Title: Status of the Nanoelectronic Modeling tool (NEMO 1-D and 3-D) and its planned extension to Spintronics
1Status of the Nanoelectronic Modeling tool (NEMO
1-D and 3-D) and its planned extension to
Spintronics
- Gerhard Klimeck, Fabiano Oyafuso, Timothy B.
Boykin, and R. Chris Bowen - Jet Propulsion Laboratory, California Institute
of Technology - University of Alabama in Huntsville
- Email gekco_at_jpl.nasa.gov
- Web http//hpc.jpl.nasa.gov/PEP/gekco
- Outline
- Atomistic modeling agenda
- Tight binding parameter fitting
- Parallel eigenvalue solver
- Parallel strain calculation
- Alloy disorder calculation
- AlGaAs bulk
- InGaAs quantum dot
- Planned Extensions
2High Performance Computing at JPL
- JPL is the Lead Center for Robotic Space
Exploration - Deep Space Missions
- Earth Observing Missions
- JPL Builds Observational Systems for Remote
Sensing - Imaging instruments across all wavelengths
- Spectroscopic and in-situ instruments for
planetary investigation - Fundamental technology development for new
instruments
http//mars.jpl.nasa.gov
http//www-misr.jpl.nasa.gov
- JPL High Performance Computing is a Key
Technology - Modeling and simulation of devices and
instruments - Rapid data reduction and analysis
- Advanced software design, implementation and
application - Integrated Design, Optimization and Synthesis
3Progressive Spacecraft Miniaturization
Cassini
1000 kg
Mars Pathfinder
Clark
Lewis
Solar Probe
NEAR
Pluto/Kuiper Express
Mars 98 Lander/Orbiter
100 kg
Spacecraft Mass
Europa Orbiter
Stardust
10 kg
Microspacecraft
Past
Present
Future
Low weight, low power and high efficiency Have a
special meaning to NASA
4Limit of Military Interest
Limit of Commercial Interest
NASA radiation and temperature requirements are
outside commercial and military interest
5Quantum Dot Simulation forRevolutionary
Computing and Sensing
DesignedOptical Transitions Sensors
Quantum Dot Arrays Computing
Atomic Orbitals size 0.2nm
Structure
- Opportunity
- Nanoscale electronic structures can be built!gt
Artificial Atoms / Molecules - Problem
- The design space is huge choice of materials,
compositions, doping, size, shape. - Approach
- Deliver a 3-D atomistic simulation tool
- Enable analysis of arbitrary crystal structures,
atom compositions and bond/structure
configurations.
6Quantum Dot Simulation forRevolutionary
Computing and Sensing
DesignedOptical Transitions Sensors
Quantum Dot Arrays Computing
Atomic Orbitals size 0.2nm
Structure
- Approach
- Use local orbital description for individual
atoms in arbitrary crystal / bonding
configuration - Use s, p, and d orbitals Use genetic algorithm
to determine material parameter fitting - Compute mechanical strain in the system.
- Develop efficient parallel algorithms to generate
eigenvalues/vectors of very large matrices
(N40million for a 2 million atom system). - Develop prototype GUI for (NEMO-3D)
- Problem
- Nanoscale device simulation requirements
- Cannot use bulk / jellium descriptions, need
description of the material atom by atom gt use
pseudo-potential or local orbitals - Consider finite extent/transport, not infinitely
periodic gt local orbital approach - Need to include about one million atoms. gt need
massively parallel computers - The design space is huge choice of materials,
compositions, doping, size, shape. gt need a
design tool
7Leverage NEMO 1-DA User-friendly Quantum Device
Design Tool
- NEMO was developed under a government contract to
Texas Instruments and Raytheon from 1993-97 - gt50,000 person hours of RD
- 250,000 lines of code in C, FORTRAN and F90
- Based on Non-Equilibrium Green function formalism
(Datta, Lake, Klimeck). - NEMO in THE state-of-the-art heterostructure
design tool. - Used at Intel, Motorola, HP, Texas Instruments,
and gt10 Universities.
Transport/ Engineering
Quantum Mechanics / Physics
Testmatrix
8Mapping of Orbitals to Bulk Bandstructure
- Bulk Semiconductors are described by
- Conduction and valence bands, bandgaps (direct,
indirect), effective masses - 10-30 physically measurable quantities
9Semiconductor Compoundscation In, Ga,
Alanion Sb, As, P
In
Ga
Al
Sb
- Match experimental data in various electron
transport areas of the Brillouin zone - Effective masses of electrons at G, X and L
- Effective masses of holes at G
- Bandedges at G, X and L
As
- Each individual material poses a 15 dimensional
fitting problem.
P
10Eigenvalue Solver for 108x108 MatricesHermitian
and Non-Hermitian Matrices
11Parallelization of NEMO 3-D
- Divide Simulation domain into slices.
- Communication only from one slice to the next
(nearest neighbor) - Communication overhead across the surfaces of the
slices. - Limiting operation sparse matrix-vector
multiplication - Enable Hamiltonian storage or re-computation on
the fly.
12Parallel Eigenvalue Solver on a Beowulf(32 node,
dual CPU Pentium III, 800MHz, Linux)
- Eigenvalue computation ranging from 1/4 to 16
million atoms - Large problems are too big for a single CPU
(memory requirements) - sp3s basis set , Matrix sizes up to 1.6 108x1.6
108 - Recompute Hamiltonian matrix on the fly.
- Measure time for 30 Lanczos iterations, Full
problem 1000-5000 iterations - 1million atoms 5000 iteration 1 CPU 48 hours
20 CPUs 3.4 hours - Computation time linear in system size.
13Atomistic strain calculationfollowed by sp3sd5
tight binding eigenvalue solution
- Dot Formation Due to Strain
- Self-Assembly induced by strain in GaAs/InAs and
Si/Ge material systems. - Bond length and orientation distortion
14Parallelization of Strain Calculation
- Problem (1million atoms)
- Serial strain computation 43 min.
- Serial electronic structure calculation (1000
iterations) 9.6 hours - Parallel electronic structure computation on 20
CPUs41 min. - Solution
- Parallelize strain calculation as well
15Alloy Disorder in Quantum Dots
- Results
- Simulated 50 dots with random cation
distributions. - Inhomogeneous broadening factor of 5meV due to
alloy disorder. - Impact
- Fundamental uniformity limit for ensemble of
alloy-based quantum dots.
- Problem
- Cations are randomly distributed in alloy dots.
- Does alloy disorder limit electronic structure
uniformity for dot ensembles? - Approach
- Simulate a statistical ensemble of alloyed dots.
- Requires atomistic simulation tool.
In0.6Ga0.4As Lense Shaped Dot
Diameter30nm,Height5nm, GaAs embedded
1,000,000 Atom Simulation, sp3s basis
In and Ga atoms are randomly distributed
Inhomegenious Broadening?
163-D Random Alloy Simulation ofAlGaAs Band Gap
- sp3s tight binding model
- VCA derived from pure GaAs and AlAs results in an
wrong bandgap (parameter averaging) - Perform 3-D alloy simulation of the bandedges.
- Represent each individual atom in the chunk of
material - 3-D random alloy simulation matches experimental
data well.
- Some questions to be addressed
- What is the noise in such a system?
- How large is the configuration noise?
- How large is the concentration noise?
- How many atoms need to be included in the
simulation? - What is the dependence of the noise on the Al
concentration? - What is the effect of clustering?
17Configuration Noise
- The actual placement of the Al atoms in the 3-D
domain differs in each sample-gt configuration
noise - For a system containing 1000 atoms with the exact
concentration, the variation is about 2-5 meV - Conduction band noise increases at the G-X
transition at Al0.45 - Valence band dependence is much smoother.
18Configuration and Concentration Noise
- Concentration may vary stochastically as well.
- Concentration noise is larger than configuration
noise. - For a system containing 1000 atoms, the variation
is about 10-15 meV - Conduction band noise shows a significant feature
at the G-X transition (Al0.45) - Valence band dependence is much smoother.
- What do the distributions look like?
19Configuration NoiseDistribution Functions at
Al0.2
- 3000 atoms
- 2800 samples
- x0.2, s0 Configuration noise only
- Valence band edge shows some asymmetry
- Conduction band symmetric
- Band gap has asymmetry due to valence band
asymmetry.
20Configuration and Concentration
NoiseDistribution Functions at Al0.2
- 3000 atoms
- 2800 samples
- Concentration noise dominating the configuration
noise. - Valence and conduction band have symmetric
distributions
21Configuration and Concentration
NoiseDistribution Functions at Al0.2
- Configuration noise uncorrelated Ec and Ev
- Concentration noise correlated Ec and Ev.
22Size Dependence of Bandedge and Deviation
- Configuration noise only.
- 1 million atom simulations !
- Standard deviation decreases approximately as
N-1/2
23Alloy Disorder in Quantum Dots
- Results
- Simulated 1000 dots with random cation
distributions. - Inhomogeneous broadening factor of 5meV due to
alloy disorder. - Impact
- Fundamental uniformity limit for ensemble of
alloy-based quantum dots.
- Problem
- Cations are randomly distributed in alloy dots.
- Does alloy disorder limit electronic structure
uniformity for dot ensembles? - Approach
- Simulate a statistical ensemble of alloyed dots.
- Requires atomistic simulation tool.
In0.6Ga0.4As Lense Shaped Dot
Diameter30nm,Height5nm, GaAs embedded
1,000,000 Atom Simulation, sp3s basis
In and Ga atoms are randomly distributed
Inhomegenious Broadening?
24InGaAs Bond Length Distribution
- VCA on the bond length is incorrect.
- InAs and GaAs maintain their bondlength character
- gt proper strain treatment
25Local Bandstructure in an Alloyed QD
- In-As bonds compressed in x-y
- -gt Ec raised from bulk value of 0.58eV to
1.2eV - -gt Ev HH raised from bulk value of 0.22eV to
0.3eV - Ga-As bonds compressed in x-y and stretched in z
inside dot - -gt Ec raised from bulk value of 1.42eV to
1.55eV - -gt Ev raised from bulk value of 0eV to 0.1eV
26Spatial Irregularity in the Hole Ground State
- VCA / no Disorder Disorder Sample 1 Disorder
Sample 2
27Spatial Irregularity in the Hole States
28Spatial Irregularity in the Electron States
29gt1000 Alloyed Quantum Dot Samples
- Atomistic granularity
- s 2.3meV
30gt1000 Alloyed Quantum Dot Samples
- Atomistic granularity
- s 2.3meV
- Cell granularity
- s 4.5meV
31gt1000 Alloyed Quantum Dot Samples
- Atomistic granularity
- s 2.3meV
- Cell granularity
- s 4.5meV
- Ec and Ev strongly correlated
- Ec and x weakly corelated
32NEMO 3-D Conclusion / Future Vision
Quantum Dots
- Atomistic Simulation (NEMO 3-D)
- Fitting tight binding sp3s,sp3d5s
- General structure input
- Several million atom solutions
- Parallel eigenvalue solvers
- Strain simulations
- Spin is explicitly included in the basis.
Grading
Atomistic Simulation
Abrupt
Graded
gt Quantum Computing
End of SIA Roadmap
- Extension of NEMO 3-D to Spintronics
- Arbitrary magnetic fields
- Magnetic impurities
- Many-body interactions (electrons, phonons,
photons) - Open boundary conditions
- Extension of NEMO 1-D to Spintronics
- DC Spin transport RTD-like structures
- Time dependent spin - transport
Dopant Fluctuations in Ultra-scaled CMOS
Electron Transport in Exotic Dielectrics
(Ba,Sr)TiO3
TiO2