You are on page 1of 5

Radiation Physics and Chemistry 161 (2019) 29–33

Contents lists available at ScienceDirect

Radiation Physics and Chemistry


journal homepage: www.elsevier.com/locate/radphyschem

Microdosimetry approach for determining single-event upsets in T


microelectronic devices
Amir Moslehi
Radiation Applications Research School, Nuclear Science and Technology Research Institute, AEOI, P.O. Box: 11365-3486, Tehran, Iran

A R T I C LE I N FO A B S T R A C T

Keywords: Decreasing the dimension of p-n junctions in microcircuits down to micron sizes and the analogy to the di-
Microdosimetry mensions of biological cells and sub-cell components, provides a motivation for using microdosimetry in mi-
Single-event upset (SEU) croelectronics. Since single-event upsets (SEU) are the most important errors in microelectronic devices, in this
Microelectronic device work, the microdosimetric one-hit detector model was evaluated to determine the SEU cross-section for mi-
croelectronic devices. There were three main assumptions: 1) the device cells are the sensitive volumes. 2) The
cells possess the same critical charges and 3) the probability of SEU induction in a cell is a step function.
Moreover, the critical charge was considered as the free parameter of the model. Therefore, the number of upsets
was obtained equal to the number of effective particle traversals (events) in the device, derived from the single-
event microdosimetric distributions. Then, the SEU cross-section was determined. In the calculations there was
no need to the electronic response of the device.
To examine the model, a 1 k (i.e. 210 bytes memory) 65 nm CMOS bulk SRAM was considered for which the
SEU cross-section due to low energy protons obtained by RPP model and experiment had been reported in the
literature. Each cell was a rectangle with a dimension of 1.2 μm × 0.6 μm × 1 μm. The single-event distributions
for 9 proton energies from 2 MeV to 10 MeV were calculated by Geant4 simulation toolkit. Obtained results
showed that for a critical specific energy of 70 Gy (equivalent to a critical charge of 35.56 fC), the SEU cross-
sections were found in the best agreement with the reported experimental values within a maximum difference
about 40%. Also, the results were comparable with those obtained by the conventional Rectangular
Parallelepiped (RPP) model. Eventually, the microdosimetry approach proposed in this work may be considered
a potential method to determine the SEU cross-section in any microelectronic device.

1. Introduction which an energy more than a critical value is required to be deposited


in the device leading to an error. SEE can be divided into two categories
Since 1950's, significant developments have been occurred in elec- of soft errors (single event upset, SEU) and hard errors. When a soft error
tronics technology. The physical size of microcircuits has been de- happens, the device experiences a transient upset which can be resolved
creasing and more and more transistors can be placed in a chip. by rewriting the device and it can be used again. Alternatively, if a hard
Microelectronic devices have been used in many fields of technology, error happens, the device cannot be used any more. Generally, the more
e.g. spacecrafts and satellites. When utilized in radiations environment, decrease in the size, the more decrease in the critical energy and the
these devices may be damaged or may experience some errors in their more device sensitivity to ionizing radiation (Messenger and Ash,
functionality. Sensitive regions in the microelectronic devices are 1997).
chiefly p-n junctions made of semiconductors. Depend upon the device In 1962, it was realized that packing density of transistors in an
application and its physical and electrical characteristics, the damages integrated circuit would cause individual transistors to suffer from
or errors may be induced by a collection of particles (collective effects) damage due to cosmic rays (Messenger and Ash, 1997). In 1975, soft
or by a single particle (single-event effect, SEE) (Bagatin and Gerardin, errors due to cosmic rays in the electronics of spacecrafts were con-
2015). The collective effects include total ionizing dose effect and dis- firmed. It was concluded that the observed errors in satellites (about
placement damage which change irreversibly the functional properties 10−3 error per transistor per year) were comparable to those expected
and hence permanently damage the device. The SEE originates from the to be induced by galactic cosmic rays (Binder et al., 1975). Considering
concept of the first order model (threshold model) (McNulty, 1993), in the fluence of space radiations and the size of microelectronic devices,

E-mail address: ammoslehi@aeoi.org.ir.

https://doi.org/10.1016/j.radphyschem.2019.03.046
Received 2 December 2018; Received in revised form 17 February 2019; Accepted 29 March 2019
Available online 31 March 2019
0969-806X/ © 2019 Elsevier Ltd. All rights reserved.
A. Moslehi Radiation Physics and Chemistry 161 (2019) 29–33

SEE was known the dominant error in the satellites and spacecrafts. detector model is evaluated in order to determine SEUs in microelec-
There are too many works on SEE and SEU in the literature. For tronic devices. SEU cross-sections for a 1 k 65 nm CMOS bulk SRAM (as
instance, cosmic rays errors in MOS devices were investigated (Pickel an example) due to low energy protons are calculated and compared
and Blandford, 1980). Some geometric considerations such as Earth's with the conventional RPP based-simulation and experimental data
shadow and angular dependence of device sensitivity to SEU were reported in the literature (Wu et al., 2017). In the calculations only
studied (Letaw et al., 1985). Mechanism of charge collection during interaction of radiation with the device is concerned and its electronic
SEU in microelectronic devices was modeled (McNulty et al., 1991). response is not included. Also, for simplicity multiple cell upsets
Rate of single-event errors and their geometrical dependencies were (MCUs) are ignored.
also surveyed (Petersen et al., 1992, 1993). Shape of the SEU cross
section induced by heavy ions in SRAMs were determined (Xapsos 2. Theory
et al., 1993). In addition, the impact of ion energy on SEU in CMOS
SRAMs was explored (Dodd et al., 1998). It was found that there was no 2.1. Microdosimetry
significant difference between the device response to both high and low
energy ions. Mechanism of destructive types of SEE in silicon devices In this section the required microdosimetry concepts were briefly
and ICs were studied (Sexton, 2003). Simulation of SEE based on its reviewed. Microdosimetry deals with energy deposition in microscopic
physical mechanism was performed (Dodd, 2005). Also, a Monte Carlo volumes exposed to ionizing radiations. Despite the conventional do-
simulation of SEE revealed that event-to-event variations have sig- simetry, it applies stochastic quantities that their probability distribu-
nificant role in understanding these errors (Weller et al., 2010). More tions represent the pattern of energy deposition events (ICRU, 1983). By
recently, SEU induced by electrons in the nanometer-technology de- definition, an event is production of statistically correlated points
vices was investigated (Caron et al., 2018; Inguimbert et al., 2015). where the energy is transferred, e.g. track of a charged particle and/or
Gorbunov and Boruzdina presented a semi-empirical method for esti- its secondary electrons.
mation of SEU for SRAM dice cells (Gorbunov et al., 2016). Moreover, The microdosimetric quantity used in this paper is the specific energy
SEE in the stratospheric radiation environment was modeled (Hands which is defined as the ratio of energy imparted to the volume con-
et al., 2017) and SEU sensitivity of SOI SRAMS were studied (Raine sidered, ε and the mass of the volume, m (ICRU, 1983):
et al., 2018). ε
The key concept conventionally used for predicting SEU is the linear z=
m (1)
energy transfer (LET) of the incident particle. It is usually supposed that
the microelectronic devices contain a number of sensitive volumes as The specific energy may be considered as the microscopic coun-
rectangular parallelepiped (RPP) shapes. On this basis, two models are terpart of the absorbed dose and is in the unit of gray (Gy). The energy
developed: the chord length model and the effective flux model (Petersen imparted to the volume may be due to one or more events. Supposing
et al., 1992; Pickel, 1996). The first is a microscopic viewpoint relying the events to be Poisson distributed, the general probability distribution
on the chord length distributions to calculate the number of interactions of z is written as:

that can upset the cells in terms of LET distribution. The second is a nk
macroscopic viewpoint which treats the device as a black box which f (z; n) = ∑ e−n × f (z )
k! k
k=0 (2)
applies the test data to predict SEU rate on the basis of the measurement
in ground test, without detailed concern for individual cells (Pickel, where n is the mean number of events and fk (z ) is the distribution of z
1996). corresponding to exactly k events given in the following equation:
The substantial limitation of LET is the lack of concerning the par- z
ticle track structure and the statistical nature of ionization (ICRU, fk (z ) = ∫ f1 (z′) fk−1 (z − z′) dz′
1983). Therefore, Xapsos et al. (1993) used the critical charge as a 0 (3)
pertinent quantity, regarding the stochastic phenomena in charge de- Eq. (3) describes that higher order distributions of z can be de-
position. His model reproduced the SEU cross-section curve in agree- termined by the convolution of single-event distribution, f1 (z ) . Since n
ment with the experimental data (Xapsos et al., 1993). Also, the effect ∞

of radial dose around the heavy ions tracks on the SEU cross-section is related to the absorbed dose D [i.e. n = D / z¯F with z¯F = ∫ zf1 (z ) dz as
0
was investigated (Boorboor et al., 2015). It was shown that heavy ions the first moment of f1 (z ) ], one can use f (z; D) instead of f (z; n) . Note
with identical LET may have different SEU cross-section because of that the distributions of specific energy depend strongly on the geo-
different delta-ray and related radial dose distributions. metry of the microscopic volume of interest.
On the other hand, the analogy of the SV size in the microelectronic
devices to the size of biological cells and sub-cell components, provided 2.2. One-hit detector model and SEU cross-section
a motivation in a few works to apply microdosimetry to microelec-
tronics. In 1987, the microdosimetric spectra of space mesons and pions In accordance with the microdosimetric one-hit detector model, it
was measured using a silicon proportional counter and their correlation was assumed that a microelectronic device contains a large number of
to soft errors in memories was discussed (Dicello, 1987). In another similar SVs. In the first step, the yield of device survival from SEU was
work in 2001, the relative variance of energy deposition in microscopic determined. After irradiation of the device with the absorbed dose D,
semiconductor volumes was investigated (Xapsos et al., 2001). the yield of survival from SEU was determined by Eq. (4):
The fact that traversal of a single charged particle may induce SEU
in a microelectronic device, is obviously similar to the microdosimetric SSEU (D) = ∫ ps (z ) f (z;D) dz (4)
one-hit detector model (Zaider, 1990). It assumes that for some passive
detectors containing a big number of similar SVs, traversal (hit) of a in which ps (z ) was the survival probability of a SV and f (z; D) was the
single charged particle through a SV is sufficient to provide the detector distribution of z in the device. According to the threshold model, the
response. Consequently, the response can be deduced from the single- survival probability was approximated by a step function as follows:
event microdosimetric distribution determined in the SV. The model 1, z < z c
has been used for calculating the energy response of thermo- ps (z ) = ⎧
⎨ 0, z ≥ z c (5)
luminescence dosimeters (TLD) (Olko, 2002) and the radiographic films ⎩
(Moslehi et al., 2017). where z c was the critical specific energy corresponding to the critical
In the present work, for the first time the microdosimetric one-hit charge, Qc (Qc = 6.25 × 1018 z c me / w , with z c in Gy, m as the SV mass in

30
A. Moslehi Radiation Physics and Chemistry 161 (2019) 29–33

kg, e the electron charge in C and w = 3.6 eV, the energy needed for critical charges).
producing an electron-hole pair in Si). Thus, the survival yield changed Protons of 2, 3, 4, 5, 6.3, 7, 8, 9 and 10 MeV energy were considered
to: as the incident particles. Among them, the experimental SEU cross-
zc section was available only for 2, 6.3 and 10 MeV protons (Wu et al.,
SSEU (D) = ∫ f (z; D) dz 2017). The particles emitted form a thin circular surface, located at a
0 (6) distance of 0.5 m, mono-directionally impinging on the top side of the
SRAM. Radius of the source was big enough to irradiate the SRAM
Inserting Eq. (3) in (2) and placing the obtained result in Eq. (6),
completely.
together with doing some mathematics, SSEU (D) was derived in the
The “QGSP_BERT_HP” and “LowE_Livermore” classes were used in
form of an exponential equation in terms of the single-event distribu-
Geant4 for modeling the hadronic and electromagnetic interactions,
tion f1 (z ) (Zaider, 1990):
respectively. A range cut-off equal to zero was also set to ensure pro-
⎛ D zc ⎞ ∞ duction of low energy secondary particles. The number of protons
⎛ ⎞
SSEU (D) = exp ⎜− ⎡1 − ∫ f1 (z ) dz⎤⎥ ⎟⎟ = exp ⎜− zD¯F ∫ f1 (z ) dz ⎟ considered for any energy was 1.7 × 107, leading to a fluence of
⎜ z¯F ⎢ 2.68 × 1012 cm−2. Thus, relative statistical error in total energy de-
⎝ ⎣ 0
⎦⎠ ⎝ zc ⎠ (7)
∞ ∞ position in the whole sensitive layer was found less than 0.1%. The
While ∫ f1 (z ) dz = 1 and z¯F = ∫ zf1 (z ) dz . space between the source and the SRAM was assumed to be the va-
0 0 cuum.
In the second step, the number of upsets (SEU yield) was calculated
To calculate the single-event distribution of specific energy for any
as the inverse yield of survival, i.e.:
proton energy, energy imparted to the SRAM cells corresponding to any
⎛ D

⎞ single-particle event was calculated and divided by the cell mass
NSEU (D) = 1 − SSEU (D) = 1 − exp ⎜−
z¯F
∫ f1 (z ) dz ⎟ (1.82 × 10−15 kg) (ICRU, 1983). An algorithm was written in Geant4 to
⎝ zc ⎠ (8) make 250 logarithmic specific energy bins from 10−2 Gy to 103 Gy
(with 50 logarithmic bins per decade) and to place any specific energy
For low absorbed doses where D < < z̄F (ICRU, 1983), the number
in its relevant bin. When all the particles were emitted, the value of
of upsets per unit of absorbed dose was approximated as:
zf1 (z ) was calculated for each bin where z and f1 named the average

1 specific energy value and the probability distribution of events in that
NSEU ≅
z¯F
∫ f1 (z ) dz bin, respectively. After providing the distributions, z̄F , NSEU and σSEU
zc (9)
were derived.
In this equation, 1/ z¯F denoted the number of all particle traversals
(events) per unit of absorbed dose and the integral specified the prob- 4. Results
ability of depositing specific energy in excess of z c per event.
Equivalently, the integral meant the probability of SEU. Thus, Eq. (9) 4.1. Single-event distributions
implied the fraction of events resulting in SEU (i.e. effective events).
Finally, the SEU cross-section in unit of cm2 bit−1 was calculated by Eq. Fig. 1 shows the normalized single-event distributions of specific
(10): energy for 2, 6.3 and 10 MeV protons calculated in the cells of 1 k
NSEU 65 nm CMOS bulk SRAM. The area under the distributions were pro-
σSEU = portional to the number of events occurred in the cells. For the three
φS (10)
above energies, the specific energies corresponding to registered events
in which φ denoted the fluence of incident particles and S specified the were between 0.01 Gy and about 250 Gy. Although the distributions
number of cells in the device (its size). Equations (9) and (10) declared were attributed to the primary incident proton energies, the protons hit
that the only requirement to determine SEU in a device is the single- the cells with a modified energy spectrum accompanying secondary
event distribution, f1 (z ) , in its SVs. It must be mentioned that the cri- particles (electrons, silicon and oxygen ions), due to the presence of 20
tical specific energy, z c , was considered as the free parameter of the layers above the sensitive region. Indeed, the energy imparted to the
model which would be found by fitting the calculated cross-section to cells was provided by the direct ionization of protons as well as the
the experimental data. Moreover, the model was based on the as- indirect ones. In the latter case, recoil ions of silicon and oxygen were
sumption that the SV geometry to be fixed for all hitting events. It was a produced during the elastic and inelastic collisions of protons with the
regulatory for determination of microdosimetric quantities and their silicon and oxide layers. The energy distributions and contributions of
distributions in the regional microdosimetry language (ICRU, 1983). these ions in SEU induction for different proton energies have been
investigated before (Wu et al., 2017). They found that the maximum
3. Methods

In order to evaluate the model given in the previous section, a 1 k


65 nm CMOS bulk SRAM (Wu et al., 2017) was considered. The SRAM
geometry was simulated by version 10.00 of Geant4 toolkit. It included
a sensitive layer of silicon with an effective thickness of about 1 μm,
containing two adjacent 32 × 16 arrays of cells. Each cell was a rec-
tangle with the dimension of 1.2 μm × 0.6 μm × 1 μm (Wu et al.,
2017). A 500 μm silicon substrate was placed beneath the arrays. Also,
20 alternate layers of Cu and SiO2 were located above the arrays.
Wu et al. determined the SEU cross-section by experiment and also
simulation based on the RPP model. They characterized 13 rectangular
SVs in each SRAM cell (8 NMOS and 5 PMOS transistors) by using 3D
TCAD simulation. Since based on the regional microdosimetry concept,
only a fixed SV had to be used for calculation of microdosimetric dis- Fig. 1. Normalized single-event distributions of specific energy for 2 MeV,
tributions, each SRAM cell was considered as a SV. In addition, it was 6.3 MeV and 10 MeV protons calculated in the cells of a 1 k 65 nm CMOS bulk
assumed that all the cells are similarly sensitive to radiation (i.e. similar SRAM. The distributions are plotted in log-log scale to show all details.

31
A. Moslehi Radiation Physics and Chemistry 161 (2019) 29–33

Fig. 2. Variation of z̄F with the proton energy calculated in the cells of a 1 k Fig. 3. Least square fitting of the SEU cross-sections calculated by the micro-
65 nm CMOS bulk SRAM. By increasing the energy, z̄F gradually decreased. dosimetry model to the experimental data (Wu et al., 2017) for 2, 6.3 and
10 MeV protons, considering a few critical specific energies. The best fit oc-
curred at z c = 70 Gy.
kinetic energy of the recoil ions produced by sub-10 MeV protons is less
than 2 MeV. The calculations in this work revealed that the proton
events (direct ionizations) dominated in z < 100 Gy region, whereas Table 1
the ion events (indirect ionizations) dominated chiefly in z > 100 Gy Calculated SEU cross-sections in cm2 bit−1 (with one standard deviation) in-
duced in a 1 k 65 nm CMOS SRAM (z c = 70 Gy), compared with the simulation
region. Because the stopping power of recoil ions was bigger than that
and experimental data reported by Wu et al. (2017) for different proton en-
of protons and consequently more energy was imparted to the cells.
ergies.
Therefore, the specific energies corresponding to the recoil ions got
bigger values than protons. The behavior of f1 (z ) with the proton en- Proton Energy Microdosimetry RPP Model Experiment
(MeV) (This Work) (Wu et al., 2017) (Wu et al., 2017) a
ergy can also be seen in Fig. 1. When the energy increased, the dis- a

tributions shifted to smaller specific energies due to decrease of the


average stopping power of protons. This caused a gradual decrease in z̄F 2.0 (3.0 ± 0.2) × 10−15 2.8 × 10−15 2.5 × 10−15
values by increasing the energy. As illustrated in Fig. 2, z̄F values re- 3.0 (2.2 ± 0.1) × 10−15 2.5 × 10−15 –
4.0 (2.0 ± 0.3) × 10−15 2.7 × 10−15 –
duced from 18.3 to 4.4 as the proton energy increased from 2 MeV to
5.0 (2.2 ± 0.3) × 10−15 2.3 × 10−15 –
10 MeV. Along with the shift to smaller specific energies, the distribu- 6.0 – 2.8 × 10−15 –
tions became wider because of the increase of straggling in the energy 6.3 (4.6 ± 0.3) × 10−15 – 4.0 × 10−15
imparted to the cells. 7.0 (2.7 ± 0.2) × 10−15 3.0 × 10−15 –
8.0 (3.9 ± 0.3) × 10−15 4.0 × 10−15 –
9.0 (5.7 ± 0.5) × 10−15 4.8 × 10−15 –
4.2. SEU cross-section 10.0 (8.6 ± 0.6) × 10−15 5.4 × 10−15 1.5 × 10−14

a
As mentioned in section 2.2, z c was assumed as the free parameter The values are deduced from the cross-section curve and are approximate.
of the microdosimetry model. To find z c , the SEU cross-sections due to
2, 6.3 and 10 MeV protons for 7 values of z c between 40 Gy and 100 Gy
were calculated using Eq. (10). Then, the least square fitting of these
values to their corresponding experimental data was determined by Eq.
(11):
2
σical . − σiexp . ⎞
F= ∑ ⎜⎛ σiexp .

i ⎝ ⎠ (11)

where σical.
and σiexp.
were the calculated and experimental SEU cross-
sections, respectively and index i (= 1, 2 and 3) referred to three above
energies. Fig. 3 shows the variation of F with z c . As can be observed, F
reached to its minimum value at z c = 70 Gy. Therefore, this value was
chosen as the desired critical specific energy of each SRAM cell. It was
equivalent to a critical charge of 35.5 fC. Note that TCAD simulation of
Fig. 4. The SEU cross-sections for a 1 k 65 nm CMOS SRAM calculated by the
Wu et al. (2017) performed for the studied SRAM resulted in the total
microdosimetry model (z c = 70 Gy) along with the experimental and RPP si-
critical charges of 1.6 fC and 2.4 fC for the internal NMOS and PMOS mulation data (Wu et al., 2017) for low energy protons.
transistors, respectively. It was expected to get a bigger critical charge
when the cell was chosen as the SV.
Table 1 lists the SEU cross-sections calculated by the micro- 10 MeV the microdosimetry model made a better agreement to the
dosimetry model beside the experimental and simulation values given experiment, with a smaller maximum difference than the RPP model
by Wu et al. (2017). The cross-sections for all the proton energies are (i.e. 43% compared with 63%). In addition, a comparison between the
plotted in Fig. 4. For 2 and 6.3 MeV, microdosimetry model over- microdosimetry model and RPP model declared that from 2 to 9 MeV
estimated the SEU cross-section by a difference of 20% and 15%, re- the both models agreed within a maximum difference of 26%, while for
spectively. For 10 MeV this model underestimated the experimental 10 MeV their discrepancy was 60%.
cross-section by a difference of 43%. At 2 MeV, RPP model performed
by Wu et al. (2017) provided slightly better agreement with the ex-
periment than the microdosimetry model. Otherwise, for 6.3 and

32
A. Moslehi Radiation Physics and Chemistry 161 (2019) 29–33

5. Discussion 6. Conclusion

In this part of the paper let compare the microdosimetry model with In this paper, the microdosimetric one-hit detector model was
the conventional RPP model. In the RPP model, before estimation of evaluated in order to determine the SEU cross-section in microelec-
SEU some electronic simulations are necessary (Wu et al., 2017). The tronic devices. Supposing each device cell as a SV with no need to the
transistors inside the cells should be calibrated and their electrical electronic response of the device, the number of upsets was determined
characteristics (e.g. current-voltage curves for the drains and gates) using the single-event microdosimetric distributions. The cross-sections
need to be determined. Then, the dimensions of the possible SVs in a calculated for a 1 k 65 nm CMOS bulk SRAM due to low energy protons
cell and their charge collection coefficients must be specified. On the were found compatible with the both experimental and RPP-simulation
other hand, in the microdosimetry model a simpler vision is confronted. data reported in the literature. Since the microdosimetry approach
There is no need to know the electrical information inside any cell. Only proposed in this work is quite general, it can be considered a potential
the pattern of energy deposition in the cells (considered as the SVs) is tool for determination of SEU in any other microelectronic device.
enough to estimate the upsets. This pattern as the specific energy dis-
tribution includes the stochastic nature of energy deposition such as References
changing the track length of particle traversals, changing the particles
LET and also energy straggling. The critical charge is a free parameter Bagatin, M., Gerardin, S., 2015. Ionizing Radiation Effects in Electronics: from Memories
and can be properly set. The microdosimetry model is a general method to Imagers. CRC Press.
Binder, D., Smith, E.C., Holman, A., 1975. Satellite anomalies from galactic cosmic rays.
and can be expected to work satisfactorily for any other device. As IEEE Trans. Nucl. Sci. 22, 2675–2680.
mentioned before, the basic assumptions of this model are: Boorboor, S., Feghhi, S., Jafari, H., 2015. Investigation of radial dose effect on single
event upset cross-section due to heavy ions using GEANT4. Radiat. Meas. 78, 42–47.
Caron, P., Inguimbert, C., Artola, L., Chatry, N., Sukhaseum, N., Ecoffet, R., Bezerra, F.,
1 The device cells were considered as the SV. 2018. Physical mechanisms inducing electron single event upset. IEEE Trans. Nucl.
2 All the cells possessed the same critical charges. Sci. 65 (8), 1759–1767.
3 The survival probability of a cell was considered ideally as a step Dicello, J., 1987. Microelectronics and microdosimetry. Nucl. Instrum. Methods Phys.
Res. Sect. B Beam Interact. Mater. Atoms 24, 1044–1049.
function.
Dodd, P., Musseau, O., Shaneyfelt, M., Sexton, F., D'hose, C., Hash, G., Martinez, M.,
Loemker, R., Leray, J.-L., Winokur, P., 1998. Impact of ion energy on single-event
The first assumption is discussed above. One point that can be added upset. IEEE Trans. Nucl. Sci. 45, 2483–2491.
Dodd, P.E., 2005. Physics-based simulation of single-event effects. IEEE Trans. Device
is when the device is partially depleted; the funneling effect inhibits to
Mater. Reliab. 5, 343–357.
define a well-defined geometry for the p-n junction. Therefore, con- Gorbunov, M.S., Boruzdina, A.B., Dolotov, P.S., 2016. Semi-empirical method for esti-
sidering the cell as the SV can solve this problem. mation of single-event upset cross section for SRAM dice cells. IEEE Trans. Nucl. Sci.
Although the second assumption works reasonably, in many modern 63, 2250–2256.
Hands, A., Lei, F., Ryden, K., Dyer, C., Underwood, C., Mertens, C., 2017. New data and
devices the cells despite their similar geometry are unequally sensitive modelling for single event effects in the stratospheric radiation environment. IEEE
to radiation and their critical charges are different. Hence, distribution Trans. Nucl. Sci. 64, 587–595.
of these charges must be included in the calculations. Xapsos et al. ICRU, 1983. Microdosimetry. ICRU Report 36.
Inguimbert, C., Ecoffet, R., Falguère, D., 2015. Electron induced SEUs: microdosimetry in
(1993) mentioned that when the number of cells is sufficiently large, a nanometric volumes. IEEE Trans. Nucl. Sci. 62, 2846–2852.
Gaussian distribution can be related to their critical charges. Equiva- Letaw, J.R., Silberberg, R., Tsao, C., Adams, J.H., 1985. Geometric considerations in
lently, such a distribution may be applied to the critical specific en- single event upset estimation. IEEE Trans. Nucl. Sci. 32, 4212–4215.
McNulty, P., 1993. Single event upsets in microelectronics and similar radiation effects in
ergies: biological systems. Radon Monit. Radioprotection, Environ. Earth Sci., INIS 24 (17).
2
McNulty, P., Abdel-Kader, W., Lynch, J., 1991. Modeling charge collection and single
1 ⎡ 1 z − z¯c ⎞ ⎤ event upsets in microelectronics. Nucl. Instrum. Methods Phys. Res. Sect. B Beam
Pz c = exp ⎢− ⎜⎛ c
2 ⎝ Szc ⎠ ⎥
⎟ Interact. Mater. Atoms 61, 52–60.
2π S z c (12)
⎣ ⎦ Messenger, G.C., Ash, M.S., 1997. Single Event Phenomena I, Single Event Phenomena.
Springer, pp. 179–231.
in which z̄ c is the mean critical specific energy and S z c is the standard Moslehi, A., Raisali, G., Hamidi, S., Gheshlaghi, F., 2017. The microdosimetric one-hit
deviation of this distribution. Both z c and S z c are now the fitting detector model to calculate the energy response of radiographic films. Radiat. Meas.
105, 33–38.
parameters. Thus, the number of upsets in the device can be obtained Olko, P., 2002. The microdosimetric one-hit detector model for calculating the response
by integrating Eq. (9) over all z c values, i.e.: of solid state detectors. Radiat. Meas. 35, 255–267.
Petersen, E., Pickel, J., Adams, J., Smith, E., 1992. Rate prediction for single event effects-
∞ a critique. IEEE Trans. Nucl. Sci. 39, 1577–1599.
NSEU =
1
z¯F
∫ dz c Pz ⎛⎜∫ f1 (z ) dz⎞⎟
c
Petersen, E., Pickel, J., Smith, E., Rudeck, P., Letaw, J., 1993. Geometrical factors in SEE
rate calculations. IEEE Trans. Nucl. Sci. 40, 1888–1909.
⎝ zc ⎠ (13) Pickel, J.C., 1996. Single-event effects rate prediction. IEEE Trans. Nucl. Sci. 43, 483–495.
Pickel, J.C., Blandford, J.T., 1980. Cosmic-ray-induced errors in MOS devices. IEEE Trans.
Finally, the step function in the third assumption introducing the Nucl. Sci. 27, 1006–1015.
survival probability is ideal. Actually, there are complex phenomena Raine, M., Gaillardin, M., Lagutere, T., Duhamel, O., Paillet, P., 2018. Estimation of the
single-event upset sensitivity of advanced SOI SRAMs. IEEE Trans. Nucl. Sci. 65,
triggering an upset which lead to the step function to be no more 339–345.
adequate for describing the survival probability. Therefore, more rea- Sexton, F.W., 2003. Destructive single-event effects in semiconductor devices and ICs.
listic functions such as exponential or Weibull need to be considered. IEEE Trans. Nucl. Sci. 50, 603–621.
Weller, R.A., Mendenhall, M.H., Reed, R.A., Schrimpf, R.D., Warren, K.M., Sierawski,
Consequently, it imposes more fitting parameters to the model which B.D., Massengill, L.W., 2010. Monte Carlo simulation of single event effects. IEEE
depend on the device of interest. For instance, imagine ps (z ) = e−αz , in Trans. Nucl. Sci. 57, 1726–1746.
which α is a saturation parameter defining the probability of upset per Wu, Z., Chen, S., Yu, J., Chen, J., Huang, P., Song, R., 2017. Recoil-ion-induced single
event upsets in nanometer CMOS SRAM under low-energy proton radiation. IEEE
unit increment of specific energy. The number of upsets given by Eq. (9)
Trans. Nucl. Sci. 64, 654–664.
will change to Eq. (14): Xapsos, M., Summers, G., Burke, E., Poivey, C., 2001. Microdosimetry theory for micro-

electronics applications. Nucl. Instrum. Methods Phys. Res. Sect. B Beam Interact.
1 Mater. Atoms 184, 113–134.
NSEU =
z¯F
∫ (1 − e−αz ) f1 (z ) dz Xapsos, M.A., Weatherford, T.R., Shapiro, R., 1993. The shape of heavy ion upset cross
zc (14) section curves (SRAMs). IEEE Trans. Nucl. Sci. 40, 1812–1819.
Zaider, M., 1990. Microdosimetry and Katz's track structure theory: I. One-hit detectors.
where α is another parameter of the model. Radiat. Res. 124, S16–S22.

33

You might also like