The National Nanotechnology Infrastructure Network: Synergy between Experiment and Computation

1
Cyberinfrastructure in Materials Science NSF
08/06
The National Nanotechnology Infrastructure
Network Synergy between Experiment and
Computation
Michael Stopa
Harvard University
NSEC
Nano by Numbers
2
Many Nanoscale Problems

• are resistant to solution by established
  theoretical methods and therefore require
  computational tools that go beyond packages that
  are on the shelf.
• Example transport through single molecules. Must
  include electronic structure of small molecule,
  many body correlation, image effect of connecting
  leads and electro-mechanical effects of
  deformation coupled to electron transport. Not
  easy.

J. Chen et al., Science 286, 1550 (1999).
NSEC
3
Theoretical treatments (examples)
J. Taylor et. al. cond-mat/0212191.
Non-equilibrium Greens function (TranSiesta
package). Self-consistent calculation of current
for finite bias. Ionic structure is not
calculated self-consistently but is assumed to be
given. Find no influence of the side-group unless
neighboring molecule interaction is included.
C. Majumder et. al. Jpn. J. Appl. Phys. 41, 2770,
(2002). HF-DFT calculation including geometrical
optimization. Threshold voltage calculated
assuming (a) current proceeds through extended
states and (b) no voltage drop in molecule. Also
calculated electronic structure of anion
(molecule plus electron).
M. Hettler et. al. cond-mat/0207483. Photon
assisted tunneling through pi-orbitals of
benzene. Model calculation single pi orbital at
each carbon atom.. Concludes that current
blocking results from population of a state which
cannot decay, due to symmetry, to the drain
lead.Ignores conformational change of molecule
entirely.
NSEC
4
Hanabusa Itcho (1652-1724)
NSEC
5
NNIN/C Plan and Philosophy

• Mission develop modeling, design and analysis
  resources that complement and extend nanoscale
  experimental and theoretical research
• Begin from core codes and hardware facilities at
  NNIN computational sites
• Be a flexible resource many classes of users
• Be a dynamic resource expand and extend the code
  base to address new research problems
• Multiple user classes
• Remote and local users of NNIN codes run on NNIN
  computers
• Download users large packages and source-code
  modifiers
• Hardware-only users
• Code contributors

6
National Nanotechnology Infrastructure Network

• The members of NNIN include the following major
  nanotechnology user facilities
• The Cornell Nanoscale Facility at Cornell
  University
• The Stanford Nanofabrication Facility at Stanford
  University
• The Solid State Electronics Laboratory at the
  University of Michigan
• The Microelectronics Research Center at the
  Georgia Institute of Technology
• The Center for Nanotechnology at the University
  of Washington                                   
            
• The Penn State Nanofabrication Facility at the
  Pennsylvania State University
• Nanotech at the University of California at Santa
  Barbara
• The Minnesota Nanotechnology Cluster (MINTEC)  at
  the University of Minnesota
• The Nanoscience at the University of New Mexico
• The Microelectronics Research Center at
  University of Texas at Austin
• The Center for Nanoscale Systems at Harvard
  University
• The Howard Nanoscale Science and Engineering
  Facility at Howard University
• The Triangle National Lithography Center at NCSU
  ( DUV lithography only ) (Affiliate)

NWChem
7
(No Transcript)
8
Core Codes
HARES (High performance fortran Adaptive grid
Real space Electronic Structure) Abinit plane
wave pseudopotential first principles code. EDIP
(Environment Dependent Interatomic Potential)
SETE (Single Electron Tunneling Elements) LM
Suite Linear Muffin tin orbital
software. NWChem is a computational chemistry
package. SEMC-2D (Schrödinger Equation
Monte-Carlo). UTQUANT is a quasi-static CV
simulator for 1D silicon MOS structures. ANEBA
(Adaptive Nudged Elastic Band Approach). MIT
Photonic Bands (MPB).  UT-MARLOWE is a neutron
transport code. TOMCAT (TOpography based Monte
CArlo Transport). CPMD (Carr-Parrinello
Molecular Dynamics code). PARSEC (Pseudopotential
Algorithms for Real Space Energy
Calculations). Support and licenses also for a
variety of commercial codes (e.g. FEMlab,
Silvaco, Transiesta, Matlab) and scientific
libraries.
9
Core codes at Harvard
HARES (High performance fortran Adaptive grid
Real space Electronic Structure)
SETE (Single Electron Tunneling Elements)
EDIP (Environment Dependent Interatomic Potential)
ANEBA (Adaptive Nudged Elastic Band Approach).
HARES and SETE electronic structure at different
scales
HARES (E. Kaxiras)
SETE (M. Stopa)

• atomistic
• density functional theory, pseudo-potentials
• real space (highly parallel okay)
• periodic or isolated molecules

• effective mass
• density functional or configuration interaction
• full heterostructure device fidelity
• treatment of device disorder

adenine
Triple dot rectifier
10
Branching of Flow in a Disordered Medium

• Heller group simulations on flow through quantum
  point contacts
• Branching of flow due to multiple small
  scattering events
• Characteristic branching pattern, also in
  magnetic field an with scanning probe microscope
  tip present.

11
NNIN Hardware facilities At Harvard University
-- NNIN users currently have access to the DEAS
Cluster, comprised of 48 dual 32 bit Xeon blades
(3 GHz) each with 2 ½ GB of RAM with gigabit
ethernet. P655 IBM Power 4 Plus processors, total
of 20 processors with 80 GB RAM. 4 units of 4-way
32 GB Opterons from SUN Microsystems, for a total
of 128 GB RAM. Currently being installed NNIN
AMD Opteron cluster 112 processors, 56 connected
with Infiniband, 56 with gigabit
ethernet. At Cornell University -- 48 node
dual processor Xeon (3.06 GHz) cluster connected
by gigabit Ethernet lines donated by Intel. (16
nodes currently running due to cooling
constraints). 15 64 Bit Opteron workstations. 
The Opteron workstations were donated by AMD
Corporation. At University of Texas, Austin
ray-Dell PowerEdge Xeon Cluster with 600
3.06GHz Xeon processors within 282 Dell
dual-processor PowerEdge 2650 compute-I/O
server-nodes and 2 Dell dual-processor PowerEdge
2650 login/development nodesIBM Power4 System,
with 224 total 1.3 GHz Power4 processors and over
500GBof memory across four SMP nodes IBM
64-processor Intel Pentium III Linux cluster with
64 1GHz processors and 32GB of memory. At
Stanford University 12 processor linux cluster
for NNIN users
12
In nanoscience, computation and experiment are
synergistic

• Nanoscience concerns the fabrication, probing
  and manipulation of small structures and, as
  such, is intrinsically interdisciplinary.
• The description of nanoscale materials typically
  involves multiple, often nonlinear phenomena with
  energy scales in the same range.
• Nanoscale systems are often highly specific as
  regards material composition, geometry, and
  connection to the outside, larger scale world.
• ? Cant solve the Hamiltonian
• Experiments are expensive, modeling reduces
  fabrication.
• Consequently, computational modeling is an
  essential component to the understanding of these
  systems and the interpretation of experiments
  thereon

13
(No Transcript)
14
Example SETE 2DEG heterostructure modeling code
SETE 2DEG heterostructure and quantum dot
modelling
Electrostatic potential of triple dot rectifying
device charging ratchet.

• User Community Sandia Laboratories, Dartmouth,
  Delft Institute of Technology, Stanford,
  University of British Columbia,
  Ludwig-Maximillian University, Cambridge
  University, University of Tokyo, National
  Research Council of Canada

15
SETE input parser screen shot
16
Photoshop Gate Parser
17
Double-Double dot interaction
SETE output
2 ?m
Density (0, 2 x 1011 cm-2)
Potential (-15meV , 300 meV)
18
Code contributors
19
Path Integral Simulations of Semiconductor
NanostructuresJohn Shumway NSF-CAREER DMR 0239819

• The thermal density matrix of an interacting
  electron system can be described by a path
  integral,
• We sample this path integral for electrons in a
  realistic device, such as a gated quantum point
  contact

Example of charge density calculated in a gated
channel. (We use parallel processing with MPI)
Illustration of imaginary time paths.
20
Other Code Contributors

• Normand Modine, (Sandia National Laboratories).
  Socorro. Highly parallel electronic structure
  code. DFT.
• Toshiaki Iitaka (RIKEN) and Shintaru Nomura
  (Tsukuba), solution of time-dependent Schrödinger
  equation in 2D billiard-like systems.
• Metin Muradoglu, (Koc University, Istanbul).
  Computational model of interfacial flows in
  Micro/Biofluidic systems.

Mixing in a drop evolution of mixing patterns
Microfluidic Application I Drops As Chemical
Reactors
21
How does it stand with respect to
cyberinfrastructure ?
Data mining, grid-based computing (note see talk
of Joy Sircar this afternoon), visualization,
algorithm development all essential features of
CI.
NNIN/C focus vis a vis CI

• Make codes user-friendly, recruit code creators
  rescue from code mortuaries
• Maintain feedback between experimentalists and
  computational scientists
• Modify or create codes to address problems of
  experimental interest. If it cant be measured,
  its mathematics (or string theory).

22
Conclusions

• NNIN/C is an NSF-funded user network dedicated to
  establishing a computational infrastructure for
  research in the nanoscale sciences.
• Nanoscience is multiscale, interdisciplinary and
  heterogenous
• Many problems can (must) be approached with
  multiple computational tools. Find, modify or
  create the tool that is fit for the system under
  study.