Transcript Gordonx - Bartol Research Institute

Single-Event Upsets and Microelectronics
(Why neutrons matter to the electronics industry)
Michael Gordon, Ken Rodbell
IBM TJ Watson Research Center
Yorktown Heights, NY 10598
Introduction
 Single Event Upsets
 The neutron radiation environment
 Some neutron-induced SEU results on commercial SRAM
 Summary
 Questions and answers
2
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Single Event Upsets
 What is a soft fail?
– Errors in chips (logic, memory) that don’t cause permanent damage
– Created by passage of energetic ionizing radiation through the sensitive volume of chips
– This is a reliability problem for servers, laptops, smart devices, pacemakers, airplanes,
autonomous cars, drones…
 Sources of soft fails:
– a-particles from the natural radiation in chip packaging (ceramic, silicide, insulators,
wafers, copper)
– High energy neutrons which create highly ionizing particles when they interact w/ silicon
(spallation)
– Thermal neutrons interacting with 10B through the 10B(n,a) reaction
3
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Single Event Upsets in the News (Sun Microsystems)
4
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Single Event Upsets in the News (defibrillators)
5
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Single Event Upsets in the News (Pluto mission)
Just after the Jupiter flyby, New Horizons suffered its first computer glitch. For spacecraft outside Earth’s protective
atmosphere, high-energy cosmic rays occasionally zip through computer memory, causing a crash and restart.
Calculations indicated that there would be one such crash during the nine-and-a-half-year trip to Pluto. Instead, they
occurred almost once a year. But none caused lasting damage, and they proved good learning experiences.
6
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Two complete issues of IBM J. Res. Dev. devoted to SEU research
IBM J. Res. Devel. Vol 52. No 3, 2008
7
IBM J. Res. Devel. Vol 40. No 1, 1996
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Radiation environment: a-particles near chips
 Alpha particles
– Sources: Solder bump (C4’s), underfill,
metals other packaging materials
– The a-particle energy and flux are
attenuated by passing through
other materials
– Measurable effects on SEU for aparticle emissivities of <2a/khr-cm2
(1.5 a/hr on a 300 mm dia. wafer)
– Can reduce the influence of aparticles on SEU by reducing flux
(shielding, screening)
– But their range is small, < 100 mm,
fortunately!
8
E. Cannon, ICICDT, 2007
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
 deposit lots of charge
 but are very short ranged
 Cause SEU through
direct ionization
dQ/dX (fC/um)
Alpha particles
16
80
14
70
12
60
10
50
8
40
6
30
4
20
2
10
0
Range (um)
Charge Deposited and Range of a particles in Si
0
0
2
4
6
8
10
Energy (MeV)
9
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Cosmic Rays and their effect on Single Event Upsets
 Cosmic rays
– Cosmic rays (mostly protons) interact with atoms in the atmosphere
– Neutrons observed on the ground come from interactions of energetic protons and
14N or 16O in the atmosphere
– Energy range- meV (thermal) to GeV
– Velocity of neutrons: 150 MeV neutrons travel at ~c/2
 Ionizing radiation in chips
– Terrestrial neutrons striking chips interact with silicon atoms or metals near the
transistors
– “Spallation” occurs through 28Si(n,X) reaction and generates recoil ions
– These recoil ions generate charge and can cause SEU
– High energy protons (E>50 MeV) cause same effect as neutrons on chips
•
•
•
10
So some labs use external beam of protons as proxy for neutrons
More availability of proton beams (proton therapy centers)
Significantly higher flux for protons compared to neutrons from spallation sources
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Terrestrial Neutrons and effect on semiconductors
 Two types of neutron-induced SEU experiments
– Accelerated testing using external beam of neutrons
• Individual chips, or system tested
• Use the acceleration factor, ~ 1E6
• Flux from source integrated over energy > 10 MeV / terrestrial (NYC) flux
– “Life” (or real-time) testing using natural flux of neutrons at high elevation
•
•
•
•
•
•
Assess SEU in “real-world” conditions
Acceleration factor, ~ 10-15
Use calculated neutron flux (or measured if available)
See influence of neutrons and alpha particle
Systems tested rather than individual chips
Tests often take months to get adequate statistics
 Life-testing can be run underground to get a measure of the alpha flux
from the packaging (effect can be subtracted from the altitude meas.)
 Testing usually follows prescription of JESD89a
– Data taken by IBM and Paul Goldhagen, using Paul’s Bonner sphere system, and
resulting model are “gold standard” for the semiconductor industry
11
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Terrestrial Neutron Measurements
data
Model from original
JESD89 spec.
analytic
model
from Goldhagen
Figure E.2.1 — The differential flux of cosmic-ray-induced neutrons as a function of
neutron energy under reference conditions (sea level, New York City, mid-level solar
activity, outdoors). The data points are the reference spectrum, the solid curve is the
analytic fit to the reference spectrum, and the dashed curve is the model from the
previous version of this standard, JESD89 (2001).
M. Gordon, et. al, IEEE Trans. Nucl. Sci. 51(6), 3427, Dec 2004
12
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Why are cosmic ray neutrons important?

High-energy neutrons striking silicon
nuclei can cause spallation reactions,
releasing light and heavy ions.

The ions have high LET (linear energy
transfer) and which can initiate SEU

The following table shows the
threshold energy needed for the
reactions to occur–
higher energy neutrons- more exit
channels
25Mg
+a
2.75 MeV
28Al
+p
4.00 MeV
27Al
+d
9.70 MeV
24Mg
27Al
+ n,a
+ n,p
26Mg
+3He
10.34 MeV
12.00 MeV
12.58 MeV
21Ne
+ 2a
12.99 MeV
27Mg
+ 2p
13.90 MeV
24Na
+ p,a
15.25 MeV
F. Wrobel, et. al., IEEE Trans. Nucl. Sci. Vol 47, No. 6, pp. 2580, Dec. 2000
13
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Normalized high energy neutron flux sources
Normalized to JEDEC from 1 MeV-1 GeV
Flux @ 40,000’
180 MeV P + W
800 MeV P+ W
500 Mev P + Pb/Fe
800 MeV P + W (mod)
392 MeV P + Pb
C. Slayman, IEEE Trans, Nucl. Sci., Vol. 57, No. 6, 3163, Dec. 2010.
14
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Secondary ion production for n(si,x) atmospheric and monoenegetic neutrons
S. Serre, et al., IEEE Trans, Nucl. Sci., Vol. 59, No. 4, 714, Aug. 2012.
15
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Thermal Neutrons, a Problem?
The magnitude/shape of the thermal neutron spectrum depends on the
scattering of neutrons from the environment (roof, walls, building materials)
10B
and other isotopes have large capture cross sections (high
probability of capture). 20% of B is 10B.
n + 10B
7Li
+ a (1.5 MeV);
104
103
3He(n,p)
10B(n,a)
10-2
16
6Li(n,a)
100 Neutron energy (eV)
Cross section 3840 B
Thermal neutrons will cause SEU’s
if there is an appreciable amount of
10B immediately surrounding
transistors
Solution is to eliminate B or enrich to
11B
From Knoll
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Failure rate caused by neutrons and alpha particles
Commercial 130 nm, 65 nm and 40 nm SRAM devices
a-particle emissivity
2.3 a/khr-cm2
FIT= error in 109 hours
0.92 a/khr-cm2
ASTEP is in the French Alps,
with an active neutron monitor
J.L. Autran, et al., Radiation Effects Data Workshop (REDW), RADECS, IEEE, 2014, 1
17
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
ASTEP (French Alps), 2552 m, AF=6 & underground test facility
J. L. Autran, et al., Radiation Effects Data Workshop, IEEE, 2014, pp. 1-8.
18
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Real-time, modeling and accelerated neutron testing
Good correlation
between modeling,
real time and
accelerated testing
J. L. Autran, et al., Radiation Effects Data Workshop, IEEE, 2014, pp. 1-8.
19
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
ASTEP neutron monitor
counts/hr
atmospheric pressure
J.L. Autran, et al., IRPS, 2012, 3C.5.1 - 3C.5.9
20
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Leadville neutron monitor in new home, 2009
21
Leadville, CO (10,200’)
Durham, NH (~50’)
388 kcounts / hr per tube
42.1 kcounts / hr per tube
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Effect of solar activity on SER (need for real-time neutron monitor)
Rodbell, IEEE NSREC short course, 2013
22
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation
Summary
 Single event upsets are a major reliability issue in the electronics industry
 Scaling device size (Moore’s law) helps, but the critical charge required to
flip bits is shrinking as well
 Real-time and neutron (or proton) accelerated testing is required to assess
the device or system SER
 Having a real-time neutron monitor at or near the real-time test site makes
the determination of the acceleration factor more precise, compared to
estimating the neutron flux from a calculator
 We at IBM support funding to keep US neutron monitors operational
23
Neutron Monitor Community Workshop, October 24, 2015
© 2006 IBM Corporation