An Overview of PRISM Jayathi Murthy Alejandro Strachan Purdue University

1
An Overview of PRISMJayathi MurthyAlejandro
StrachanPurdue University
2
Outline

• Overarching application
• Science focus and prediction goals
• Software approach
• Uncertainty quantification and VV
• Organization

3
PRISM Mission

• Accelerate substantially the integration of MEMS
  technologies into NNSA stockpile monitoring and
  weapons systems
• Significantly improve understanding of long-term
  reliability of MEMS and survivability in extreme
  environments
• Achieve this goal by simulating rigorously, and
  at multiple scales, the physics of failure
• coupled electrical, mechanical, thermal and
  materials behavior
• from atoms to devices
• uncertainty quantification
• verification and validation

4
MEMS Performance Requirements

• MEMS must satisfy stringent requirements before
  inclusion in NNSA stockpile
• Survive billions of cycles of operation
• Survive dynamic impact conditions
• High g up to 30,000 g over milliseconds
• Large range in operating temperatures
• Typically -50C to 80C
• Must function after long storage

5
MEMS Today
6
Failure Mechanisms
7
Target PRISM Device
Anchor

• Contacting capacitive RF MEMS switch
• Used for contact actuators and capacitive
  switches
• Metal membrane makes periodic contact with thin
  dielectric layer

8
Failure Mechanisms
Dielectric Charging
Mechanisms poorly understood but of fundamental
scientific interest!
Contact Area Damage
Current Channeling
Mechanical Failure
9
Uncertainty
Simulation inputs are inherently uncertain!
Electrodeposited LIGA Ni microtensile specimens
(Hemker and Sharpe, Ann. Rev. Mater. Res, 2007)
10
Multiphysics/Multiscale Simulation
Vbias 0
11
Outline

• Overarching application
• Science focus and prediction goals
• Software approach
• Uncertainty quantification and VV
• Organization

12
Multi-physics, Multi-scale Phenomena
Solid / fluid dynamics Fluid damping
Characterization and validation experiments
Initial microstructure Performance and failure
12
13
PRISM Multi-physics Integration

• Trapped charges in dielectric

Predictions
Electronic processes
Validation Experiments Microstructure evolution,
device performance reliability
PRISM Device simulation MPM FVM

• Elastic, plastic deformation, failure

Micromechanics

• Defect nucleation mobility in dielectric

• Fluid damping

Fluid dynamics

• Temperature species

• Dislocation and vacancy nucleation mobility in
  metal

Thermal and mass transport

• Fluid-solid interactions

• Thermal electrical conductivity

Input Experiments Surface roughness,
composition, defect densities, grain size and
texture
13
14
Dielectric Charging
Problem Definition
Impact induced

• Electron/hole pairs injected into dielectric
• Resulting distortion of electric field
• Joule heating
• Dynamic positive feedback loop
• Complicated by dynamic impact effects

Heating-induced
electron-induced
Goal Predict dielectric charging in dielectric
under operating conditions

• Research challenges
• Role of impact coupling to atomistics
• Multi-scale in time/multi-physics
• How to make long time-scale predictions?
• Very large simulation domain
• Previously 5x5 mm2, now 100x100 mm2

15
Micromechanical Modeling
PI Marisol Koslowski
Goal Predict plastic deformation of metallic
bridge
Previous Experience

• Research challenges
• Size and microstructure effects
• Grain boundaries free surfaces
• Long timescales

• Challenges II
• Micromechanical model in device simulation
• Crystal plasticity size effects
• Contact
• Solves individual grains
• Creep

15
16
Multiscale Models of Fluid Damping
PI Alina Alexeenko
Goal To accurately simulate gas damping across
Knudsen regimes Research challenges
Transitioning between continuum and rarefied
regimes high computational cost
Fluid-structure accommodation coefficients
from atomistics
Streamlines and pressure contours for moving
micron-sized plate at Kn0.1 (ES-BGK solution)
16
17
Atomistic Modeling of Contact Physics
PI Alejandro Strachan
Goal Provide insight and first principles
characterization of materials processes and
properties that govern PRISM device

• Research challenges
• Role of initial microstructure surface
  roughness, moisture and impact velocity on
• Mechanical response
• Force-separation relationships (history-dependent)
  
• Generation of defects in metal roughness
  evolution
• Generation of defects in dielectric (charging)
• Electronic properties
• Thermal role of electrons
• Current channeling/Joule heating
• Surface chemical reactions

Previous work
Mesodynamics
Chemistry
18
Device Level Simulations
PIs J. Murthy, D. Sulsky, S. Mathur

• Goal
• Full PRISM device simulation
• Membrane, dielectric, and other solid structures
• Nitrogen atmosphere moisture
• Surrounding package

• Research challenges
• Structural deformation
• Electrostatics, charge transport
• Continuum and rarefied gas dynamics
• Fluid-structure interaction
• Thermal transport
• Contact physics through compact models of
  atomistics
• Micromechanical material models

Length and time scales 100s ?m 1-1000s of cycles
19
Ultimate and Intermediate Prediction Goals

• Ultimate
• Multiple-cycle device simulations to predict
• Damage accumulation, performance evolution
• Cumulative behavior that enables longer-time
  prediction

• Intermediate goals
• Electrical performance without contact (device
  always closed capacitor)
• Dielectric charging (minus defects produced by
  contacts)
• Metallic bridge dynamics without contact
• Fluid-solid interactions, solid dynamics
• Stress relaxation when contact is closed
• Micromechanical model of metallic bridge

20
Outline

• Overarching application
• Science focus and prediction goals
• Software approach
• Uncertainty quantification and VV
• Organization

21
Software Architecture

• C/C building blocks combined with Python
• Modularity, maintenance, extensibility
• Flexible orchestration of solver suite

22
Parallel Computing and Scalability

• Excellent scaling to 65,536 processors
• Results used to form reduced order models and
  populate look-up tables
• Coupled through compact models to FVM/MPM

Image from LAMMPS web site http//lammps.sandia.go
v/bench.html
23
Parallel Computing and Scalability

• ParMETIS / Chaco for domain decomposition

24
Parallel Computing and Scalability
FVM
MPM

• ParMETIS / Chaco for domain decomposition
• Coupling between FVM / MPM through Immersed
  Boundary Method (IBM)
• Iterative solvers pre-conditioners SPIKE,
  Trilinos, PETSc, Aztec, Hypre, etc.
• Use Vtune / Tau / PAPI for optimization

25
Software Engineering Tools

• Subversion source code control
• Autoconf/Make or SCons build system
• Doxygen for integrated documentation
• Wiki for project documentation
• Ticketing system for bug tracking

26
Outline

• Overarching application
• Science focus and prediction goals
• Software approach
• Uncertainty quantification and VV
• Organization
• Closure

27
UQ Techniques

• Generalized polynomial chaos (gPC)
• Stochastic Galerkin
• Stochastic collocation
• Sensitivity analysis
• C templating and operator overloading

Jemcov Mathur, 2006
28
Generalized Polynomial Chaos
Stochastic Collocation Methods
Solver
Choice of inputs from collocation
realizations
Easy to implement Minor changes to code - Not
as efficient as Galerkin
- Harder to implement - Significant changes to
code More efficient than collocation
â Good solution for LAMMPS â Good starting point
for other codes
â Better for FVM / MPM codes â Optimization
beyond collocation
29
UQ in MEMS Switch
PDF of tip displacement using gPC and MC, V7.0 V
Vertical displacement with error bars, V7.0 V
Aluru and co-workers
30
Experimental Program

• Material and interface phenomena characterization
• Single physics experiments focusing on
  electrical, mechanical and thermal domains
• Multi-domain experiments coupling two domains
• System-level validation experiments
• Uncertainty quantification experiments
• Leverage from Sandia and IMPACT Center at UIUC

31
Microstructure characterization (initial,
intermediate, final)
Dielectric charging, thermal, electro-thermal,
creep experiments
Experimental Program
Detailed structural vibration measurements
Uncertainty quantification experiments
32
Uncertainty Quantification
Experiments Determine input uncertainty for MEMS
material microstructure. Determine output
uncertainty for MEMS material properties How
Measured Electronic microscopes Test structures
for extracting material properties by precise
measurements of electrical quantities What
Microstructure of metal and dielectrics
capacitance, voltage and frequency for test
structures
Rms roughness 13nm
Jensen et al., 2005
ADI ADXRS150
33
Uncertainty Quantification Micro/Nanostructure
Example of Nickel fatigue/chemical analysis
Example of polysilicon fatigue analysis

• Crystallography
• Dislocation density
• Texture distribution

Example of polysilicon grain and chemical analysis
34
Key Ideas for New Metrology

• Exploit structures symmetry
• Perform differential measurements
• Link desired property to a precisely measured
  quantity, e.g.
• off-chip capacitance attoF
• on-chip capacitance zeptoF (ADI ADXRS150)
• One test structure? many material and
  geometrical properties

35
Dielectric Charging
Experiments Charging on static
metal-insulator-metal (MIM) capacitors How
Measured Axopatch 200B ultra low noise
pre-amplifier (current resolution level 10 fA for
10 Hz bandwidth) in a faraday cage What Measure
current through dielectric as a function of time
by applying bias voltage
Al/SiO2 switch
Pictures by J. Hwang, IMPACT center
36
Dielectric Charging
Goldsmith et al., 2006
37
Executive Director, Discovery Park, Purdue
University
Organizational Structure
Leadership Council Managing Director Associate
Directors Team Leaders
Advisory Board
Cross-Cutting Technologies
38
Estimated Computational Resources
Simulation Single and multiple contact events
Method MPMFVM simulations Predict Membrane
response, dielectric charging, micromechanical
failure Size (100 mm)3 / 10-100M DOF Time 1
to 400 ms / 0.25-100M steps Computation
resources15M CPU hours per run (based on Cray
XT3 CPUs)
Device-level simulation
Atomistic simulation
Simulation Single and multiple contact events
Method parallel MD Predict Contact
physics Size (250 nm)3 / 200M 2B atoms Time
200ps to 10 ns / 2M steps Computation
resources 10 M CPU hours per run (based on BGL
CPUs)
39
Why PRISM ?