Transcript Kickoff Meeting Expectations - Institute for Space and Defense

UNCLASSIFIED
Fundamental Aspects of Radiation
Event Generation for Electronics and
Engineering Research
Robert A. Weller
Institute for Space and Defense Electronics
School of Engineering
Vanderbilt University
UNCLASSIFIED
UNCLASSIFIED
Overview
• Introduction
• Research Program Background
• Technical Objectives
• Approach
• Expected Research Results
• Technology Transfer
UNCLASSIFIED
UNCLASSIFIED
Introduction
• The Question:
1. If you model all the physical processes by which radiation interacts
with materials, and by which electric charge moves in solids, as
accurately as current knowledge permits, can you predict radiation
effects in semiconductor devices from first principles?
2. If not, then why not?
• The essential issue - Complexity!
UNCLASSIFIED
UNCLASSIFIED
Background
• Existing methods for predicting the rate of single event
upsets in semiconductor devices have begun to fail
because their of their basic assumptions.
• New computational methods developed at Vanderbilt
have exposed gaps in basic science that limit our ability
to make accurate predictions of single event effects.
• Two important areas needing basic work are:
• The generation of final states of ionizing particles following
nuclear reactions.
• The microstructure of energy deposition and charge generation
by ions, including the spatial and energy distributions of carriers.
• The detailed motion of charge in matrices of sub 100 nm
structures is also important but is not the primary focus
of this work.
UNCLASSIFIED
UNCLASSIFIED
SEU Rate for a Modern SRAM
•
•
Comparing a calculation with data from
a real SRAM flown by NASA
Observed Average SEU Rate:
•
•
Vendor predicted rate using
CREME96:
•
•
•
1x10-9 Events/Bit/Day
2x10-12 Events/Bit/Day
Classical Method nearly a
factor 500 lower than the
observed rate
VU-ISDE rate:
•
•
•
All relevant physics with Geant4
1.3x10-10 to 1.3x10-9 Errors/Bit/Day
Wide error bar from Geant4
ion-ion physics uncertainty
UNCLASSIFIED
Multilayered
Stack
UNCLASSIFIED
The Impact of Complexity
NMOS
PMOS
12-T DICE Latch
Kevin Warren, VU ISDE
UNCLASSIFIED
NMOS
UNCLASSIFIED
A Focus on the Nuclear Physics
UNCLASSIFIED
UNCLASSIFIED
Charge Generation
Si-Nitride 0.4 µm
SiO2 1.0 µm
TiN 0.1 µm
Al 0.84 µm
TiN 0.1 µm
SiO2 0.60 µm
TiN 0.1 µm
Al 0.45 µm
SiO2 or W 0.6 µm
TiN 0.1 µm
Al 0.45 µm
TiN 0.1 µm
SiO2 0.6 µm
Si 0.25 µm
0
50 µm
Nuclear Reactions
UNCLASSIFIED
UNCLASSIFIED
Charge Motion
1018 e-/cm3
1014 e-/cm3
UNCLASSIFIED
UNCLASSIFIED
Technical Objectives
• Establish the available data, theory and computation for
ion-ion nuclear reactions
• Build virtual experiments that test these models for best
fit to semiconductor data
• Identify the best available nuclear reaction codes
• Interface and/or adapt the best codes for MRED
• Improve the nuclear-physics/charge-transport interface
as necessary for < 100 nm structures
• Establish a roadmap for any necessary nuclear reaction
research
UNCLASSIFIED
UNCLASSIFIED
Approach - Strategy
• Identify the best available
physics.
• Code the physics (whenever
possible) in timeless algorithms.
• Use supercomputer-based highfidelity simulations to extract the
physics that arises from
complexity.
• Establish validity by comparison
with experimental data.
• Let computer evolution deal with
any power shortfall (if possible).
UNCLASSIFIED
/* int patmat(pattern, string)
char *pattern, *string ;
patmat() examines the string s 'pattern' and 'string ' for equality and returns YES
(=1) if they are equal and NO (=0) if not. The string 'pattern' may contain '*'
and '?' characters which will match any substring and any singl e character
respectiv ely. Eith er symbol may appear at any lo cation in the pattern. The '*'
will also match a NULL string -- that is, the absence of any characters in the
designated position. The '?' character always requires that there be an explicit
(but arbitrary) character present in the test string .
*/
#defin e YES 1
#defin e NO 0
int patmat(pattern,string)
char *pattern, *string ; {
register char *p, *s;
register int match = NO;
p = pattern; s = string ;
whil e(*p != '*') {
if(*p != *s && *p != '?')
goto retNO;
if(*p++ == '\0')
goto retYES;
if(*s++ == '\0')
goto retNO;
}
if(*++p != '\0')
whil e(!patmat(p,s))
if(!*s++)
goto retNO;
retYES: ++match;
retNO: return(match);
}
/* Then *s must also. This is a match. */
/* No more string, but still some pattern
/* which is not a '*'. No match.
*/
*/
/* '*' at the end matches anything , even a
/* NULL string.
*/
*/
/* Robert A. Weller */
/* December, 1990 * /
Croc image: http://crocodilian.com/crocfaq/
UNCLASSIFIED
Approach - Implementation
• Identify available nuclear reaction models.
• Adapt models to mred/Geant4 if necessary.
• Develop structures to simulate nuclear physics
and radiation effects experiments.
• Conduct simulated experiments for comparison
with experimental data.
• Identify the best available models for practical
applications.
• Identify essentially-related issues outside the
scope of this effort.
UNCLASSIFIED
UNCLASSIFIED
Manage Complexity with Computing
UNCLASSIFIED
UNCLASSIFIED
Predicting Experiments
UNCLASSIFIED
UNCLASSIFIED
The Structure of MRED
• Python: The common
system language.
• MRED: A Python module
called mredPy.
• Target: VU Linux cluster.
• Python writes
submission scripts,
controls job execution,
and merges results.
UNCLASSIFIED
Python
SWIG
MRED
C++
Geant4
UNCLASSIFIED
Expected Results
• Identify optimum nuclear modes for MRED.
• Validate nuclear models against semiconductor
data.
• Identify issues related to the interface between
charge generation and transport in sub-100 nm
structures.
• Deeper understanding of the intricacies of
charge generation.
• Reduced uncertainty in SEU prediction.
UNCLASSIFIED
UNCLASSIFIED
Technology Transfer
• ISDE Engineering
• Collaborative R&D, e.g. NRL/Vanderbilt
• NASA MSFC/Vanderbilt CREME-MC Site
• Major semiconductor supplier relationships
• NASA Center collaborative R&D
• Through students
UNCLASSIFIED