Single-event upset rate estimator
Estimate single-event upset rates from Weibull cross-section parameters or the figure of merit method, for heavy ions and protons, with scrubbing and error correction.
ReliabilityFree to useHow it worksSourcesWhere it helps
Fills the inputs below with typical figures. Replace any of them with your own.
Want your own spectrum? Start from this one and edit the rows.
- Upset rate, per device-day
- 0.004395
- Per bit per day
- 1.048e-9
- Per device per day
- 0.004395
- 4,194,304 bits
- Whole system per year
- 1.605
- 1 device
- FIT per device
- 183,100
- 45,780 FIT per Mbit
- Mean time between upsets
- 227.5 days
- Whole system
- Expected upsets in the mission
- 16.05
- 10 years
- Weibull cross section, share of saturation (solid line, circles)
- Step at L0.25 (figure of merit) (dashed line, squares)
Show the data as a table
| LET (MeV cm2/mg) | Weibull cross section, share of saturation | Step at L0.25 (figure of merit) |
|---|---|---|
| 0 | 0 | 0 |
| 1.83 | 0.0046 | 0 |
| 3.66 | 0.026 | 0 |
| 5.49 | 0.056 | 0 |
| 7.32 | 0.092 | 0 |
| 9.15 | 0.13 | 0 |
| 11 | 0.17 | 0 |
| 12.8 | 0.22 | 0 |
| 14.6 | 0.26 | 1 |
| 16.5 | 0.31 | 1 |
| 18.3 | 0.35 | 1 |
| 20.1 | 0.4 | 1 |
| 22 | 0.44 | 1 |
| 23.8 | 0.48 | 1 |
| 25.6 | 0.52 | 1 |
| 27.5 | 0.56 | 1 |
| 29.3 | 0.6 | 1 |
| 31.1 | 0.63 | 1 |
| 32.9 | 0.67 | 1 |
| 34.8 | 0.7 | 1 |
| 36.6 | 0.73 | 1 |
| 38.4 | 0.75 | 1 |
| 40.3 | 0.78 | 1 |
| 42.1 | 0.8 | 1 |
| 43.9 | 0.82 | 1 |
| 45.8 | 0.84 | 1 |
| 47.6 | 0.86 | 1 |
| 49.4 | 0.87 | 1 |
| 51.2 | 0.89 | 1 |
| 53.1 | 0.9 | 1 |
| 54.9 | 0.91 | 1 |
| 56.7 | 0.92 | 1 |
| 58.6 | 0.93 | 1 |
| 60.4 | 0.94 | 1 |
| 62.2 | 0.95 | 1 |
| 64.1 | 0.95 | 1 |
| 65.9 | 0.96 | 1 |
| 67.7 | 0.96 | 1 |
| 69.5 | 0.97 | 1 |
| 71.4 | 0.97 | 1 |
| 73.2 | 0.98 | 1 |
| 75 | 0.98 | 1 |
| 76.9 | 0.98 | 1 |
| 78.7 | 0.98 | 1 |
| 80.5 | 0.99 | 1 |
| 82.4 | 0.99 | 1 |
| 84.2 | 0.99 | 1 |
| 86 | 0.99 | 1 |
| 87.9 | 0.99 | 1 |
| 89.7 | 0.99 | 1 |
| 91.5 | 0.99 | 1 |
| 93.3 | 1 | 1 |
| 95.2 | 1 | 1 |
| 97 | 1 | 1 |
| 98.8 | 1 | 1 |
| 101 | 1 | 1 |
| 102 | 1 | 1 |
| 104 | 1 | 1 |
| 106 | 1 | 1 |
| 108 | 1 | 1 |
| 110 | 1 | 1 |
- Heavy ions0.004395100%
Show the data as a table
| Item | Upsets per device per day |
|---|---|
| Heavy ions | 0.004395 |
Sign in to save your inputs and come back to them later.
| Environment | Weibull integral | Figure of merit | Protons | Total for the method chosen |
|---|---|---|---|---|
| GEO, galactic cosmic rays, solar minimum | 0.004395 | 0.002118 | 0 | 0.004395 |
| GEO, worst week of a solar particle event | 0.2235 | 0.1271 | 0 | 0.2235 |
| LEO, 400 km, 51.6 degrees | 0.0009019 | 0.0004235 | 0 | 0.0009019 |
| LEO, polar or sun-synchronous | 0.00267 | 0.001271 | 0 | 0.00267 |
| Interplanetary space, solar minimum | 0.005494 | 0.002753 | 0 | 0.005494 |
| Threshold scaled by | L0 (MeV cm2/mg) | Rate, per device-day |
|---|---|---|
| 0.5 | 0.5 | 0.007321 |
| 0.75 | 0.75 | 0.00546 |
| 1 | 1 | 0.004395 |
| 1.5 | 1.5 | 0.003186 |
| 2 | 2 | 0.002501 |
| 4 | 4 | 0.001299 |
How it works
A memory or logic device in orbit, or at altitude on Earth, is upset by ions, protons and neutrons that flip stored bits. This tool estimates how often, from the device's measured cross section and the particle environment: by integrating a Weibull cross section against an LET spectrum, by Petersen's one-line figure of merit, or from proton and neutron cross sections and fluxes. It then shows what error correction with scrubbing leaves uncorrected.
- Describe the cross section. Heavy-ion test data are fitted with a Weibull curve: zero below the threshold LET, then rising to a saturation cross section. The width sets how fast it rises and the shape how it bends. Cross sections can be entered per bit, in square micrometres per bit, or per device.
sigma(L) = 0 for L <= L0; sigma(L) = sigma_sat x (1 − exp(−((L − L0) / W)^s)) for L > L0 - Integrate against the LET spectrum. The environment is an integral LET spectrum: the flux of ions a day per square centimetre with LET above each value. The upset rate is the cross section folded with the differential spectrum. Written by parts it becomes an integral over the Weibull density, which the tool evaluates with Simpson's rule. Between spectrum rows the flux is interpolated log-log; above the last row it is zero.
rate = integral of sigma(L) x (−dPhi/dL) dL = sigma_sat x integral of Phi(L0 + W x) x s x^(s − 1) x exp(−x^s) dx, per bit per day - Or use the figure of merit. Petersen's figure of merit replaces the whole calculation with one number. It treats the cross section as a step at the LET where the Weibull reaches a quarter of saturation, and the spectrum as a power law. The orbit enters through one coefficient. The tool reads L0.25 from the Weibull curve or takes your value.
L0.25 = L0 + W x (−ln 0.75)^(1/s); rate per bit-day = C x sigma_sat / L0.25^2 - Add protons and neutrons. Protons and neutrons upset devices through nuclear reactions, so they are handled by a cross section and a flux above the device's threshold energy rather than by LET. For terrestrial work the neutron flux is the sea level reference scaled by an altitude factor.
proton rate per bit-day = sigma_p x flux above threshold; neutron rate per bit-day = sigma_n x flux (n/cm2/h) x altitude factor x 24 - Scale to a device, a system and FIT. The rate per bit becomes a rate per device, per year, and in FIT (failures per billion device hours, with 1 Mbit taken as 1,048,576 bits). The mean time between upsets is the inverse of the system rate, and the mission total is that rate times the mission length.
per device-day = rate per bit-day x bits; FIT = per device-day / 24 x 10^9; MTBU (days) = 1 / (per device-day x devices) - Error correction and scrubbing. With a single-error-correcting, double-error-detecting code, one upset in a word is repaired the next time the word is read or scrubbed. A word is lost when a second upset reaches it before that. Single upsets reach a word as a Poisson process; the share of upsets that flip two or more bits of the same word is uncorrectable at once.
x = (data bits + check bits) x rate per bit-hour x (1 − multi-bit share) x scrub interval; P(2 or more) = 1 − exp(−x) x (1 + x); uncorrectable per device-hour = words x P / interval + multi-bit share x upsets per device-hour
Assumptions
- The orbit LET spectra and proton fluxes are editorial examples with the usual shape and order of magnitude. They are not output of CREME96, SPENVIS or OMERE; replace them with the mission's own environment analysis.
- The figure of merit orbit coefficient is the tool's own, read from the editorial spectrum (the integral flux at 10 MeV cm2/mg times 100). It is not Petersen's tabulated coefficient; enter his value for your orbit and part class if you have it.
- The cross section is the measured, effective one for the whole device, so the geometry of sensitive volumes and the angle of incidence are already inside it; no rectangular parallelepiped model is applied.
- Upsets are independent and arrive as a Poisson process; the multi-bit share is a fraction of events and the check bits are exposed like data bits.
- A scrub or a read repairs every correctable word, and a double hit is detected but not corrected. With no scrub interval entered, errors accumulate for the whole mission.
Limits and warnings
- Single-event upsets in memory cells. Single-event latch-up, burnout, gate rupture, transients and functional interrupts need their own cross sections and are not covered.
- The figure of merit is a screening method; it overstates the rate for soft parts with a low threshold and understates it for hard ones, and the Weibull parameters from a fit are not unique.
- Rates behind more or less than about 100 mil of aluminium, or at different solar conditions, differ; use the spectrum for your case.
- Protons act through a measured cross section above a threshold energy; a full proton energy spectrum (the Bendel method) is not integrated.
- Neutron altitude scaling is a single factor; real flux varies with latitude, solar cycle and building shielding.
Sources
The methods and default figures rest on these public sources. The wording here is our own; read the source for its full text.
- Suggested Single Event Upset Figure of Merit (opens in a new tab), IEEE Transactions on Nuclear Science, volume 30, issue 6 (E. L. Petersen, J. B. Langworthy, S. E. Diehl), December 1983. The original figure of merit: a step cross section, a power-law spectrum and one orbit constant.
- SEE Rate Calculations Using the Effective Flux Approach and a Generalized Figure of Merit Approximation, IEEE Transactions on Nuclear Science, volume 42, issue 6, pages 1995 to 2003 (E. L. Petersen), December 1995. The figure of merit restated with L0.25, the LET at 25% of the limiting cross section.
- The SEU Figure of Merit and Proton Upset Rate Calculations (opens in a new tab), IEEE Transactions on Nuclear Science, volume 45, issue 6, pages 2550 to 2562 (E. L. Petersen), December 1998. Orbit rate coefficients for heavy ions and a proton coefficient, different for hardened and unhardened parts. Cited for the method; its coefficient tables are not reproduced.
- 3.4. CREME96 and Related Error Rate Prediction Methods (opens in a new tab), NASA Marshall Space Flight Center (J. H. Adams Jr.), NASA Technical Reports Server 20120016823, 2012. Survey of rate methods: the RPP model, the figure of merit and its restatement with L at 25% of the limiting cross section.
- ECSS-E-HB-10-12A, Calculation of radiation and its effects and margin policy handbook (opens in a new tab), ECSS Secretariat, ESA-ESTEC, 17 December 2010. Clause 8.5: the RPP and integrated RPP methods, the Weibull cross section and proton and neutron rates from flux times cross section.
- Soft Error Rate in Space: A Unified Analytical Approach (opens in a new tab), arXiv 2501.06260, January 2025. The rate as a cross section folded with the LET spectrum, a comparison with Petersen's figure of merit (it overstates the rate at low thresholds and understates it at high ones) and why Weibull parameters from a fit are not unique.
- Reliability of Scrubbing Recovery Techniques for Memory Systems, IEEE Transactions on Reliability, volume 39, issue 1, pages 114 to 122 (A. M. Saleh, J. J. Serrano, J. H. Patel), 1990. The model of a single-error-correcting code with scrubbing: a word fails on a second upset between scrubs.
- Multiple Cell Upset Partitioning for Simulation of Soft Error Rates in Space Systems with Error Correcting Codes (opens in a new tab), arXiv 1704.07271 (G. I. Zebrev and others), 2017. Scrubbing efficiency of a simple SEC-DED code when some upsets flip several bits of one word.
- Heavy Ion Orbital Environment Single-Event Effects Estimations (application note SLVK046B) (opens in a new tab), Texas Instruments, May 2020, revised June 2025. A worked rate from an integral flux at the onset LET; the GEO worst-week point at LET 65 follows its example.
- JESD89A, Measurement and Reporting of Alpha Particle and Terrestrial Cosmic Ray-Induced Soft Errors in Semiconductor Devices, JEDEC Solid State Technology Association, October 2006. Cited by name. It takes New York City at sea level as the reference for soft error FIT rates; the reference neutron flux is commonly quoted as about 13 n/cm2/h above 10 MeV, so confirm it against the standard.
Where it helps
The lifecycle stages this tool serves: 3 stages in 1 lifecycle.
More reliability tools
- Failure rate (FIT) calculator
- Stress test acceleration factor calculator
- Qualification test planner
- Defect level (DPPM) estimator
- Error rate confidence calculator
- Lot sampling plan calculator
- Stack and multi-die yield calculator
- Radiation dose margin calculator
- Component derating checker
- Mission profile builder
- Warpage and thermal mismatch estimator
Method notev1Formula, ranges, sources and limits
Estimates single-event upset rates for a memory from Weibull cross-section parameters and an LET spectrum, from Petersen's figure of merit, or from proton and neutron cross sections, and the uncorrectable rate once error correction and scrubbing are applied.
Formula
sigma(L) = 0 for L <= L0; sigma(L) = sigma_sat x (1 − exp(−((L − L0) / W)^s)) for L > L0; rate = integral of sigma(L) x (−dPhi/dL) dL = sigma_sat x integral of Phi(L0 + W x) x s x^(s − 1) x exp(−x^s) dx, per bit per day; L0.25 = L0 + W x (−ln 0.75)^(1/s); rate per bit-day = C x sigma_sat / L0.25^2; proton rate per bit-day = sigma_p x flux above threshold; neutron rate per bit-day = sigma_n x flux (n/cm2/h) x altitude factor x 24; per device-day = rate per bit-day x bits; FIT = per device-day / 24 x 10^9; MTBU (days) = 1 / (per device-day x devices); x = (data bits + check bits) x rate per bit-hour x (1 − multi-bit share) x scrub interval; P(2 or more) = 1 − exp(−x) x (1 + x); uncorrectable per device-hour = words x P / interval + multi-bit share x upsets per device-hour
| Input | Unit | Range it holds for | Comes from |
|---|---|---|---|
| Heavy-ion method | choice | weibull, fom, none | Your figure |
| Cross section unit | choice | cm2bit, um2bit, cm2dev | Your figure |
| Saturation cross section | as set by the unit | from 0 | Your figure |
| Threshold LET | MeV cm2/mg | 0 to 100 | Your figure |
| Weibull width | MeV cm2/mg | from 0.001 | Your figure |
| Weibull shape | ratio | 0.1 to 20 | Your figure |
| LET spectrum | choice | geo-gcr, geo-worst, leo-iss, leo-polar, deep-space | Your figure |
| Figure of merit coefficient | choice | orbit, own | Your figure |
| Own coefficient | ions MeV2 cm2/mg2 per cm2 per day | from 0 | Your figure |
| L0.25 | choice | weibull, entered | Your figure |
| L0.25 entered | MeV cm2/mg | from 0 | Your figure |
| Include protons | choice | yes, no | Your figure |
| Proton cross section | as set by the unit | from 0 | Your figure |
| Proton flux above threshold | protons/cm2/day | from 0 | Your figure |
| Include terrestrial neutrons | choice | yes, no | Your figure |
| Neutron cross section | as set by the unit | from 0 | Your figure |
| Neutron flux above 10 MeV | n/cm2/h | from 0 | Your figure |
| Altitude factor | ratio | 0 to 1000 | Your figure |
| Memory size | as set by the unit | from 0 | Your figure |
| Memory size unit | choice | bit, kbit, mbit, gbit | Your figure |
| Devices | count | from 1 | Your figure |
| Error correction | choice | none, secded | Your figure |
| Data bits per word | bits | 8 to 512 | Your figure |
| Scrub interval | as set by the unit | from 0 | Your figure |
| Scrub interval unit | choice | s, min, h, day | Your figure |
| Multiple-bit share of upsets | % | 0 to 100 | Your figure |
| Mission length | years | from 0 | Your figure |
Outputs: Upsets per bit per day (1/bit/day); Upsets per device per day (1/device/day); Upset rate per device (FIT); Mean time between upsets, whole system (days); Uncorrectable errors per year, whole system (1/year).
Assumptions
- The orbit LET spectra and proton fluxes are editorial examples with the usual shape and order of magnitude. They are not output of CREME96, SPENVIS or OMERE; replace them with the mission's own environment analysis.
- The figure of merit orbit coefficient is the tool's own, read from the editorial spectrum (the integral flux at 10 MeV cm2/mg times 100). It is not Petersen's tabulated coefficient; enter his value for your orbit and part class if you have it.
- The cross section is the measured, effective one for the whole device, so the geometry of sensitive volumes and the angle of incidence are already inside it; no rectangular parallelepiped model is applied.
- Upsets are independent and arrive as a Poisson process; the multi-bit share is a fraction of events and the check bits are exposed like data bits.
- A scrub or a read repairs every correctable word, and a double hit is detected but not corrected. With no scrub interval entered, errors accumulate for the whole mission.
Uncertainty: The arithmetic is exact (the integral to better than 0.1%). The orbit spectra are editorial examples good to an order of magnitude, and Weibull parameters from a fit are not unique, so expect a factor of several against a full mission analysis.
Known limits
- Single-event upsets in memory cells. Single-event latch-up, burnout, gate rupture, transients and functional interrupts need their own cross sections and are not covered.
- The figure of merit is a screening method; it overstates the rate for soft parts with a low threshold and understates it for hard ones, and the Weibull parameters from a fit are not unique.
- Rates behind more or less than about 100 mil of aluminium, or at different solar conditions, differ; use the spectrum for your case.
- Protons act through a measured cross section above a threshold energy; a full proton energy spectrum (the Bendel method) is not integrated.
- Neutron altitude scaling is a single factor; real flux varies with latitude, solar cycle and building shielding.
Where the method comes from
The Hub's own implementation of the Weibull-LET integral, Petersen's figure of merit and the Poisson model of SEC-DED scrubbing. The orbit spectra, fluxes and figure of merit coefficients are editorial examples.
Reference example, checked by test: Hand-worked example in this method note: figure of merit method with L0 = 1, W = 30, s = 1.5 so L0.25 = 1 + 30 x 0.28768^(1/1.5) = 14.074; coefficient C = 10; saturation cross section 1e-8 cm2 per bit; rate = 10 x 1e-8 / 14.074^2 = 5.049e-10 per bit-day; 1 Mbit of 1,048,576 bits gives 5.294e-4 per device-day, 22,058 FIT and a mean time between upsets of 1,889 days. Upset rate per device 22,058; Mean time between upsets, whole system 1,889; Uncorrectable errors per year, whole system no result.
Versions
- v1 (2026-10-25): First version, built from the tool idea.
Terms on this page
- Figure of merit
- For a switch, the product of on-resistance and gate charge, or of on-resistance and output charge, which captures the conduction versus switching trade-off in one number.
- Micron
- One millionth of a metre, the unit behind pre-2000 node names such as 0.35 micron, and still the working unit for package features, bump pitch, wafer thickness and defect sizes.
- Neutron flux
- The rate of atmospheric neutrons, created by cosmic ray showers, passing through a device.
- FIT
- The standard failure rate unit, one failure per billion device hours of operation.
- Latch-up
- Triggering of the parasitic PNPN thyristor formed by the well and substrate of a CMOS structure, creating a low impedance path from supply to ground that persists until power is cycled and can dest...
- Shielding
- Routing grounded or supply-tied wires beside or above and below a sensitive net so that coupling to switching neighbours is intercepted, limiting crosstalk delay and noise on that net.
Tell us what you're after and leave your email: we'll connect you with the right person on the Hub.