Soft-Error Rate Induced by Thermal and Low Energy Neutrons in 40 nm SRAMs

1. SOFT-ERROR RATE INDUCED BY THERMAL AND
LOW ENERGY NEUTRONS IN 40 NM SRAMS
J.L. Autran, S. Serre, S. Semikh, D. Munteanu
IM2NP-CNRS, Aix-Marseille Université, France
G. Gasiot, P. Roche
STMicroelecronics, Crolles, France
2012 IEEE Nuclear and Space Radiation Effects
Conference, Miami, Florida – July 16-20, 2012

2. Purpose
• Explore the question of thermal and low energy neutron-
induced soft errors in 40 nm SRAMs from both
experimental and simulation point-of-views
• Study conducted in three steps:
• Preliminary investigations on thermal neutrons
interactions with boron-doped Si
• Experimental measurements on 40 nm SRAM using
a thermal neutron source
• Monte Carlo simulations of 40 nm SRAM and SER
estimation
2

3. Outline
• Introduction
• Geant4 analysis of neutron interactions with bulk
Silicon doped with natural boron
• Thermal neutron irradiation of 40 nm SRAM at LLB
facility (Laboratoire Léon Brillouin)
• Introduction of the new TIARA-G4 Monte Carlo
simulation code
• SER simulation results on 40 nm SRAM
• Conclusion
3

4. Introduction
• Atmospheric neutrons play a major role in the Soft-Error Rate
(SER) of electronic circuits at ground level
• Neutron energies are ranging from thermal energy to hundred
of GeV. Neutrons below 1 MeV represent a non-negligible part
of the spectrum :
4
10
-10
10
-8
10
-6
10
-4
10
-2
10
0
10
2
10
4
10
6
0
2
4
6
8
10
12
Neutronfluencerate
perlethargy(x10
-4
cm
-2
s
-1
)
Neutron energy (MeV)
Experimental data courtoisy from Paul Goldhagen (U.S. Department of Homeland Security)
Part #1
7.6 n/cm2/h
Part #2
16 n/cm2/h
Part #3
20 n/cm2/h

5. Introduction
5
• Thermal neutrons mainly interact with
10B present in IC materials following
the 10B(n,a)7Li reaction
• Boron is composed of two isotopes:
10B (19.9%) and 11B (80.1%)
• In modern electronics, the principal reservoir of boron (BPSG)
has been totally removed=> the only residual source of 10B is now
located at silicon level (boron-doped areas) because implantations
are generally not selective in isotopes
→ Objective of this study: quantify the contribution of thermal
and low energy neutrons in the SER of a 40nm SRAM
thermal n
(~0.0026 eV)

6. 6
Preliminary Geant4 simulations
• Objective: compile and analyze event databases resulting
from the interactions of atmospheric neutrons with natural
boron-doped silicon
• Methodology:
o Geant4 dedicated code
o Particle Source: full atmospheric spectrum (Goldhagen)
o QGSP_BIC_HP physics list
o Target: 1 cm2 x 20 µm thick Silicon layer with natural boron
doping level ranging from 0 to 3×1020 cm-3
o Databases compiled for 25×106 h of natural radiation at sea-level
o Total of 1.09×109 neutrons for the complete simulation

7. Preliminary Geant4 simulations
Simulation results :
7

8. 8
Preliminary Geant4 simulations
500
1000
1500
2000
2500
20000
40000
60000
80000
100000
DATABASE ANALYSIS Part #1 (1.9x10
8
thermal neutrons)
Target: 1cm
2
x 20 µm
10
B fission
neutron capture by silicon
neutron capture by boron
doped silicon with natural boron
Boron concentration [cm
-3
]
3x10
20
10
20
10
19
10
18
10
17
10
16
Numberofevents
undoped
silicon
Database analysis for Part #1 (thermal neutrons):

9. 9
Preliminary Geant4 simulations
Normalized graph
Number of 10B fissions
per mm2.µm and per 109 h
10
17
10
18
10
19
10
20
0.1
1
10
100
1000
Numberof
10
Bfissions
/(mm
2
.µm.10
9
h)
Natural boron doping level (cm
-3
)
Simple analytical model for
thermal neutron SER
Considering the volumes of
sensitive drains per Mbit of SRAM
and the criterion “1 fission in a
sensitive drain = 1 SEU”
130 nm 90 nm 65 nm 40 nm 32 nm 22 nm
0
5
10
15
20
ThermalneutronSER
[FIT/Mbit]
Technological node

10. 10
Experimental characterization
• “Pool” type reactor named ORPHÉE
• Thermal power of 14 MWh
• Total thermal neutron flux of
3×1014 n/cm²/s.
• Experiment conducted on the
G3-2 beam line
• Thermal neutron flux reduced to
7.88×108 n/cm2/s
• Beam surface area of 25×50 mm2
• Neutron energies in the range
1.8-10 meV
Thermal neutron irradiation conducted at LLB facility (Saclay, France)

11. 11
Experimental characterization
• Fabricated by STMicroelectronics using
a commercial CMOS low power process
• 40 nm technology = optical shrink of the
45 nm technological node
• Boro-Phospho-Silicate Glass (BPSG)-free
Back-End Of Line (BEOL)
• Tests performed on a 7 Mbit array of
standard single-port SRAM
(layout cell area of 0.374 µm2)
• Nominal core voltage: 1.1 V
40 nm SRAM test chip
D
SP RAM1
5 Mbit
SP RAM1
2 Mbit
SP RAM2
5 Mbit
SPRAM22Mbit
DP RAM1
896 kbit

12. Experimental characterization
Thermal neutron accelerated test results
12
Comparison with the analytical
model for thermal neutron SER
130 nm 90 nm 65 nm 40 nm 32 nm 22 nm
0
5
10
15
20
ThermalneutronSER
[FIT/Mbit]
Technological node
Experiment
Model
10
20
30
40
50
60
70
80
90
100
Events(%)
MCU5
MCU4
MCU3
MCU2
SBU
Experiment
Measured bit flip SER = 4 FIT/Mbit

13. 14
TIARA-G4 Simulation
Cell/Circuit
construction
Radiation event
generator
Interactions,
particletransport
and tracking
SRAM electrical
response
Bit-flipoccurrence
Upset criteria
SoftError Rate
Cross-section
Error bitmap
3D Circuit geometry
from layout (GDS)
Natural radiation
environment models
Additionalparameters
extracted from TCAD
or SPICE simulations
Charge transport
models
FEOL & BEOL
materialsand doping
TIARA-G4
simulation flow Geant4
G4 GeneralParticle
Source (GPS)
VirtualGeometry
Model(VGM)
ROOT Event
Visualization
Particle Source
Characterization
ROOT Result
Analysis
G4 Geometry classes
G4 Elements
G4 Materials
Directimpact model
Collection-diffusion
model
Specific C++ routines
implemented in the
Geant4 code using
G4 Geometry classes
Artificial sources
Physicallist
QGSP_BIC_HP
J.L. Autran et al. in Numerical Simulation, Intech (2012)
New release (2012) of the TIARA code → full Geant4 application
Silicon
substrate
Tungsten
plugs
Si3N4
Al
SiO2
Cu
Cu
SiO2
SiO2
SiO2
Nwells
Pwell Pwell Pwell
CompleteBEOL
18 layers
Total thickness
8.75 µm
PMOS
drains
NMOS
drains
SiO2
Cu

14. 13
TIARA-G4 Simulation
a
7Li
Siliconsubstrate
BEOL
P-wells N-wells P-mos drains N-mos drains
[B] = 3×1020 cm-3
40nm SRAM block (20×20) • = reaction vertex
a
7Li
n
P-well
N-mos drains
N-well
P-well
P-mos drains
ROOT vizualization tool screenshots

15. TIARA-G4 Simulation
15
10
20
30
40
50
60
70
80
90
100
Events(%)
MCU5
MCU4
MCU3
MCU2
SBU
Experiment Simulation
Monte Carlo simulation results
0
2
4
6
8
10
12
14
16
18
400
500
600
700
800
Exp.
real-time
ASTEP
Simul.
1+2+3
Simul.
Part #3
Simul.
Part #2
Exp.
LLB
SERinbitflips[FIT/Mbit]
Simul.
Part #1
Single-port 40 nm SRAM
Standard density
682
thermal neutrons
Atmospheric (full spectrum)
500.5482
10
44.5
Thermal neutrons
(Atmospheric
spectrum Part #3)
* J.L. Autran et al. IRPS (2012)
*

16. 16
Conclusion
• This work investigated thought experiment and simulation the
question of thermal and low energy neutron sensitivity of a state-
of-the-art SRAM technology for which the only residual source of
10B is the natural boron doping in silicon
• SER characterization using a thermal neutron source
demonstrated a thermal neutron SER of 4 FIT/Mbit for the 40 nm
SRAM under test
• Geant4 simulations used to calculate the 10B fission rate in
natural boron-doped silicon predicts a thermal SER around
a few FIT/Mbit for technology nodes down to 22 nm
• For the 40 nm SRAM, Monte Carlo simulation using the new
release of the TIARA code (TIARA-G4) confirms the negligible
impacts of thermal and low energy neutrons in the total SER