(SEE) Testing - Radiation Effects & Analysis Home Page - NASA

Abstract-- The complexity of SEE testing for space
applications, has increased over the years. This trend has been
mainly driven by advances in integrated circuit technologies.
This paper reviews how the test environment relates to the
real space environment. It shows that medium energy beams
(25 to 200 MeV/amu) are more and more necessary for SEE
testing of state-of-the-art microcircuits.
T
Implications of Advanced Microelectronics
Technologies for Heavy Ion
Single Event Effect (SEE) Testing
Christian Poivey, Member, IEEE, Janet A. Barth, Member, IEEE, Robert Reed, Member, IEEE,
Epaminondas G. Stassinopoulos, Member, IEEE, Kenneth A. LaBel, Member, IEEE,
and Michael Xapsos, Member, IEEE
I. INTRODUCTION
HE complexity of SEE testing for space applications
has increased over the years. This trend has been
mainly driven by advances in integrated circuit
technologies. SEE testing of state-of-the-art devices has
raised many issues [1]:
- delidding processes for devices encapsulated in
plastic material.
- Increased complexity of the functions to test.
- Increased number of the input and output
interfaces.
- Increased clock speed.
- Increased SEE sensitivity and new phenomena
[2].
Shortcomings of SEE ground testing have been widely
discussed [1], [3], [4]. In this paper, we focus on how
heavy ions used for SEE ground testing relate to the actual
space environment. We review the basic concepts and
assumptions that underlie SEE ground testing techniques
using various ions to simulate the space radiation
environment. The limitations of these concepts for state-ofthe-art
devices are then presented.
Manuscript received September 10, 2001. This work was sponsored
by the NASA Electronics Parts and Packaging (NEPP) Program’s
Electronics Radiation Characterization (ERC) Project.
C. Poivey is with SGT-Inc. in support of NASA Goddard Space Flight
Center, Greenbelt, MD 20771 USA (telephone: 301-286-2128, e-mail:
cpoivey@pop500.gsfc.nasa.gov).
J.A. Barth, R. Reed, E. Stassinopoulos, K. LaBel, and M. Xapsos are
with NASA Goddard Space Flight Center, Greenbelt, MD 20771 USA.
II. SIMULATION OF THE SEE SPACE ENVIRONMENT
AT GROUND LEVEL
Fig. 1 shows representative Galactic Cosmic Ray (GCR)
energy spectra as modeled by CREME96 software [5].
CREME96 is considered to be the best available model for
this application. Its limitations are discussed in [6]. GCR
Particles are always present, but their intensities vary with
the solar cycle. During solar minimum, GCR intensities
attain their highest levels and during solar maximum they
are at their lowest levels. The most important characteristic
of GCRs is their very high energy (up to several GeV per
nucleon). They are highly penetrating in terms of
spacecraft shielding.
Flux (#/cm 2 -s-MeV/n)
1.E-02
1.E-03
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
1.E-09
100 mils, SOLMIN
1.E-10
1.E-01 1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05
C
Mg
Fe
Energy (MeV/nucleon)
Fig. 1: Representative Cosmic Ray energy spectra, CREME 96 solmin.
The Fig. 2 shows representative solar energetic particle
spectra. Note the very large fluxes (compared to GCR) at
low energies and the very rapid decrease of the flux with
energy. Because the solar particle spectra are
comparatively soft, typical spacecraft shielding does have a
more noticeable effect.
Ion beams dissipate energy in matter through a series of
collisions with the atoms of the material. In microcircuits
ionization leads to the creation of a dense plasma of
electron-hole pairs along the ion track. Existing electric
fields then transport the deposited charge to various nodes.
In the study of SEE effects in microcircuits, it has become
customary to describe the ion radiation quality by the
amount of energy lost per unit length of track, or the Linear
Energy Transfer (LET). As each material has a unique
H
He
1

ionization potential, the LET is easily converted to a
charge deposition rate. For example, in silicon, a charge
deposition rate of 1 pC/μm corresponds to a LET of 98
MeV/(mg/cm 2 ).
Flux(#/cm 2 -s-MeV/n)
1.E+04
1.E+03
1.E+02
1.E+01
1.E+00
1.E-01
1.E-02
1.E-03
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
Fe
GCR He
100 mils Al shielding
He
C
Mg
H
1.E-09
1.E-01 1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05
Energy (MeV/nucleon)
Figure 2: The energy spectra of solar energetic
particles, CREME96 worst day. The spectrum of cosmic ray Helium is
showed for comparison.
It is possible to obtain the same LET for different types
of ions and energies. However, if it is assumed that all ions
of a given LET have the same effect on the circuit, a LET
spectrum can be used to estimate the upset rates in space.
Such a characterization is called a Heinrich spectrum [7].
Flux(#/cm 2 -s)
1.E-01
1.E-02
1.E-03
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
1.E-09
1.E-10
1.E-11
1.E-12
Fe
1.E-13
1.E-14
100 mils Al shielding
1.E-01 1.E+00 1.E+01 1.E+02
LET (MeVcm 2 /mg)
Figure 3: GCR LET spectrum, CREME 96 solmin.
H
He
Li
Total Z=1 to 92
The GCR integral LET spectrum for silicon is shown in
Fig. 3 along with contributions of some individual species.
The lightest ions are dominant in the lowest LET (0.1 to 1
MeV/(mg/cm 2 )) part of the spectrum. The heaviest ions are
dominant in the highest LET part (30 to 100
MeV/(mg/cm 2 )). However, the fluences in this latter
portion are very small. For LETs from 1 to 30
MeV/(mg/cm 2 ), the LET spectrum is dominated by the Fe
and Ni ions. The Solar Particle Event (SPE) integral LET
spectrum is shown in Fig. 4. The lowest LET part of the
spectrum is dominated by protons and alpha particles.
Similar to GCR, the heaviest ions are dominant in the
highest LET (> 30 MeV/(mg/cm 2 )) range of the spectrum,
but their overall contribution is extremely small. In the
U
Au
Ni
Pb
medium LET range (1 to 30 MeV/mg/cm 2 )), Fe is the most
significant ion.
Integral Flux (#/cm 2 -s)
1.E+03
1.E+02
1.E+01
1.E+00
1.E-01
1.E-02
1.E-03
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
1.E-09
1.E-10
1.E-11
1.E-12
H
He
Fe
Au
Li
U
Pb
Ni
100 mils Al shielding
total Z=1 to 92
1.E-13
1.E-01 1.E+00 1.E+01 1.E+02
LET (MeVcm 2 /mg)
Figure 4: SPE LET spectrum. CREME96 worst day.
A plot of LET and range versus energy for Fe in silicon
is shown in Fig. 5. We can see that the highest LET values
correspond to the low to medium energy range (< 100
MeV/u).
LET (MeVcm 2 /mg)
40
30
20
10
0
10 -2
10 -1
10 0
Fe in Silicon
10 1
Energy (MeV/u)
Figure 5: LET and range in Silicon versus Energy, Fe ion.
As the SEE sensitive regions of many microcircuits are
relatively thin (several μm), ground testing is conducted
using ions with lower energies than GCR, but with similar
LET. The energy range at the SEE facilities commonly
used [8,9,10,11] is of the order of several MeV/u and the
range of ions is about from 30 μm for the heaviest particles
to 100 μm for the lightest particles.
The use of different angles of incidence is used to vary
the energy deposition in the sensitive volume of a
microcircuit and therefore obtain more experimental
measurements with the same ion. This gives rise to the
concept of effective LET, which assumes the energy
deposition is determined by the incident particle’s LET and
its angle of incidence. However, this does not account for
several factors that may influence the measured results.
First, this concept may not be valid for low energy ions
because the LET value may decrease as the ions traverse
the sensitive region. Secondly, the concept does not
10 2
10 3
10 4
2
10 6
10 5
10 4
10 3
10 2
10 1
10 0
range (um)

account for statistical fluctuations in the energy deposition
process. Thirdly, it assumes that effects due to ion track
structure are negligible. Quantitative estimates of these last
two effects have been discussed in [12]. Finally, it assumes
that the sensitive volume is a thin parallelepiped. Some
SEE, such as SEB and SEGR cannot be correlated with tilt
angle [13].
When the GCR LET spectrum is defined and the SEU
cross section curve versus LET is measured, the upset rate
can then be calculated assuming a Rectangular
Parallelepiped (RPP) sensitive volume and a constant LET
in this sensitive volume [14].
III. DISCUSSION AND EMERGING ISSUES
As discussed in the preceding section, an incident ion’s
LET is usually assumed to be a valid parameter for
describing SEE effects in microcircuits. But in modern
devices, sensitive volumes are very small and there may be
a strong difference between the energy lost by the ion and
the energy locally deposited in the sensitive volume. The
ion LET (and effective LET) overestimates the energy
deposition as shown in Fig. 6. This Figure shows the
fraction of an ion’s energy loss as a function of the
volume’s average chord length (ie the average travel length
of the particles in the sensitive volume) for different
incident ion energies [12]. We can see in the figure that the
smaller is the sensitive volume and the higher is the ion
energy, the higher is the overestimation. Then, charge
collection is a complex phenomenon affected by the size of
sensitive regions and the structural arrangements of
multiple junctions. Therefore, it is controlled not only by
the characteristics of the incident ions [15] but also by
geometrical arrangement of junctions and the properties of
the semiconductor material.
Fion
1
0.9
0.8
0.7
0.6
0.3 1 3 10 30
Curves labeled in MeV/u
0.5
0.001 0.01 0.1 1 10 100
Average chord length (mm)
Figure 6: fraction of an ion’s energy loss within the sensitive volume as a
function of the volume’s average chord length for different incident ion
energies [12].
In addition, the use of low energy ions for SEE testing
introduces experimental limitations: LET variations over
the sensitive volume, low range inducing angle effects and
limiting the funneling charge collection. Comparative SEE
100
300
1000
testing studies with high and low energy ions have shown
that standard low energy test methods were adequate to test
previous generations of microcircuits (down to 0.5 μm
feature size) [16, 17]. But, discrepancies have already been
observed in high density microcircuits and vertical
sensitive volumes like DRAMs [18,19].
Another significant factor that must be considered for
SEE testing of modern microcircuits is to use ions that
reach the sensitive regions. Some devices can have up to 5
metallization layers, and are constructed with a metal lead
frame extending over the die. Others have the entire die
encapsulated in plastic material [20]. Plastic removing
techniques have been used with success for several years,
but removing the lead frame of a DRAM is a much more
complex operation. These factors have resulted in the need
for ions with a longer range (ie of higher energy) to
penetrate to sensitive regions through plastic, lead frames
and/or metallization layers. It is not uncommon now to
have to go through on the order of 100 μm of material
before reaching the sensitive region of a device (for
example the thickness of a lead frame on a 64M DRAM is
about 150 µm). For certain types of devices such as flip
chip devices, this thickness could be even greater. The
heavy ion accelerators currently used for SEU testing can
not deliver this range.
Because of the complexities of the charge collection
process and new package technologies in advanced
microcircuits, it is becoming more and more necessary for
SEE testing to use ions with an energy range representative
of the SEE space environment.
Fig. 7 shows the ratio of particles of low energy (0.1 to
25 MeV/u), medium energy (25 to 200 MeV/u) and high
energy (> 200 MeV/u) in the GCR LET spectrum. We can
see in the Figure that high energy particles dominate the
low LET part of the spectrum, but are negligible for LET
greater than about 3 MeV/(mg/cm 2 ). The low energy
particles, (the particles having similar energy range to
those often used for SEE testing) dominate the high LET
spectrum from about 10 MeV/(mg/cm 2 ) to 100
MeV/(mg/cm 2 ). The medium energy particles dominate the
LET spectrum from about 1 to 10 MeV/(mg/cm 2 ).
fraction of the integral fluence
1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
> 200 MeV/u
25 to 200 MeV/u
0.0
0.1 1.0 10.0 100.0
LET (MeVcm 2 /mg)
0.1 to 25 MeV/u
Figure 7: repartition of low, medium and high energy ions in the GCR
LET spectrum.
3

Above a LET of 10 MeV/(mg/cm 2 ), space heavy ions
have low or medium energy, and their range is limited. For
example, for the Fe ion, above a LET of 28
MeV/(mg/cm 2 ), the energy is less than 1.6 MeV/u and the
range in silicon is less than 18 μm. As can be seen in
Figure 3, Fe dominates the GCR spectrum for LETs
between 10 and 28 MeV/(mg/cm 2 ). This implies that most
particles with LETs in this range that reach the sensitive
structure have a limited penetration range, which turns out
to be between 18 and 400 μm as shown in Fig. 5. This is
consistent with the penetration range of particles used for
ground testing for this LET range.
The most significant part of the LET spectra for
advanced microcircuits is the range between 0.1 to 10
MeV/(mg/ cm 2 ). Most sensitive devices have a threshold
LET less than 1 MeV/(mg/cm 2 ) and their cross section
reaches the saturation before 10 MeV/(mg/cm 2 ). It has
been shown that the threshold LET of microcircuits with
sensitive volume dimensions less than micrometers has not
changed significantly with the device scaling [2].
Therefore, except for optoelectronic devices, we do not
expect in the near future threshold LET lower than 0.1
MeV/(mg/cm 2 ), which implies that SEE can be produced
by direct ionization due to protons.
Considering the incident particle energies, the most
significant energy range is the medium energy range from
tens of MeV/u to 200 MeV/u.
Maximum energy (MeV/u)
1000
100
10
1
0.1
TAMU
Fe
GANIL
200 MeV/u
25 MeV/u
0 10 20 30 40 50 60 70 80 90
Particle atomic number
Figure 8: available beams in facilities used for SEE testing.
IPN
LBL
The Fig. 8 shows the maximum beam energies available
in facilities used for SEE testing. We can see that the most
currently used facilities (BNL, UCL, IPN) with a maximum
beam energy lower than 10 MeV/u do not cover this energy
range at all. LBL and TEXAS A&M facilities cover a part
of this range. GANIL and MSU facilities give the best
coverage of the medium energy range, but it is generally
difficult and expensive to get some beam for SEE testing in
these facilities.
UCL
BNL
MSU
IV. CONCLUSION
We have reviewed how the SEE test environment relates
to the real space environment. Low energy beams are not
representative of the most significant part of the space
environment for modern microcircuits. Medium energy
beams are a better representation of the space SEE
environment, and they also give longer ranges to allow
access to the sensitive volumes of modern devices. For
these reasons medium energy beams are more and more
necessary for SEE testing. These beams are not well
covered in the existing facilities currently used for SEE
testing.
V. REFERENCES
[1] R. Koga, “Single Event Effect Ground test issues,” IEEE TNS, vol.
43, n°2, 661-670, April 1996.
[2] A. H. Johnston, “Radiation Effects in Advanced Microelectronics
Technologies,” RADECS 1997 Proceedings, pp. 1-16.
[3] E. G. Stassinopoulos, “Shortcomings in Ground Testing,
Environment simulations, and Performance Predictions for Space
Applications,” RADECS 1991 Proceedings, pp.3-16.
[4] S. Duzellier, R. Ecoffet, “Recent Trends in Single Event Effect
Ground Testing,” ,” IEEE TNS, vol. 43, n°2, pp. 671-677, April
1996.
[5] A.J. Tylka & al., “Single Event Upsets caused by Solar Energetic
Heavy Ions,” ,IEEE TNS, NS43, pp. 2758, Dec. 1996.
[6] J. Barth “Modeling Space Radiation Environment” 1997 IEEE
NSREC short course.
[7] W. Heinrich, “Calculation of LET Spectra of Heavy Cosmic Ray
Nuclei at Various Absorber Depths”, Radiation Effects, vol. 34,
pp.143-148, 1977.
[8] E.G. Stassinopoulos & al. ,“SEU measurements at the BNL test
facility,” Trans. Am. Nucl. Soc., vol. 62, pp. 230-232, 1990.
[9] M.A. Mc Mahan, ‘Radiation Effects Testing at the 88 inch
Cyclotron” RADECS 1999 proceedings, pp. 142-147.
[10] G. Berger & al., “The Heavy Ion Irradiation Facility at CYCLONE-
A dedicated SEE beam line,” 1996 IEEE NREC data workshop, pp.
78-83.
[11] M. Labrunee & al., “Heavy ion component test coordination,”
RADECS 1991 proceedings, pp. 577-581.
[12] M.A. Xapsos, “Applicability of LET to Single Events in
Microelectronic Structures”, IEEE Trans. Nucl. Sci., vol. 39, pp.
1613-1621, Dec. 1992.
[13] D.K. Nichols & al., “update of Single Event failure in power
MOSFETs,” IEEE NSREC 1996 dataworkshop, pp. 67-72.
[14] J. C. Pickel, “Single Event Effects Rate Prediction,” IEEE Trans.
Nucl. Sci., vol. 43, n°2, April 1996.
[15] H. Dussault, “The effects of ion track structure in simulating Single
Event Phenomena,”, Proceedings of RADECS 1993, pp. 509-516.
[16] R. Koga & al., “Comparative SEU sensitivities to relativistic heavy
ions,” IEEE TNS, vol. 45, n°6, pp. 2475-2482, 1998.
[17] P.E. Dodd & al. “Impact of ion energy on Single event Upset,”
IEEE TNS, vol. 45, n°6, pp. 2483-2491, 1998.
[18] S. Duzellier & al. “SEE results using high energy ions,” IEEE
TNS, vol. 42, n°6, pp. 1797-1802, 1995.
[19] R. Ecoffet “Low LET cross section measurements using high
energy carbon beam,” IEEE TNS, vol. 44, n°6, 1997.
[20] R. Koga & al., “SEE sensitivity determination of high density
DRAMs with limited range heavy ions,” IEEE NSREC 2000 data
workshop, pp45-52.
4