-million particle molecular dynamics (MD) simulations to investigate properties and processes in novel ceramic materials, atomic-level stresses in silicon/silicon nitride nanopixels, hypervelocity impact damage in thermal-barrier coatings, and nanoindentation in ceramics. A couple of these simulations are presented below.

Dynamic Fracture in Nanophase Ceramics

Advanced ceramics such as Si₃N₄, SiC, SiO₂, and Al₂O₃ are highly desirable materials for applications requiring extreme operating conditions. Light weight, elevated melting temperatures, high strengths, and wear and corrosion resistance make them very attractive for applications in aeronautics, automotive, electronics, and advanced manufacturing industries. The only serious drawback of ceramics is that they are brittle at low to moderately high temperatures.

In recent years a great deal of progress has been made in the synthesis of ceramics that are much more ductile (and metals that are harder) than conventional coarse-grained materials. These so called nanophase materials are fabricated by in-situ consolidation of nanometer size clusters. Despite a great deal of research, many perplexing questions concerning nanophase ceramics remain unanswered. Experiments have yet to provide information regarding the morphology and dynamics of cracks or why nanophase materials fracture much more gracefully than crystalline ceramics with the same chemical constituents.

Using large-scale MD simulations, we have investigated the morphological and dynamic aspects of fracture in nanophase Si₃N₄. We find that the critical strain at which the nanophase system fractures is enormous compared to the strain crystalline Si₃N₄ can sustain. This is due to: i) plastic deformation in interfacial regions; ii) deflection and arrest of cracks by nanoclusters; and iii) multiple crack branching. None of these mechanisms are operative in the crystalline system. This demonstrates the dramatic effect of nanostructures on fracture in nanophase ceramics.

Simulation of Internal Stresses at Semiconductor/Dielectric Interfaces

In silicon and gallium arsenide semiconductor technology, the use of SiO₂ and Si₃N₄ thin films as dielectrics, encapsulation layers or mask layer is ubiquitous. One of the main issues concerning as-grown processed semiconductor devices is stress/strain distributions in dielectric layers and semiconductor structures as a function of the pixel size. Device pixellation and isolation create edge and corner stresses that are highly inhomogeneous, evolve with the device processing step (and hence each processing step can be impacted by the preceding steps), and in the final configuration, impact the electronic behavior and long-term reliability of the device.

Historically, stress related problems in semiconductor (mostly Si) technology have been dealt with in terms of the classical linear elasticity theory of continuum bodies with the aid of the finite-element method (FEM). However, as the minimum feature size in Si integrated circuit (IC) technology approaches the nanometer regime, the applicability of the FEM to examining stress (particularly at the edges and corners) becomes highly questionable. The impact of the discrete atomic nature of the system on such stresses and of such stresses, in turn, on stabilizing local atomic configuration, needs to be self-consistently examined.

Recently we have performed 10-million-atom MD simulations to investigate the atomic level stress/strain distributions in Si/Si/Si₃N₄ nanopixels, see Figures 7 and 8. These simulations provide insight into the nature and degree of stresses in mesas as a function of temperature, density and stoichiometry of the dielectric films.

  

Figure 7: Results of a 30-million-atom MD simulation for shear stress distributions in SiC at 300 K (top) and 1500 K (bottom).

  

Figure 8: Stress distributions in (a) crystalline and (b) amorphous Si/Si/Si₃N₄ nanopixels.

Multi-scale Algorithms for Materials Simulations

Large-scale simulations discussed above have been performed with:

• Space-time multi-resolution MD based on the O(N) fast multipole method and a multiple time-scale scheme;
• Fuzzy-clustering approach to hierarchical dynamics including quaternion-based rigid-body dynamics, symplectic implicit integration, and normal-mode analysis;
• Wavelet-based adaptive load balancing scheme using curvilinear-coordinate transformation;
• Multilevel preconditioned conjugate gradient method for variable-charge MD.

Efforts are underway to develop a novel multi-scale simulation approach which will combine the electronic structure, MD, and finite-element methods (Fig. 9) to study the mechanical behavior of materials at mesoscopic length scales.

  

Figure 9: Schematic of a multi-scale simulation approach which combines electronic structure method with molecular dynamics (MD) to describe properly bond breakage and bond formation phenomena. The MD, in turn, is embedded in the finite-element method to take into account nonlinearities in the system.

NERSC 3 Greenbook

Next: First Principles Simulation of Up: Basic Energy Sciences Previous: Basic Energy Sciences