prediction of proton cross sections for seu in srams and...

Microelectronics Reliability xxx (2015) xxx–xxx

MR-11668; No of Pages 5

Contents lists available at ScienceDirect

Microelectronics Reliability

j ourna l homepage: www.e lsev ie r .com/ locate /mr

Prediction of proton cross sections for SEU in SRAMs and SDRAMs using the METISengineer tool

C. Weulersse a,⁎, F. Miller b, T. Carrière c, R. Mangeret da Airbus Group Innovations, Toulouse, Franceb Airbus Group Innovations, Suresnes, Francec Airbus Defence and Space, Les Mureaux, Franced Airbus Defence and Space, Toulouse, France

⁎ Corresponding author.E-mail address: [email protected] (C. Weu

http://dx.doi.org/10.1016/j.microrel.2015.06.1170026-2714/© 2015 Elsevier Ltd. All rights reserved.

Please cite this article as: C. Weulersse, et alMicroelectronics Reliability (2015), http://dx

a b s t r a c t
a r t i c l e i n f o
Article history:Received 19 May 2015Accepted 26 June 2015Available online xxxx

Keywords:Space radiation environmentSoft errorSingle Event UpsetsProtonHeavy ion

METIS, SIMPA and PROFIT are engineer tools based on heavy ion cross section for the assessment of Single EventUpsets induced by protons. Whereas SIMPA and PROFIT were based on analytical models; METIS has the partic-ularity to rely onMonte-Carlo simulations of nuclear reactions and simple assumptions for upset triggering. Suchtools are very useful for end-users because no information about the technology is needed to perform the sensi-tivity prediction, not even the feature size. The work presents the prediction results achieved on sub-100 nmtechnology SRAMs and SDRAMs with METIS and compares them with the ones obtained using SIMPA andPROFIT,widely applied for space radiation environment.METIS givesmuchmore accurate results than the formeranalytical models.

© 2015 Elsevier Ltd. All rights reserved.

1. Introduction

Single Event Upsets (SEUs) occur when a single ionizing radiationevent produces a burst of hole–electron pairs that is large enough tochange the state of a storage element, such as configuration memorycell, user memory, and register.

In space, SEUs are caused by charge collection at sensitive circuitnodes following a proton or a heavier ion strike. The heavy ions pro-duce the effect directly while high energy protons induce upsetsthrough ions emitted as a result of their nuclear interactions withthe device matter. Furthermore, low energy protons may causedirect ionization SEU in very sensitive devices. Less than 1 protonin 105 will encounter a nuclear reaction in a silicon device. Althoughthis seems like a very rare event, for low Earth orbits, the trappedproton fluxes are so high that they will contribute to more upsetsthan the heavy ion cosmic rays.

In order to reduce the cost and the number of radiation tests, it isinteresting to develop bridges from one particle to another.

Among the large number of analytical predictions methods forSEU induced by protons based on heavy ion data, SIMPA [1] andPROFIT [2] are the most widely used by the European space commu-nity, as they are included in the OMERE package [3]. These models

lersse).

., Prediction of proton cross se.doi.org/10.1016/j.microrel.2

were developed in the middle of 90s, both require to specify thesensitive thickness and both rely on simplifications which haveshown their limits for process technologies below 130 nm (seeTable 1).

Monte Carlo codes based upon more physical triggering criteriacompute the energy deposition by tracking all the secondary parti-cles generated from nuclear interactions. Despite being more timeconsuming, they can be more accurate and detailed than analyticalapproximations [4].

Unfortunately, for an end-user, and so in space engineer workingin radiations, technological data required for full physical simula-tions can be hardly obtained. An irradiation experiment oftenremains the only way to characterize the device sensitivity to radia-tion effects. The METIS (Monte-Carlo Engineering Tool for Ion-induced SEE) tool offers to extrapolate heavy ion data to proton orneutron cross section, which makes it therefore very close to theSIMPA and PROFIT approaches. Based on a nuclear database, and sim-ple assumptions, it constitutes a relevant trade-off between accuracyand access to technological data. For SEU investigations in SRAMs,METIS is perfectly suitable for the end-user needs because no informa-tion about the technology is needed, not even the feature size, and thecalculation is fast.

The tool has been extensively validated for SEU calculation on SRAMdevices, bulk or SOI up to 45 nm [5].We propose here to further validatethe tool for very deep submicron SRAMs (45 and 28 nm) and to extendthe validation to SEU in SDRAMs. Results are also compared with theones achieved by SIMPA and PROFIT.

ctions for SEU in SRAMs and SDRAMs using the METIS engineer tool,015.06.117

http://dx.doi.org/10.1016/j.microrel.2015.06.117mailto:[email protected] logohttp://dx.doi.org/10.1016/j.microrel.2015.06.117http://www.sciencedirect.com/science/journal/www.elsevier.com/locate/mrhttp://dx.doi.org/10.1016/j.microrel.2015.06.117

Table 1Features of the 3 prediction methods.

METIS [5] SIMPA [1] PROFIT [2]

Inputs Heavy ion cross sectionSensitive thicknessonly required forSEU in DRAM.

Sensitive thickness(default value 2 μm)

Average diffusion angle(default value 90°)

Collection IRPP Rectangular surfaceCriteria Critical energy Critical LETNuclear data Monte-Carlo

selectionExperimentalenergy depositionin diodes

Analytical expressionsfor LET (ESi) and ∑nucl

Time-efficiency ~7 s/energy b1 s b1 s

Table 2Weibull parameters for the SRAM and DRAM devices.

Weibullparameters

Saturation crosssection (cm2/bit)

LETth W S

(MeV·cm2/mg)

45 nm SRAM DNW 6.0E−8 1.0 105 1.0no DNW 1.8E−8 1.0 80 1.0

45 nm SRAM-based-FPGA BRAM 1.8E−8 0.4 30 1.0CRAM 1.5E−8 0.8 20 1.8

28 nm SRAM-based-FPGA BRAM 7.94E−9 0.8 5 0.7CRAM 1.43E−8 1.9 125.3 0.78

45 nm SOI L2 cache 2.5E−9 0.4 32 0.8DRAM Samsung 1.7E−9 1.5 22 2.0DRAM Manufacturer C 1.0E−9 2 35 2.5

Fig. 1. Heavy ion cross sections for the 45 nm SRAMs.

2 C. Weulersse et al. / Microelectronics Reliability xxx (2015) xxx–xxx

2. METIS overview

METIS relies on the same expression as the SIMPAmethodology for a“gateway” tool between heavy ion and proton. The SEU proton crosssection is given as a function of the heavy ion cross section by:

σP Ep� � ¼

ZσHI Edð Þ : ΡEp Ed;Vð Þ : dEd

with Ep the incident proton energy, σHI the heavy ion cross section, andPEp(Ed), the probability to obtain a deposited energy Ed from the second-ary ions in a given volume V.

Based on the concept of sensitive sub-volumes having rectangularparallelepiped (RPP) shape, an upset occurs when the energy depositedby the ions in the charge collection region exceeds the threshold energy.For the estimations performed in this paper, around 30 nested sensitivevolumeswere selected along aWeibull curve of the experimental heavyion cross section. For each sensitive sub-volume, the threshold energy isthe product of the heavy ion LET per the sensitive depth. In METIS, theprobability for energy deposition is calculated using SRIM tables [6]and a Monte-Carlo selection of nuclear interactions from a pre-calculated database, which differs from the analytical expression usedby the SIMPA method.

The nuclear database is constituted by listing the characteristic(type, energy and direction) of the secondary ions induced during elas-tic scattering and nuclear reactions. It was built using the MC-RED(Monte Carlo Recoil Energy Determination) code and was experimen-tally validated [7,8]. In this nuclear physics code, elastic reactions withsilicon nuclei are studiedwith the ECIS [9] calculation code using the op-tical parameters from Koning and Delaroche [10]. Non-elastic reactionsare simulated following, step by step, the nucleus decay. The pre-equilibrium stage is calculated using the 1D excitonmodel [11,12]. Equi-libriumprocesses are simulated using theHauser-Feshbach formulation[13]. The angular distributions are evaluated using the Kalbach system-atics [14]. In the following studies, only nuclear reactions with siliconinduced by protons are considered. Nuclear interactions with otherelements than silicon used in the manufacturing process can be easilycovered by including the proper database.

Table 1 presents a comparison between METIS and the formermethodologies SIMPA and PROFIT.

3. SEU in SRAMs

For SEU studies in SRAMs with METIS, all input parameters are ex-tracted from the measured heavy ion cross section. No technologicaldata is required, and we assume that the heavy ion cross section fullycharacterizes the device sensitivity.

The sensitive areas are directly related to the measured heavy ioncross section curve. For submicron feature sizes, charge collection ismainly driven by diffusion mechanism and the sensitive thickness istherefore strongly linked to the sensitive surface (and is no more a

Please cite this article as: C. Weulersse, et al., Prediction of proton cross seMicroelectronics Reliability (2015), http://dx.doi.org/10.1016/j.microrel.2

constant value as it is in PROFIT and SIMPA). The sensitive thicknessesare then sized as half of the areal dimensions to approximate a hemi-sphere shape and are then increasing along the heavy ion cross sectioncurve.

In order to achieve a good trade-off between computing time and ac-curacy, the nuclear interactions are generated in a silicon layer of 20 μmbelow the active surface and an upper layer of 5 μm, simulating the backend of line.

More details about the tool algorithms can be found in [5].In [5], the METIS tool was validated for SEU simulation in SRAMs

using a large number of commercial CMOS memories. For most of thedevices, especially for deep submicron technologies, this tool givesmore accurate predictions than SIMPA and PROFIT and the margin of10 recommended in the ECSS Standard [15] when using SIMPA andPROFIT as a prediction tool could thus be reduced to about 3 whenusing METIS. Since then, new data are becoming available and thiswork investigates SEU in advanced SRAMs. Weibull fits, tabulated inTable 2, were made to all of the heavy ion data.

3.1. 45 nm SRAM

In [16], heavy ion (see Fig. 1) and proton (see Fig. 2) accelerator dataare presented on 2 types of 45 nm SRAM from ST, with deep N-welllayer (DNW) and without deep N-well structure. METIS simulationsshow fairly well accordance with experiments for both devices (seeFigs. 2 and 3). SIMPA and PROFIT mainly underestimate the protonexperiments when using 0.2 to 2 μm sensitive thicknesses and 90° ofaverage diffusion angle for PROFIT.

3.2. 45 nm SRAM-based-FPGA

The susceptibility of a SRAM-based-FPGA based on a 45 nm technol-ogy was characterized under heavy ions and protons at UCL (see Fig. 4).


http://dx.doi.org/10.1016/j.microrel.2015.06.117

Fig. 3. Proton cross sections for the 45 nm SRAM without deep N-well layer.

Fig. 5. Proton cross sections for 45 nm BRAM cells.

Fig. 6. Proton cross sections for 45 nm CRAM cells.

Fig. 2. Proton cross sections for the 45 nm SRAMwith deep N-well layer.

3C. Weulersse et al. / Microelectronics Reliability xxx (2015) xxx–xxx

As expected, the CRAM cells were found slightly less sensitive thanBRAM cells at low LET values.

Predictions given by METIS, PROFIT and SIMPA are compared inFigs. 5 and 6 for BRAM and CRAM respectively. METIS results are invery good agreement with proton testing. As already highlighted,PROFIT and SIMPA underestimate the proton cross sections.

Fig. 4. Heavy ion cross sections for the 45 nm BRAM and CRAM cells.


3.3. 28 nm SRAM-based-FPGA

The SEE response of the Xilinx 28 nm Kintex-7 FPGA under heavyions is shown in Fig. 7 [17]. The Weibull curves illustrating the CRAMand the BRAM cross sections allow us, using the 3 predictions tools,to compute the proton sensitivity and to compare with proton data

Fig. 7. Heavy ion cross sections for the 28 nm BRAM and CRAM cells.



Fig. 10. Heavy ion cross section for the L2 cache.Fig. 8. Proton cross sections for 28 nm BRAM cells.

4 C. Weulersse et al. / Microelectronics Reliability xxx (2015) xxx–xxx

[18,19]. For BRAM cells (see Fig. 8), METIS gives an overestimation of afactor 2. For CRAM cells (see Fig. 9), METIS matches very correctly theexperiments with a cross section value at 85% of the 63 MeV protondata.

3.4. 45 nm SOI

The P2010 is a microprocessor from the QorIQ family developed byFreescale manufactured on a 45 nm SOI (Silicon On Insulator) process.Figs. 10 and 11 provide the heavy ion and proton data obtained on theL2 cache memory [20].

The increase of the cross section measured below 10 MeV is proba-bly due to the direct ionization of lower energy protons encounteredin the beam. Direct ionization is also taken into account inMETIS by cal-culating the sensitivity to protons having a LET of 0.54 MeV·cm2·mg−1

(highest LET value of a proton given by SRIM) but no experimental datatowards proton direct ionization was carried out.

METIS underestimates the proton cross section by a factor 3. Wethen adapt theMETIS simulation in order to be closer to the SOI technol-ogy by limiting the sensitive thickness to the one corresponding to theSOI layer thickness. That means, if the thickness corresponding to a sen-sitive volume is lower than 150 nm, it remains unchanged. If this thick-ness is greater, we limit it to 150 nm. As shown in Fig. 11, this has noimpact on cross sections predicted by METIS. As expected, the sensitiv-ities estimated by PROFIT and SIMPA fall well below the accelerator datawhen using a 0.15 μm sensitive thickness.

Fig. 9. Proton cross sections for 28 nm CRAM cells.


4. SEU in SDRAMs

The experimental data on SDRAM are often more ambiguous thanthe one related to SRAMs. The sensitivity depends on the pattern andon the mode of operation, the effective LET used during a test at tiltedangles is not adequate, the presence of error code correction is not al-ways mentioned. This makes it difficult to validate METIS using biblio-graphical data. Moreover, some heavy ion data collected on veryadvanced SDRAMs show that the upset cross section does not reflectanymore the sensitivity at memory cell level [21]. In addition to theheavy ion curve, a SEU calculation in SDRAM by METIS involves onekey parameter, the thickness of the sensitive volumes (as it would befor the PROFIT and SIMPA approaches). The diffusion model seems tobe not applicable for DRAM due to the high level of doping and thepresence of isolation well.

Since the METIS results strongly depend on the sensitive thicknessfor DRAM sensitivity and the limitations mentioned above, the use ofMETIS to predict SEU in this kind of devices is limited. Above all, thecomputation time is greatly increased in order to reach very low crosssections.

However, we conduct a study on two old generations of SDRAMs.Two 512 Mb SDRAMs (0.12 μm technology) were tested under heavyions and protons [22]. METIS yields good result using a sensitive thick-ness of 0.5 μm (see Figs. 12 and 13). In the published data, most of theDDR2 and all the DDR3 SDRAMs tested under protons exhibit very lowSEU saturation cross section, between 1E−20 and 1E−19 cm2/bit,which can probably not be imputed to discharge of the DRAM cellcapacitor. In this case, METIS is not applicable.

Fig. 11. Proton cross sections for the L2 cache.



Fig. 13. Proton cross sections for the 512 Mb SDRAMs.

Fig. 12. Heavy ion cross sections for the 512 Mb SDRAMs.

5C. Weulersse et al. / Microelectronics Reliability xxx (2015) xxx–xxx

5. Conclusion

In this paper, we have investigated the validity of METIS on 45 and28 nm SRAMs and shown very good correlation with experimentalvalues reported in the open literature on bulk CMOS technologies. ForSOI process devices, the results are less accurate but still rather goodand constitutes a significant improvement to the analytical methodsthat are SIMPA and PROFIT.

We have also performed a preliminary study to extend this tool forthe SEU analyses in SDRAMs. The great dependence of the proton calcu-lation on the sensitive thickness affects the capability of METIS when no


technological data on the device is known, but, based on the results ob-tained on the 2 studied components, METIS seems to be also applicablefor old technologies of SDRAMs for which the upset mechanism is at-tributed to charge collection in the memory cell capacitor. As proposedfor the SET predictions in linear devices [23], one interesting way to useMETIS tool studies could be to determine the sensitive thickness toconsider based on the knowledge of both proton and heavy ion crosssections.

METIS can also handle the simulation of neutron environment foravionic applications, by replacing the proton-induced nuclear reactionsby neutron nuclear data.

References

[1] B. Doucin, T. Carrière, C. Poivey, P. Garnier, J. Beaucour, Y. Patin, Model of SingleEvent Upsets Induced by Space Protons in Electronics Devices, RADECS, 1995.402–408 (1995).

[2] P. Calvel, C. Barillot, P. Lamothe, R. Ecoffet, S. Duzellier, D. Falguère, An empiricalmodel for predicting proton induced upset, IEEE Trans. Nucl. Sci. 43 (1996)2827–2832.

[3] http://trad.fr/OMERE,14.html.[4] R.A. Reed, et al., Anthology of the development of radiation transport tools as

applied to single event effects, IEEE Trans. Nucl. Sci. 60 (2013) 1876–1911.[5] C. Weulersse, F. Wrobel, F. Miller, T. Carrière, R. Gaillard, J.-R. Vaille, N. Buard, A

Monte-Carlo Engineer Tool for the Prediction of SEU Proton Cross Section FromHeavy Ion Data, RADECS, 2011. 376–383 (2011).

[6] OnlineAvailable www.srim.org.[7] F. Wrobel, J.M. Palau, M.C. Calvet, P. Iacconi, Contribution of SiO2 in neutron-induced

SEU in SRAMs, IEEE Trans. Nucl. Sci. 50 (2003) 2055–2059.[8] S. Rocheman, F. Wrobel, F. Saigne, J.-R. Vaille, C. Weulersse, N. Buard, F. Miller, T.

Carriere, Measurement and calculation of charge deposition in a silicon diode irradi-ated by 30 MeV protons, J. Appl. Phys. 104 (2008).

[9] J. Raynal, Notes on ECIS95, CEA Saclay, Rep. CEA-N-27721994.[10] A. Koning, J.-P. Delaroche, Local and global nucleon optical models from 1 keV to

200 MeV, Nuclear Physics A 713 (3–4) (2003) 231.[11] C. Kalbach, Surface effects in the exciton model of preequilibrium nuclear reactions,

Phys. Rev. 32 (1985) 1157.[12] C. Kalbach, The griffinmodel, complex particles and direct nuclear reactions, Z. Phys.

A 283 (1977) 401.[13] M. Uhl, Calculations of reaction cross sections on the basis of a statistical model, with

allowance for angular momentum and parity conservation, Acta. Phys. Aust. 31(1970) 245.

[14] C. Kalbach, Systematics of continuum angular distributions: extensions to higherenergies, Phys. Rev. 37 (1988) 2350.

[15] European Cooperation for Space Standardization ECSS-Q-ST-60-15C PR-Draft 1, Oct.2011.

[16] P. Roche, G. Gasiot, Presentation “SEE and TID Radiation Test Results of DigitalCircuits Designed and Manufactured in ST 40 nm/45 nm/65 nm/90 nm/130 nmCMOS Technologies”, 2011.

[17] D.S. Lee, M. Wirthlin, G. Swift, A.C. Le, Single-event Characterization of the 28 nmXilinx Kintex-7 Field-programmable Gate Array Under Heavy Ion Irradiation, IEEEREDW, 2014.

[18] A. Lesea, Single-event Effects in 28 Nanometer Configuration and Dual-Port RAMBlock Memories, 2011.

[19] M.J. Wirthlin, H. Takai, A. Harding, Soft error Rate Estimations of the Kintex-7 FPGAWithin the ATLAS Liquid Argon (LAr) Calorimeter, 2013.

[20] D. Lambert, X. Vega, D. Gauthier, B. Azaïs, Radiation Effect Characterization on theQorIQ P2010 Microprocessor, IEEE REDW, 2012.

[21] R. Koga, P. Yu, J. George, S. Bielat, Sensitivity of 2 Gb DDR2 SDRAMs to Protons andHeavy Ions, IEEE REDW, 2010.

[22] T. Langley, R. Koga, T. Morris, Single-event Effects Test Results of 512 MB SDRAMs,IEEE REDW, 2003.

[23] C. Weulersse, S. Morand, F. Miller, T. Carrière, R. Mangeret, Simulation of Proton In-duced SET in Linear Devices and Assessment of Sensitive Thicknesses, RADECS,2015. (accepted for publication).


http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0075http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0075http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0075http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0010http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0010http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0010http://trad.fr/OMERE,14.htmlhttp://refhub.elsevier.com/S0026-2714(15)30071-8/rf0015http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0015http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0085http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0085http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0085http://www.srim.orghttp://refhub.elsevier.com/S0026-2714(15)30071-8/rf0025http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0025http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0025http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0095http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0095http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0095http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0100http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0105http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0105http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0035http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0035http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0040http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0040http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0045http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0045http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0045http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0050http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0050http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0110http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0110http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0055http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0055http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0055http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0115http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0115http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0115http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0065http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0065http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0070http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0070http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0120http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0120http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0125http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0125http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0130http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0130http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0135http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0135http://refhub.elsevier.com/S0026-2714(15)30071-8/rf0135http://dx.doi.org/10.1016/j.microrel.2015.06.117

Prediction of proton cross sections for SEU in SRAMs and SDRAMs using the METIS engineer tool1. Introduction2. METIS overview3. SEU in SRAMs3.1. 45nm SRAM3.2. 45nm SRAM-based-FPGA3.3. 28nm SRAM-based-FPGA3.4. 45nm SOI

4. SEU in SDRAMs5. ConclusionReferences

prediction of proton cross sections for seu in srams and...

Documents