ITR: Billion-atom Multiscale Simulations on a Grid Priya Vashishta, - - PowerPoint PPT Presentation

ITR: Billion-atom Multiscale Simulations on a Grid

Priya Vashishta, Rajiv K. Kalia & Aiichiro Nakano

Concurrent Computing Laboratory for Materials Simulations

• Dept. of Physics & Dept. of Computer Science, Louisiana State Univ.

Email: {priyav, kalia, nakano}@bit.csc.lsu.edu URL: www.cclms.lsu.edu

September 1, 2002: Collaboratory for Multiscale Simulations Departments of Materials Science & Engineering, Physics, Computer Science, and Biomedical Engineering University of Southern California

CCLMS CCLMS CCLMS

NSF Division of Materials Research Computational Materials Theory Program Review Program Managers: Dr. Bruce Taggart & Dr. Daryl Hess Organizers: Dr. Duane Johnson & Dr. Jeongnim Kim June 20, 2002, Urbana, IL

Outline

• 1. Multiscale simulation of lattice-mismatched

nanopixels & nanomesas

• 2. Multimillion-atom molecular dynamics

simulation of semiconductor nanoparticles

• 3. GRID computing with latency tolerant

algorithms

Concurrent Computing Laboratory for Materials Simulations (CCLMS)

Faculty: Priya Vashishta, Rajiv Kalia and Aiichiro Nakano, Postdocs: Paulo Branicio, Bijaya Karki, Hideaki Kikuchi, Sanjay Kodiyalam, Maxim Makeev, Elefterios Lidorikis Dual-Degree Graduate Students: Gürcan Aral, Jabari Lee, Xinlian Liu, Zhen Lu, Brent Neal, Cindy Rountree, Ashish Sharma, Satyavani Vemparala, Weiqiang Wang, Cheng Zhang Undergraduate Students: DeAndra Hayes (Xavier), Paul Miller, Wei Zhao Visitors: Simon de Leeuw (Delft, The Netherlands), Ingvar Ebbsjö (Uppsala, Sweden), Hiroshi Iyetomi (Niigata, Japan), Shuji Ogata (Yamaguchi, Japan), José Rino (São Carlos, Brazil), Fuyuki Shimojo (Hiroshima, Japan) Systems Manager: Monika Lee Coordinator: Jade Ethridge

1,024 CPU System being installed at LSU under the auspices of Louisiana IT initiative.

Hybrid FE/MD Algorithm

• FE nodes & MD atoms coincide in the handshake region
• Additive hybridization

[0 1 1] [1 1 1]

_

HS

_

[1 1 1] [2 1 1]

Si/Si3N4 nanopixel

Si(111)/Si3N4(0001) Nanopixel

0.0 0.2 0.4 0.6 [Å]

r

z (top to bottom) [nm] 5 10 15 20 Displacement [Å]

• 0.5

Interface Hybrid full MD

r z y

0.5

Int. HS

Displacement from equilibrium positions

Si Si3N4

Hybrid FE/MD Full MD HS

70 nm

Stress Domains in Si3N4/Si Nanopixels

Stress domains in Si due to an amorphous Si3N4 film

• 2GPa

2GPa

Stress well in Si with a crystalline Si3N4 film due to lattice mismatch

Si Si3N4

• Epitaxially Grown Quantum Dots
• A. Madhukar (USC)

Substrate-encoded size-reducing epitaxy

GaAs (001) substrate; <100> square mesas

10nm 101 GaAs AlGaAs QD QD 001 AlGaAs

Lattice-mismatched Growth of Epitaxial Quantum Dots on Patterned Substrates

InAs island formation on a flat GaAs(001) substrate >1.6 monolayer deposition

• A. Madhukar (USC)

Self-limiting growth of 12 ML InAs on a patterned substrate

10 µm

GaAs mesa substrate [001] [010]

InAs GaAs GaAs 20nm

InAs delivery: 24ML, Base: 75nm Height: 11±1 ML

Strain relaxation suppresses 2D 3D transformation on a patterned substrate <100nm

30 25 20 15 10 5 2 4 6 8 10 12 14

InAs deposition (ML) InAs thickness (ML)

MESA SIZE ~ 750 Å

cr = 1.6ML cr = 12ML

GaAs/InAs: 7.2% lattice mismatch

Validation of Interatomic Potentials—GaAs

X-ray static structure factor Phonon dispersion High-pressure phase transition

Si3N4

amorphous

SiC

10 20 30 40 10 20 30 40

Fr equency (m eV)

Γ Γ K X L X W L Experiment (Strauch & Dorner, '90) Theory

2.3 2.4 2.5 10 20 Ga-As Distance (Å) Pressure (GPa)

MD

10 20

Expt.

[Besson et al., '91]

1 2 2 4 6 8 10 MD Experiment q (Å-1)

Amorphous GaAs

[Udron et al., ‘91]

Atomistic Stress in InAs/GaAs Square Mesa

• In-plane lattice constant in InAs
• verlayers exceeds the bulk value

at 12 ML self-limiting thickness

Vertical displacement in the first As layer above the interface

• Domain formation in larger

mesas critical lateral size for 3D island formation

Colloidal Quantum Dots, Rods & Tetrapods

Applications

• LED, display
• Biological labeling
• Pressure synthesis
• f novel materials

High-pressure structural transformation in a GaAs nanocrystal

Collaborator: Paul Alivisatos (Chemistry, Berkeley)

[from Bawendi’s group at MIT]

17.5 GPa

Multiple domains

22.5 GPa

30 Å

Nucleation at surface

Multiple Domains in a GaAs Nanocrystal

Nucleation & growth of high-pressure-phase domains

Domain Fluctuations

Third domain’s growth fluctuates with time

Shape Dependence of Transformation

Transformation is sensitive to the initial shape

Spherical Faceted Faceted Multiple domains Single domain Multiple domains

GRID Computing for a One Billion Atom One Micron Nanopixel

• One billion atom simulation for
• ne micron (1000nm) nanopixel.

The simulation will be split in two parts - top 200 million atoms

• n a 256 CPU system and the

remaining 800 million on a 1,024 CPU system for GRID

• computing. In GRID computing,

quality of service (QoS) and latency issues serious, but not killers.

• Main objective is to confirm the

nature of hexagonal pattern of stress domains at the interface. Si/Si3N4 nanopixel

GRID Computing for an Ensemble of 64 Nanoclusters

• Nanocrystals in Lennard-Jones liquid
• Isothermal-Isobaric simulations
• Nanocrystal: 20-60 Å
• Pressure: 2.5-25 Gpa
• 90% of the particles constitute

pressure medium.

• 8 to 16 processors optimum for
• ne nanocrystal.
• In GRID computing, QoS and

latency issues not serious, syncronization needed at pressure change only.

Access Grid Technology for Education and Training of Underrepresented Groups

Access Grid

Un de rg ra du a te Edu ca tio n & Tra in in g

Co m pu ta tio n a l S cie n ce Wo rks h o p fo r Un de rre pre s e n te d Gro u ps

• 1 9 pa rticipa n ts fro m 1 1 in s titu tio n s —

Ha m pto n , Cla rk-Atla n ta , Mo re h o u s e , Ja cks o n S ta te , Mis s is s ippi S ta te , Te x a s S o u th e rn , Un iv . o f Te x a s – – Pa n Am e rica n , Xa vie r, Gra m blin g , S o u th e rn & Un iv . o f Lo u is ia n a in Mo n ro e

• Activ itie s : Co n s tru ctio n o f a PC clu s te r fro m
• ff-th e -s h e lf co m po n e n ts & u s in g th is pa ra lle l

m a ch in e fo r a lg o rith m ic a n d s im u la tio n e x e rcis e s .

Summary

CCLMS CCLMS CCLMS

• 1. Multiscale simulation of lattice-mismatched

nanopixels & nanomesas

• 2. Multimillion-atom molecular dynamics

simulation of semiconductor nanoparticles

• 3. GRID computing with latency tolerant

algorithms

• 4. Educational and training activities using

access grid

Future: Biologically-inspired Nanostructures

NASA Information Power Grid Collaborators:

• A. Madhukar (USC)

Paul Alivisatos (Berkeley) Collaborator: Jonathan Trent (NASA)

Protein-nanotube-based nanostructures Bio-inspired self-assembly of epitaxical & nanoparticle quantum dots