An Analysis of the Significance of the 14N(n, p) 14C Reaction for Single-Event Upsets Induced by Thermal Neutrons in SRAMs

The thermal neutron threat to the reliability of electronic devices caused by <inline-formula> <tex-math notation="LaTeX">$^{10}\text{B}$ </tex-math></inline-formula> capture is a recognized issue that prompted changes in the manufacturing process of electronic devices with the aim of limiting as much as possible the presence of this isotope nearby device sensitive volumes (SVs). <inline-formula> <tex-math notation="LaTeX">$^{14}\text{N}$ </tex-math></inline-formula> can also capture thermal neutrons and release low-energy protons (LEPs; through the <inline-formula> <tex-math notation="LaTeX">$^{14}\text{N}$ </tex-math></inline-formula>(n, p)<inline-formula> <tex-math notation="LaTeX">$^{14}\text{C}$ </tex-math></inline-formula> reaction) that have high enough linear energy transfer (LET) to cause single-event upsets (SEUs). Typically, nitrogen is used in thin barrier layers made of TaN or TiN or even as insulator in the form of Si3N4. Numerical simulations on SVs calibrated on proton and ion experimental data and with an accurate description of the metallization layer on top of the sensitive region show that the presence of nitrogen in these thin barrier layers can be enough to justify the experimentally observed thermal neutron SEU cross Section for a static random access memory (SRAM) sensitive to LEPs. Nevertheless, the expected SEU cross Section from thermal neutrons is usually a few orders of magnitude lower than that of high-energy particles, therefore, not representing an important threat in atmospheric applications. At the same time, for high-energy accelerators, the contribution to the total soft error rate (SER) could become substantial, though easy to handle by margins.

An Analysis of the Significance of the 14 N(n, p) 14  Abstract-The thermal neutron threat to the reliability of electronic devices caused by 10 B capture is a recognized issue that prompted changes in the manufacturing process of electronic devices with the aim of limiting as much as possible the presence of this isotope nearby device sensitive volumes (SVs). 14 N can also capture thermal neutrons and release low-energy protons (LEPs; through the 14 N(n, p) 14 C reaction) that have high enough linear energy transfer (LET) to cause single-event upsets (SEUs). Typically, nitrogen is used in thin barrier layers made of TaN or TiN or even as insulator in the form of Si 3 N 4 . Numerical simulations on SVs calibrated on proton and ion experimental data and with an accurate description of the metallization layer on top of the sensitive region show that the presence of nitrogen in these thin barrier layers can be enough to justify the experimentally observed thermal neutron SEU cross section for a static random access memory (SRAM) sensitive to LEPs. Nevertheless, the expected SEU cross section from thermal neutrons is usually a few orders of magnitude lower than that of high-energy particles, therefore, not representing an important threat in atmospheric applications. At the same time, for high-energy accelerators, the contribution to the total soft error rate (SER) could become substantial, though easy to handle by margins.

I. INTRODUCTION
T HERMAL neutrons are an important contributor to any radiation environment where spallation reactions and neutron moderation are involved. This is the case of the atmosphere (ground level to flight altitudes) as well as of high-energy physics accelerators. Thermal neutrons can trigger single-event upsets (SEUs) in memory-based devices through indirect ionization following their nuclear absorption in 10 B atoms that may be present within or nearby sensitive cells, causing the release of an alpha and a Li ion. The 10 B(n, α) 7 Li nuclear reaction is illustrated in Fig. 1.
In the past [1], [2], [3], boron was employed in passivation layers for memory devices [borophosphosilicate glass (BPSG)]. Later, this process was abandoned. Nevertheless, boron is still used in semiconductor manufacturing in two other circumstances. In the first case, it is used as a dopant in p-type silicon. However, in this case, 10 B is filtered out from the implantation process by magnetic separation [4]. In the second case, it is used to reduce the resistance of the tungsten plugs [5] by deposition of B 2 H 6 , for which boron isotopic separation is not possible.
As shown in Fig. 1, the nuclear reaction releases two light ions, an alpha particle and a Li ion. In 94% of the cases, the alpha is produced with an energy of 1.472 MeV and the Li ion with an energy of 840 keV, and also, a γ photon with an energy of 480 keV is emitted. In the remaining 6% of the cases, the alpha is generated with an energy of 1.776 MeV and the Li ion with an energy of 1014 keV, and no photon is generated. Depending on the sensitivity of the device, either one or both of the secondary ions can deposit enough energy in the sensitive volume (SV) to cause an upset.
In this article, the focus is shifted on another material commonly employed in the manufacturing of memory-based devices, i.e., nitrogen. Nitrogen is used in the back-end-ofline (BEOL) as either a thin barrier layer around the metal layers (typically, TiN and TaN are used) or as insulator (Si 3 N 4 ). Therefore, given the state-of-the-art manufacturing technologies, much larger quantities of nitrogen (an actual constituent) can be expected in a memory device than of boron (more of a residual).
Nitrogen may be important, because thermal neutrons can be captured by 14 N atoms with the emission of a low-energy proton (LEP) with an energy of 584 keV (range in silicon of 7.46 µm [15]). This energy may further reduce as the proton travels through the BEOL, bringing the proton to enter the SV at an energy close to the Bragg peak (at around 50 keV) where the linear energy transfer (LET) is maximum. Therefore, given that many devices in the 28-90 nm sensitive node size have been shown in [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], and [31] to be very sensitive to LEPs, it is possible that they will also be significantly sensitive to thermal neutrons if nitrogen is used in their manufacturing process. Concerning the 14 N target nucleus, this turns into a 14 C product nucleus with an energy of just 42 keV, after the neutron capture and proton emission. Therefore, the isotope has insufficient range to reach the SV and become a hazard. This reaction is also shown in Fig. 1.
Despite the much larger abundance of nitrogen in a memory, the neutron capture cross section of 14 N is much lower than that of 10 B and of other isotopes, see Fig. 2, whose data were retrieved from the ENDF database [32]. For the thermal neutron energy, i.e., 25 meV, the 14 N capture cross section is as low as 1.8 barns (1 barn = 10 −24 cm 2 ), whereas the 10 B cross section is about 3844 barns, therefore ≈2000 times larger.
In this work, experimental thermal neutron SEU measurements on an SRAM highly sensitive to LEPs [31] and Monte Carlo simulations embedding details of the construction analysis of the metallic and insulating overlayers seem to confirm that LEPs created through the 14 N(n, p) 14 C reaction would indeed lead to upsets. In this respect, this article is further devoted to analyzing under which conditions nitrogen can become a threat as serious as that of boron for the soft error rate (SER) of a memory device in atmospheric and accelerator environments.

II. EXPERIMENTAL MEASUREMENTS AND CONSTRUCTION ANALYSIS
The targeted device is the CY62167GE30-45ZXI, which is commercially available from Infineon. The memory is based on a 65-nm bulk technology and has a total capacity of 16 Mbits and a 3.3-V I/O voltage. It shall be noted that the commercial version has an embedded error correction code (ECC) that was disabled for the purpose of the various experimental SEU measurements. This memory has been characterized in a number of different radiation fields other than thermal neutrons [33], out of which the marked LEPs sensitivity stood out [31].
Thermal neutron testing was achieved at the ILL D50 facility [34]. Therefore, high fluxes of thermal neutrons (10 9 n/cm 2 /s) are obtained by cooling down the fast neutrons emitted by the fission reactor (with a negligible fast neutron component, i.e., no SEUs were observed when covering the memory with 4 mm of B 4 C to remove the thermal neutron component completely). The experimentally measured SEU cross section upon irradiation of 16 devices is (4.76 ± 0.78) × 10 −16 cm 2 /bit. This SRAM was also tested with LEPs at CNA [31], [33], [35] with peak cross section >10 −9 cm 2 /bit. Finally, to assess the relative impact of thermal neutrons in actual environments, it is necessary to relate the corresponding thermal neutron SER to that from high-energy neutrons (HENs). Therefore, one unit of this SRAM was also tested at ChipIr with atmospheric neutrons [36], with a resulting SEU cross section of (1.16 ± 0.24) × 10 −13 cm 2 /bit. All error bars are calculated at 95% confidence level.
The structure of the BEOL of the memory was retrieved by means of scanning electron microscope (SEM) cross sectioning. Nitrogen was identified in thin barrier layers around copper and aluminum metal layers in TaN and TiN composites. In addition, some thin layers of Si 3 N 4 were also identified in the passivation layer, which is placed the farthest from the SV. Although the actual configuration of the BEOL is not just monodimensional, it is still possible to summarize one-dimensionally its various layers, as listed in Table I.
Note that the SEM analysis did not provide an exact measurement for the thickness of the individual TaN and TiN layers, but these were combined with those of Cu and Al metal layers. This is the reason why, in the table, the data are left in  TABLE I  LIST OF MATERIALS PRESENT IN THE BEOL RESULTING FROM THE SEM  ANALYSIS. THE SENSITIVE CELLS ARE PLACED AT THE BOTTOM OF  THE STACK. THICKNESS OF METAL AND THIN BARRIER LAYERS  WERE NOT GIVEN SEPARATELY. FOR THE PURPOSE OF THIS  STUDY, THE THICKNESS OF THE TIN AND TAN THIN  BARRIER LAYERS AS WELL AS THE METAL LAYERS  IS LEFT IN PARAMETRIC FORMAT, THROUGH  THE T PARAMETER parametric format with a generic thickness parameter T that will be varied throughout the numerical analysis. These thin barrier layers are placed at the top and at the bottom of each of these metal layers. In typical manufacturing processes, these thin barrier layers are obtained by depositing an amount of TaN or TiN in the order of 50-100 nm per 1 µm of metal. Therefore, as shown in the table, the thickness of these barrier layers is parametrized based on this ratio and on the overall thickness of the actual metal layer and its two thin barrier layers. For instance, taking the thin barrier layers to be 50-nm thick for 1 µm of metal, the resulting "TaN1above" and "TaN1below" layers in the table would be 7.5-nm thick and the "Metal1" layer 135-nm thick.
The SEM analysis did not reveal the presence of boron, and the manufacturer declares no boron is employed in the BEOL of the SRAM. Nevertheless, for the sake of the analysis, the presence of natural boron (80.1% 11 B and 19.9% 10 B) at the level of the tungsten plug will be considered with concentrations in the order of 10 17 -10 20 cm −3 [37].

III. NUMERICAL ANALYSIS FOR THIS DEVICE
A. Event-by-Event Scoring of LEPs G4SEE [38] Monte Carlo simulations can be used to shed light on whether the measured experimental SEU cross section is due to LEPs emitted by thermal neutron interactions with nitrogen nuclei. The characteristics of the SV of this SRAM were determined in a previous study [31] based on the fitting of experimental proton and ion SEU data.
The SV model is based on a 360-nm cubic rectangular parallelepiped (RPP) with 0.86-fC critical charge (19.35-keV critical energy). With respect to the previous study, a more detailed BEOL structure, which includes all the materials as listed in Table I, was introduced in the simulation geometry.
Simulations are performed including various amounts of nitrogen, by acting on the thickness parameter T in the range of 25-100 nm per 1 µm of metal, and of boron, by acting on the boron concentration in tungsten.
Some first simulations were done to understand whether the LEPs generated by the nitrogen in the BEOL would lead to energy deposition events in the SV that could exceed the critical charge/energy of the memory. The energy deposition distribution for various amounts of nitrogen is also compared with those arising when dispersing boron in the tungsten. Fig. 3 provides the event-by-event scoring deposited energy spectra of LEPs, alphas, and Li ions released by the nuclear reactions of 25-meV neutrons with the nitrogen and boron included in the simulation. The graph also shows which is the critical energy required to cause an SEU for this memory SV model. As the plot shows, LEP events can exceed the critical energy, and the amount of nitrogen does not significantly alter the spectra; i.e., only the counts per primary are reduced, but not the maximum deposited energies.
Alphas and Li ions can deposit higher levels of energy than the LEPs. The alpha peak occurs at an energy five times higher than for LEPs and the Li ion peak at an energy ten times higher.
As shown in Table II, given their low range, the LEPs reaching the SV mostly originate in the thin barrier layers that are the nearest to the SV. However, as proven by the "TaN5above" and "TaN5below" layers, a higher amount of nitrogen far from the SV can compensate for the longer distance to be covered to reach the SV. Energy deposition events from LEPs exceeding the critical charge are almost equally distributed among the three closest and two farthest TaN thin barrier layers. This means that even if there are fewer LEPs coming from the farther layers, these LEPs arrive at the SV after having lost much more energy in the following BEOL layers and have, thus, higher LET. As it can be seen, no LEPs from "TiNabove" and "Passivation" layer make it to the SV due to the insufficient range.
To provide further insight about the characteristics of the LEPs as a function of the layer where they have originated, Fig. 4 displays the respective kinetic energy spectra of the LEPs. As expected, even for the closest layers to the SV, the spectrum is not monochromatic, but it spreads over some tens of keV. The spread increases the farther the point of origin is from the SV, but it is only for LEPs generated in the Tan5above layer that the energy spectra positions around the proton Bragg peak energy of 50 keV. Fig. 5 shows the LET distributions of the LEPs reaching the SV as a function of the layer they are generated in. As the figure shows, the spread in terms of LET is quite small for the closest layers. The LET of all LEPs generated in all of the BEOL layers is almost always higher than  0.25 MeV/(mg/cm 2 ), which is enough to cause an upset in this SRAM as seen with carbon ion testing [31], [33].
Finally, Fig. 6 depicts the energy deposition distributions of the LEPs based on the layer where the LEPs originated. The y-axis of the plot has been cut at the level of the critical energy of the SRAM at the bottom side to show only data relevant to the triggering of SEUs. Note that, in accordance with the LET, the LEPs generated in the closest layers typically deposit in the SV amount of energies that are just slightly above the critical energy and that never exceed it by more than a factor of 2.
On the other hand, as the distance increases, the distribution widens, and the critical energy can be exceeded by even a factor of 4.

B. Nitrogen/Boron Required to Match the Experimental Thermal Neutron SEU Cross Section
The purpose of this section is to assess whether realistic amounts of nitrogen, as those previously defined, can lead to a sufficient amount of energy deposition events to justify the experimental thermal neutron SEU cross section or whether considering also the presence of boron would be required.
As previously mentioned, amounts of TaN (and TiN) in the thin barrier layers in the order of 50-100 nm per 1 µm of Cu (or Al) layer can be considered realistic. The analysis is, therefore, conducted by performing simulations with different values of the thickness parameter T , adapting accordingly the thickness of the thin barrier layers and of the metal layers. Simulations were performed using T = 25, 50, 75, and 100 nm per 1 µm of metal. Fig. 7 depicts the SEU cross section as a function of the critical charge for these values of the T parameter. Two lines also mark the critical charge of this SRAM SV model and the experimental thermal neutron SEU cross section as guidance to understand whether realistic amounts of nitrogen would yield the experimentally observed cross section.
The figure shows that the case for T = 75 nm of TaN per 1 µm of Cu passes exactly through the intersection between the critical charge and the experimental cross section. This confirms that even small and realistic amounts of nitrogen in thin barrier layers are expected to contribute significantly to the thermal neutron cross section through the generation of LEPs.
It is also quite clear from the figure that there seems to be rather direct proportionality between the amount of TaN and the SEU cross section, given that the T = 50-nm case has a cross section twice as high as that of T = 25 nm and twice as low as that for T = 100 nm.
Even if it was possible to obtain an excellent match with nitrogen alone, for the sake of the discussion, it is also relevant to try to insert natural boron into the tungsten plug and to assess if the combination of nitrogen and boron could also provide a satisfactory match. In this case, of course, the simulations are run assuming a lower amount of TaN that matching the experimental cross section, i.e., with T = 50 nm.
The simulation data including natural boron are also reported in Fig. 7 alongside those without any boron. Three natural boron concentrations have been used: 1.7 × 10 17 , 1.7 × 10 18 , and 1.7 × 10 19 cm −3 . The 10 B concentrations are just 19.9% of those of natural boron. The values of these concentrations have been chosen to find a good match for the experimentally measured thermal neutron SEU cross section in the case of T = 50 nm.
The figure shows that a uniform natural B concentration of 1.7 × 10 18 cm −3 , coupled with T = 50 nm, also satisfies the experimentally measured thermal neutron SEU cross section.
In this case, the contribution of LEPs from nitrogen to the total SEU cross section would be 71%, with just 29% due to the alphas and Li ions from boron.
Therefore, sticking at T = 50 nm, which is typically the minimum used in manufacturing processes, LEPs from nitrogen would still be the major contributor to the observed upsets, and even using a twice as lower amount (T = 25 nm) would result in at least 35% of the upsets coming from LEPs from nitrogen, i.e., not negligible.
A concentration of natural B of 1.7 × 10 17 cm −3 , even if yielding more energetic energy deposition events, would provide negligible contribution to the overall SEU cross section. On the other hand, if the natural boron concentration is raised to 1.7 × 10 19 cm −3 , then the events from nitrogen would be negligible, even when considering T = 100 nm. However, the SEU cross section would clearly become too large to match the experimental value. Therefore, it can be concluded that, if boron was present in this SRAM tungsten plug, its concentration would have to be limited to a few 10 18 cm −3 or lower.

C. Nitrogen and Boron Relative Criticality
When nitrogen is present only in the thin barrier layers, which can indeed be very thin, it does not take a very high concentration of natural boron in the tungsten plug to make the impact of nitrogen completely negligible. As shown in Fig. 7, for a natural boron concentration of 1.7 × 10 19 cm −3 , the SEU cross section is expected to grow by a factor of 3-10 at the considered critical charge with respect to the cross section that would result from the nitrogen in the thin barrier layers alone.
When considering lower critical charges, the gap reduces. At this natural boron concentration, the SEU cross section will, nevertheless, remain at least a factor of 2 higher than with nitrogen in the thin barrier layers alone. This natural boron concentration can be found quite often in tungsten plugs of commercial devices [37] and would typically result in SEU cross sections exceeding 10 −15 cm 2 /bit. Therefore, even relatively low concentrations of natural boron would, in general, make the impact of nitrogen in thin barrier layers negligible.
Among the materials that can be employed for semiconductor manufacturing, nitrogen can also be found in the form of silicon nitride (Si 3 N 4 ). This material is used in the device under consideration in the top-level passivation layer of the BEOL. Due to the distance from the SV, this layer did not contribute at all to the SEU cross section because of the low range of the released protons that would just stop inside the BEOL.
Nevertheless, it may be possible to find, in some devices, much higher amounts of Si 3 N 4 placed much closer to the SV, given that this material may be used, for instance, in place of SiO 2 . In the context of this criticality analysis, it would, therefore, be interesting to verify what would be the SEU cross section if SiO 2 in the BEOL would be fully replaced by Si 3 N 4 . Fig. 8 depicts the SEU cross section as a function of the critical charge for the cases in which SiO 2 and Si 3 N 4 are Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.  Given that the total thickness of layers containing nitrogen has increased from 420 to 2690 nm, the SEU cross section has almost proportionally increased by a factor of 5 for the critical charge of this SRAM, reaching 1.62 × 10 −15 cm 2 /bit. The increase was such that it matched also the SEU cross section from the 1.7 × 10 19 natural boron concentration case, becoming even slightly higher at lower critical charges. As it will be shown later, an SEU cross section this high would have important consequences for the hardness assurance in accelerator environments.

D. Analysis for Diverse Node Technologies
G4SEE simulations can also be performed to assess the relative importance of nitrogen and boron in contributing to the SEU cross section for technologies whose node varies from 22 to 250 nm. Since detailed models and construction analyses for devices in these nodes are not available down to the level of detail needed, this analysis is performed under some assumptions. For both the nitrogen content and placement, the BEOL used for the 65 nm SRAM under analysis will be retained. For the boron, the content will be varied, and its location will again be the tungsten plug, located in the same position as for the 65 nm SRAM. Therefore, no BPSG will be used in the simulations. This material typically characterized higher technology nodes than those considered here.
For the simulations, it is also necessary to define the SV size and critical charges characteristic of the technological nodes. For the critical charge, an extensive analysis for all the technology nodes of bulk silicon SRAMs is reported in the work from Kobayashi [39]. Multiple critical charges for the same node were evidenced in this work. Therefore, the critical charges that were adopted for the various nodes in this analysis, and reported in Table III, are interpolated from the values reported in that work. The critical charge for the 65-nm SRAM here reported (0.86 fC) is slightly lower with respect to other 65-nm examples reported by Kobayashi, but it aligns well with those of smaller nodes reported in that same work.
Concerning the SV size of the RPPs, a similar rescaling in terms of interpolation is applied, knowing that for a 250-nm SRAM, an SV size of 625 nm is suitable [40]. Then, intermediate nodes are rescaled based on this value and that for the 65-nm SRAM in this work, i.e., 360 nm. For the smaller nodes, a model for the 40 nm is available [31], and the 22 nm is rescaled accordingly. All data for the SV models used for the simulations are reported in Table III. Fig. 9 depicts the numerical SEU cross sections as a function of the technology node size and as a function of the contents of nitrogen and boron. The trend obtained for the case without any boron points out that the presence of nitrogen becomes significant only for nodes below 100 nm. As the technology node increases, the nitrogen has a more and more negligible, and even null, effect on the SEU cross section.
Boron is expected to be dominant for the nodes above 100 nm starting from concentrations larger than 2 × 10 18 cm −3 . In fact, for all the curves without nitrogen content, the SEU cross sections are seen to peak around 180 nm.
For lower technology nodes, the observations done for the 65 nm node still apply, although it is noted that the response due to the nitrogen peaks at around 65 nm and starts lowering for smaller nodes, probably because the shrinking of the SV is more impactful than the decrease in critical charge.
For cases in which both nitrogen and boron are considered, it is evident that the nitrogen provides a bump in the cross sections for nodes below 100 nm that becomes more and more negligible as the boron concentration increases.

IV. RADIATION ENVIRONMENTS OF CONCERN
Thermal neutrons can typically be found in radiation environments where fission or spallation reactions are involved. This is typically the case of nuclear power plants, high energy, and medical accelerators and of the terrestrial neutrons that can impact ground and avionics electronics.
In spite of the lower and lower reliance on natural boron in the manufacturing process of commercial devices, test standards for ground and atmospheric applications [41], [42] give high consideration to the thermal neutron hardness assurance. Thermal neutron sensitivity has also been an increasingly concerning issue for the hardness assurance of the electronics to be used in the high-energy accelerators [43], [44] involving both SRAM-and flash-based FPGAs and has brought to design specific guidelines for the design of systems that can be sensitive to them [45]. Similarly, for medical accelerators [46], thermal neutrons are considered the most significant threat to the reliability of the nearby electronics used to monitor the patient treatment.
Whereas most of these concerns pertain to the presence of boron close to the SV of the devices, the idea here is to assess whether the lower SEU cross section due to the presence of nitrogen may end up being a concern for these various environments.
The thermal neutron SER shall be put in relation to that arising from HENs (E > 10 MeV), in terrestrial environments, or high-energy hadron equivalents (HEH eq , i.e., protons, neutrons, and pions whose E > 20 MeV and intermediate energy neutrons whose E > 0.1 MeV [50], [51]), in the accelerator environment, that are typically the main concern when it comes to SEEs. The R-factor is the ratio between the thermal neutron flux and that of higher energy particles in a certain environment. Table IV reports the yearly thermal neutron and HEN (or HEH eq ) fluxes for two terrestrial cases at 0 and 12 km of altitude and for two CERN Large Hadron Collider cases with lightly shielded alcoves and heavily shielded alcoves. As the table shows, thermal neutron fluxes are expected to be more substantial when the amount of shielding (typically concrete, but it is also the case of the atmosphere) is increased. In the specific case of some of the considered CERN environment, in particular, the thermal neutron fluxes can be tens of times higher than those from HEH eq . SERs for this SRAM in the various environments can be obtained by multiplying the integral flux of thermal neutrons, and of high-energy particles, by the respective experimentally measured SEU cross section. Fig. 10 depicts the percentage contribution to the total SER of thermal neutrons and Fig. 10.
Percentage contribution to the SER of thermal neutrons and high-energy particles for the four environments under consideration. Data calculated for the CY62167GE30-45ZXI from the experimental results (labeled SiO 2 ) and for the case in which Si 3 N 4 is used in the BEOL in place of SiO 2 (labeled Si 3 N 4 ). high-energy particles in the four environments under consideration (SiO 2 labeled data).
As the bar plot shows, for this SRAM, the thermal neutrons are not expected to have a substantial impact on the atmospheric environments due to the fact that the HEN cross section is a factor of ≈200 higher than that of thermal neutron, and that the thermal neutron fluxes are lower than those from high-energy particles.
Concerning the accelerator, a more substantial effect is seen due to the relatively strong thermal neutron fluxes. For the lightly shielded alcoves, this contribution would anyway remain limited to <4%, whereas for the heavily shielded alcoves, this contribution would increase to 16%, a considerable contribution, though easy to margin. Fig. 10 also depicts how these relative contributions would change if Si 3 N 4 was used as a material in the BEOL in place of SiO 2 (Si 3 N 4 labeled data). Therefore, in this case, the data rely on the numerically simulated SEU cross section for the thermal neutrons, whereas the experimental one is still used for high-energy particles.
No substantial contribution is expected in atmospheric environments considering that the thermal neutron SEU-cross section has incremented of less than a factor of 5. For the accelerator environments, the contribution to lightly shielded alcoves has increased to 12%, whereas that in heavily shielded alcoves has reached 38%. The latter would already provide a contribution that may bring a significant error in the SER estimation if thermal neutron characterization is neglected.
The situation may be more critical in medical accelerators where the flux of thermal neutrons can be a few million times higher than both the thermal and HEN fluxes at ground level [46], resulting in R-factors >10 7 . In this case, even the tiny amount of nitrogen in the thin barrier layers will be sufficient to trigger SEUs in devices that would have, otherwise, not suffered any event from the HEN background.
This analysis allows concluding that thermal neutron sensitivity caused by the presence of nitrogen inside the BEOL of a memory device would likely not be an issue for terrestrial applications, but that the potential hazard may be substantial for the high-energy physics and medical accelerator applications or whenever the R-factor exceeds 10.

V. CONCLUSION
This article introduces the possibility that thermal neutron SEUs may not only be caused by the interaction with 10 B in the proximity of the SV, but also by the interaction with 14 N that may result in the release of LEPs. The devices that may be affected are, therefore, only those showing a marked sensitivity to LEPs. At the same time, there is no need for a very large amount of nitrides (TaN, TiN, or Si 3 N 4 ) to cause a substantial number of upsets in these kinds of devices.
The numerical simulations performed showed that it was possible to reproduce the experimentally measured thermal neutron SEU cross section without considering any boron in the BEOL of the SRAM and just based on realistic amounts (50-100 nm of TaN per 1 µm of copper) of nitrides. Nevertheless, the potential impact of both nitrogen and boron was also assessed in order to establish under which conditions boron SEUs would be dominant over nitrogen SEUs.
When Si 3 N 4 is used in place of other insulators, the SEU cross section may grow quite proportionally to the amount of Si 3 N 4 employed and can become more threatening.
The potential criticality of nitrogen induced SEUs was also assessed in the context of potential application environments of the commercial SRAM under consideration. The results show that an impact on terrestrial environments (i.e., ground and avionics) shall be disregarded. On the other hand, some substantial contributions were identified for heavily shielded areas in high-energy accelerators (and may also be applicable to some nuclear power plants and medical applications).

ACKNOWLEDGMENT
The authors would like to thank Robert C. Baumann for the insightful information about manufacturing practices and the useful discussion about the potential research lines. They would like to thank Helmut Puchner from Infineon, San Jose, CA, USA, for disclosing the information required to disable the embedded ECC in the CY62167GE30-45ZXI and on the boron usage in the manufacturing process of this SRAM. They would also like to thank the ILL, Grenoble, France, for the allocated beam time INDU-178 and INTER-499.