1. Applications of Monte Carlo Radiation Transport Simulations Techniques for Predicting Single Event Effects in Microelectronics
Robert A. Reed
Vanderbilt University
Vanderbilt Team: Robert Weller, Marcus Mendenhall, Brian Sierawski, Kevin Warren, Ronald Schrimpf, Mike King, Mike Clemens, Elizabeth Auden, Sabra, Mohammad, Aritra DasGupta, Nathaniel Dodds
NASA/MSFC Collaborators: Jim Adams, John Watts, Abdulnasser Barghouty
SLAC Collaborators: Makoto Asai, Koi Tatsumi, Dennis H. Wright
Vanderbilt Sponsors:
NASA Electronics Parts and Packaging Program
NASA Autonomous Systems and Avionics Project
DTRA Basic Research Program
DTRA Radiation Hardened Microelectronics Program

2. Outline Introduction Brief history of SEE Rate Predictions
Predictions SEEs Using Monte Carlo Radiation Transport Techniques (VU)
Transporting radiation environments (MSFC)
CRÈME website (MSFC and VU)
Characteristics of the radiation event and the

3. Transients from Single Particle Event
Charge
Generation
~ ps
Energy
Deposition
Charge
Collection
ps to s
Transient
Current Pulse Ei Eo
fs to ps
Time
Time
Soft Error Examples:
Single Event Transient: A current pulse occurring at a circuit node due to single energetic particle event
Single Event Upset: A change in a circuit’s logic state induced by a single energetic particle event

4. Analytical On-Orbit SEE Performance Predictions
Space Environment
Ground Testing
ANALYTICAL RATE PREDICTION MODELS:
Rectangular Parallelepiped (RPP) model
Bendel Model
(both are circa 1980)
On-Orbit SEE Rate

5. Scaling Trends in Electronics
Single-Event Upset and Single-Event Transient Hardness Have Worsened with Scaling
SEE have recently become a substantial Achilles heel for the reliability of even earth-based advanced CMOS technologies. As the amount of charge that represents stored information has dropped lower and lower, so the sensitivity of CMOS devices to single-particle charge collection transients has increased. In fact, single-event upsets (SEUs) are the single largest contributor to device soft failure (i.e., recoverable) rates in many of today’s CMOS technologies. In addition, as operating speeds have soared, increased sensitivity of circuits to single-event transients (SETs, which are short-lived single-event induced current/voltage transients that do not directly cause upsets but that may propagate through logic circuitry) has followed. In fact, SETs in digital circuits may set fundamental limits on the operating speed of radiation-hardened integrated circuits. SEE are therefore a significant concern not only for electronics operating in space environments, but also for high-altitude and even terrestrial systems that must operate very reliably.
Scaling Trend
Many newer ICs also exhibit complex failure modes such as single-event functional interrupts (SEFIs) that may require device reconfiguration or reset for recovery.
P. E. Dodd, et al., RADECS 2009 Short Course, Brugge, Belgium

6. New SEE Challenges Complicated charge-collection volumes
Ion tracks larger than device sizes
Overlayers affect device response
One event may affect multiple cells

7. Breakdown of Analytical Models
Proton
effects in
SOI based
memories
Early experimental work on optocouplers & optical links show need for new tool
R.A. Reed, et. al, IEEE Trans. Nuc. Sci., vol. 49, no. 6, Dec. 2002, pp – 3044.
RADHARD
CMOS
SRAM
R.A. Reed, et. al IEEE Trans. Nuc. Sci., vol. 48, no. 6, Dec. 2001, pp – 2209.
Heavy Ion
Effects 90 nm
DICE latch
K.M. Warren, et. al IEEE Trans. Nuc. Sci., vol. 48, no. 6, Dec. 2005, pp – 2131.
K. M. Warren, et. al, IEEE Trans. Nuc. Sci., vol. 54, no. 6, pp , 2007.
Heavy Ion
Effects in
SiGe HBTs
R.A. Reed, et. al IEEE Trans. Nuc. Sci., vol. 50, no. 6, Dec. 2003, pp – 2190

8. A New Paradigm:
Classical models no longer capture all physical processes that drive the response of modern technology to ionizing radiation
Solution:
It is possible to predict single event effects from first principles given sufficient knowledge of device structure and the interaction of radiation with matter

9. Monte Carlo Method Using High Performance Computing
National Electrostatics Corporation
Predicted Responses
After E. G. Stassinopoulos and J. P. Raymond,
Proc. of the IEEE 76, 1423 (1988)

10. Key Technology: MRED Python First generation Python/Geant4 application
MRED: Monte Carlo Radiative Energy Deposition
First generation Python/Geant4 application
Contains the best available physics from various sources:
Geant4
PENELOPE2008
CEM03
LAQGSM
PHITS
Geant4
MRED
C++
SWIG
Python

11. MRED+PENELOPE2008: Electrons in Silicon
Silicon Cube 10 × 10 × 10 nm3
Penelope2008 extension
State-of-the-art electron, positron and gamma transport code
Gold-standard for low energy ionizing electromagnetic interactions with atoms and solids.
Simulate 250 eV electrons
100 particle raster

12. Trends in Advanced Technology Nodes
Decreasing feature sizes are leading to an overall reduction in critical charge
IBM 65 nm SOI critical charge around 0.14 fC – 0.24 fC (1500 electrons) [1]
Recent publication from IBM details a 22 nm SOI technology node [2]
SRAM cell area of 0.1 μm2
Estimated critical charge of 0.08 fC, approximately 1.8 keV (500 electrons)
0.1 μm2 SRAM after [2].
Decreasing feature size has resulted in overall reduction of Qc in SOI circuits – cite example.
Highlight IBM 22 nm technology – SRAM areal density of .1 um^2.
By estimating the gate capacitance we were able to approximate the critical charge for this technology node – 0.08 fC – 1.8 keV.
[1] Rodbell, K.P., et al, "Low-Energy Proton-Induced Single-Event-Upsets in 65 nm Node, Silicon-on-Insulator, Latches and Memory Cells," IEEE  Trans.  Nucl.  Sci. Dec
[2] Haran, B.S. et al, "22 nm technology compatible fully functional 0.1 μm2 6T-SRAM cell,” IEDM Dec
TEM of 22 nm devices after [2].
King, TNS 2010

13. Electron Events in 50 × 50 ×50 nm3 Si Cube
Incident δ-ray
Scattering events
Energy Deposited = 2.1 keV
E.D. = 2.6 keV
Both Events Deposit Sufficient Energy to Upset 22 nm SRAM !
King, TNS 2010

14. 280 MeV Fe Strike 28 GeV Fe Strike δ-rays δ-rays SVs SVs
Sensitive Volumes
Graphical representation of the preveious slide.
28 GeV Fe has a very irratic track structure – due to highly energetic delta-rays – 280 MeV Fe has a track structure more tightly focused around the ion track.
King, TNS 2010

15. Proton Test Structures
300 μm × 780 μm area

16. Data Analysis Charge collected and histogrammed
3. Data analyzed as a function
of charge.
2. Reverse Integrated

17. Effect of Tungsten overlayers
Tungsten (W) is seen to increase charge collection cross section for high collected charge.
No difference in charge collection cross section for charge below ~0.5 pC.
Presence of W can cause SEE in devices that would otherwise not see proton-induced SEE.

18. Monte Carlo Simulations
Monte Carlo Radiative Energy Deposition tool (MRED)
Geant4-based simulation tool
Provides option of selecting nuclear physics model
Geant4 Bertini Cascade
Cascade-Exciton Model (CEM)
Validate physics model, then investigate proton-induced charge deposition mechanism

19. Proton-Silicon Reactions
Geant4 Bertini Cascade has a branching ratio orders of magnitude lower than other models for proton-induced fission in W, and thus drastically under-predicts the energy deposition
CEM hadronic physics model in MRED integrated with Geant4 transport and energy deposition models agrees well with experiment

20. Hadronic Physics Model Comparison

21. Low-Energy Proton Upsets
Previous work reported data collected by Vanderbilt and NASA Goddard on TI 65 nm bulk CMOS process [Sierawski, TNS 2009]
Consistent with evidence of proton direct ionization contributing to single event upsets (SEUs) reported for IBM 65 nm SOI process [Rodbell, TNS 2007][Heidel, TNS 2008]
3-4
orders of magnitude
increase in cross section below 2 MeV
Cross sections consistent with nuclear
Reaction events for Ep > 10 MeV
Sierawski, TNS 2009

22. Heavy-Ion SEU Cross Sections
Heavy-ion tests at TAMU revealed LET threshold less than MeV-cm2/mg confirming proton sensitivity
Low-LET cross sections were used to calibrate an MRED multiple sensitive volume model
High-energy ions provide consistent energy deposition through a well-known stopping power thus characterizing the charge collection
σion
SBU
MBU
Sierawski, TNS 2009

23. σproton Proton Predictions
Proton SEU cross sections were predicted for Ep < 32.5 MeV
Transport codes model the stochastic energy deposition and the sensitive volume model captures the process of charge collection
Simulations include all relevant physical mechanisms of energy loss with no adjustable parameters
σproton
Sierawski, TNS 2009

24. Terrestrial Environment
Neutrons:
13 cm2hr-1 for E > 10 MeV
Muons:
60 cm2hr-1 for P > 0.35 GeV/c
Cosmic rays create showers (“zoo”) of secondary particles in the atmosphere
Neutrons, while numerous, rarely interact with material leading to single event upsets
Muons are the most abundance particle at sea level
Stopping power similar to protons
Sierawski, TNS 2010

25. Preliminary Simulations
MRED Monte Carlo simulations indicated potential for upset from muon direct ionization
Technology scaling (assuming same sensitive volume dimensions) will increase susceptibility to terrestrial protons, muons
Spectra will be moderated by concrete, buildings, etc.
45nm
65nm
Technology scaling

26. Single Event Upsets Trend in upsets is indicative of direct ionization
Peak occurs when beam is tuned to stop in device active region
At higher energies, stopping power is too low for upset
At lower energies particles range out
Kinetic energy distributions were determined by simulation and related to the abscissa for SEU probability
Two other SRAMs from a different vendor showed similar response
First experimental evidence of single event upsets from muon ionization!
Sierawski, TNS 2010

27. Applications of MRED Space Applications:
Integration of various physics models from various sources for comparison to experimental data
Identify strengths and weaknesses of model
Monte Carlo Based SEU Rate Predictions of NASA, DTRA, BAE, Boeing
Simple cuboid sensitive volume model
Multiple geometry sensitive volume model
Spice-in the loop
Transients in silicon focal plane arrays electron environments behind spacecraft shielding
Low LET (< 1 MeV-cm2/mg) effects
Multiple bit upset in highly scale (90nm and 65 nm) CMOS technology
Simulation of SEU Cross-Sections Using MRED Under Conditions of Limited Device Information
Prediction of SEGR in power MOSFETs
Dose enhancement effects
CREME-MC website
Terrestrial applications of MRED
Assessing Neutron Induced Multi-bit upset
Muon single event upsets
Alpha particles from packaging

28. Galactic Cosmic Ray Flux and Dose Using GEANT4/HZETRN 95
John Watts
University of Alabama Huntsville/CSPAR
(256)
February 5, 2010

29. Galactic Cosmic Ray Dose and Flux
Transporting the Full Galactic Cosmic Ray Flux (protons through uranium and energies for 10 MeV to 105 MeV) is computationally expensive for complex geometries.
The 3-D space craft geometry is in GDML format extract from files created by ST-Viewer which can read CAD STEP files.
We are investigating developing a replacement for ST-Viewer using the open source library, JSDAI, which provides an API for processing STEP files.
To make the problem tractable GEANT4 is used to generate and output a set of vectors sampling all directions about a point in a 3-D geometry with material description and distance through each volume encountered using geantinos.
Java script reads a single vector of multiple materials and transport the input galactic cosmic ray spectrum along the vector and output a file of the fluxes and dose at the end of the vector.

30. CRÈME Capabilities
Vanderbilt / MSFC developing a website that will provide certain functionalities to the user community
CREME86
CREME96
CRÈME-MC (limited Monte Carlo)
Monte Carlo Provides:
Multilayer structures and materials
Multiple sensitive volumes
Effects of track size and nuclear interactions
Accelerated ground test simulation
On-orbit predictions
Evaluates models for space radiation environments
Natural spectra
Transported spectra

31. Backup

32. Proton-W Reactions Proton-W secondary particles sorted by LET value
Secondaries with LET > 25 originate from proton-induced fission
Isotropic emission
Max LET: 42 MeVcm2/mg
High LET particles have a range from 4 μm to 22 μm
Proton-induced fission rare for low energy protons