SEE Rate Assessment Tool

Preliminary analysis aid. Check the result against your own analysis and, where it matters, a test on your part. Full disclaimer.

1. What the tool calculates

On orbit single event rates for a parts list: heavy ion rates by integrating each part's cross section over the LET spectrum of the chosen environment, proton rates by integrating a proton cross section over the proton energy spectrum, and a system total with Monte Carlo confidence levels. Parts with no test data get a bounding rate from technology priors (the Ladbury approach); parts with a null result get a Poisson upper bound. Output is events per device per day and per system per day, the mean time between events at the 90 percent level, per part contributions sorted by size, and a PDF report. Everything runs in the browser.

2. Inputs

Environment

Built in heavy ion LET spectra and proton spectra, all from one set of CREME96 runs made on 11 and 12 September 2026 with the same settings: solar minimum (the CREME96 solar quiet mode, 1977), AP8MIN trapped protons, quiet magnetosphere unless the label says stormy, Z = 1 to 92 transported through 100 mil of aluminum, LET spectra in silicon for Z = 2 to 92. The heavy ion set: GCR at solar minimum at GEO (the worst case for heavy ions and the usual design case), ISS 400 km 51.6 degrees with the quiet and with the stormy magnetosphere, LEO 400 km 28.5 degrees, the 550 km 53 degree constellation orbit, sun synchronous 600 km 97.8 degrees, polar 1200 km 87.9 degrees, MEO 20,200 km 55 degrees, GTO 200 x 35,786 km 27 degrees, the lunar and Mars surfaces (GCR with planetary shadowing), and the worst week, worst day and peak 5 minute solar particle event models at GEO. The proton set is the Z = 1 block of the same transported spectra, trapped plus galactic plus (for the event cases) solar protons, as the integral flux above energy; a legacy AP-8 table for 1470 km 53 degrees is kept for continuity. Two cautions. The low inclination 400 km case has no heavy ion flux above LET 20 MeV cm2/mg at all: the quiet time geomagnetic cutoff at 28.5 degrees excludes the low energy heavy ions that carry the high LET, so a part with an onset above 20 gets a zero rate there; use the ISS stormy case as the conservative bound for a disturbed magnetosphere. And the MEO table is a solar minimum spectrum (an earlier label called it solar maximum; the table did not change). Or import your own from CREME96 (native .LET integral spectrum output with % header lines) or as a two column text file (LET in MeV cm2/mg, differential flux) from OMERE or anything else. Proton spectra import as a CREME96 FLUX or TRANS output (.flx or .tfx, read natively: the proton block of the differential spectrum is integrated above each energy and converted from per m2 per s per sr to per cm2 per day over 4 pi steradian, the way the built in tables were made), or as two columns, energy in MeV and the integral flux above that energy, in the unit picked next to the import button (per cm2 per s, per cm2 per day, or per m2 per s per sr). A second column that rises with energy is refused, since it is a differential spectrum. The shielding depth is whatever the spectrum was generated behind; the tool does not transport.

Correction, September 2026: earlier versions used an imported proton file as it came. This page described the second column as a differential flux while the rate used it as the integral flux, and a file in per cm2 per s gave proton rates 86,400 times too low. The built in tables were not affected. Recompute any proton rate made with an imported file.

Parts

One row per part: name, quantity, technology, and one of four data categories.

Category	You supply	Rate method
Full Weibull	sigma_sat, L_th, W, s	Spectral integration
Partial	L_th and/or sigma at one LET	Missing parameters sampled from empirical priors in the Monte Carlo
Null result	Test LET and fluence, no events	Poisson upper bound on the cross section, bounding rate versus test LET
No data	Technology only	Ladbury bounding with technology susceptibility priors

A proton cross section (Bendel or Weibull in energy) can be entered per part; otherwise it is estimated from the heavy ion parameters.

Geometry and run settings

The RPP depth to width ratio Z/X (default 0.01), mission duration, and the number of Monte Carlo iterations (default 10,000).

3. Calculation

Heavy ion rate

The rate is the integral of the cross section over the differential LET flux:

It is evaluated on a log spaced LET grid from L_th to the top of the spectrum as a summation by parts:

where Φ_eff is the effective flux at L_i described below. Written this way the rate is a Stieltjes integral of the effective flux with respect to the measured cross section, R = ∫ Φ_eff dσ, and that is the point of the formulation: everything about the environment and the geometry sits in Φ_eff, which is tabulated once per environment, and everything about the part sits in σ(L), which is what the test measured. Integrating in sigma rather than in L also puts the grid resolution where it is needed, near threshold, where the flux is steepest.

This is Larry Edmonds' formulation of the heavy ion rate calculation, developed at the Jet Propulsion Laboratory and set out in his recommendations on the use of CREME96 for heavy ion rate calculations [9]. The conventional CREME96 path converts the cross section curve into a critical charge and an RPP with a guessed depth and integrates the chord length distribution; Edmonds showed the same physics can be carried entirely in an effective flux that depends only on the environment and the aspect ratio of the sensitive volume, so that the measured curve is used directly and the unknown depth appears as one explicit parameter whose worst case can be taken. The rate engine on this site descends from an implementation of that method and carries the effective flux tables computed his way. His later work extends the same picture to devices sensitive to particle energy and range [10], [11] and to non planar sensitive volumes through the alpha law [12], which the alpha law tool implements.

Cross section model

Parameter	Units	Meaning and typical range
σsat	cm2	Limiting cross section at high LET. About 1e-8 cm2 for SEL in small parts up to 1e-3 cm2 for SEU in large memories.
Lth	MeV cm2/mg	Onset LET. Below 1 for sensitive commercial parts, above 40 for hardened ones.
W	MeV cm2/mg	Width of the rise from onset to saturation.
s	none	Shape. Near 1 is an exponential rise; larger is closer to a step.

Effective flux and the chord length correction

The isotropic flux crosses the sensitive volume along a distribution of path lengths, and a long oblique chord lets a lower LET ion deposit the critical charge. Following Edmonds, the tool models the volume as a rectangular parallelepiped of depth to width ratio Z/X, integrates over solid angle for both path length and projected area, and folds the result with the environment spectrum into an effective flux Φ_eff(L; Z/X). The path length integral uses the full chord length distribution of the box, not its mean: for each direction the area over which the chord is at least a given length is the projected area of the box shrunk along the beam, a quadratic in the chord, so the fold with the spectrum reduces to three moments of the differential flux (the integral flux and its first two inverse moments) and an angular quadrature. This is the same integral CREME96's integral RPP performs and the closed form of the I-HICUP rate on the alpha law page, and the two agree to a fraction of a percent. Six of the eleven built in environments (GTO, LEO constellation, MEO, sun synchronous, lunar surface, Mars surface) had been folded earlier with the mean chord of each direction in place of the distribution; that approximation runs 25 to 80 percent low at Z/X = 0.2 because the chords longer than the mean let the far more numerous lower LET ions upset, and those six tables were regenerated with the exact folding in September 2026. Of the five older tables, the three solar particle events and ISS already used the full distribution and are unchanged; the GCR solar minimum table, the default, was 5 to 25 percent low at Z/X = 0.2 against the exact folding of its own spectrum (the spectrum itself agrees with a 2024 CREME96 run, solar quiet, near earth interplanetary, 100 mils of aluminum, to 0.2 percent) and was regenerated from that run, 1002 bins from 0.001 to 110 MeV cm2/mg. The tables are precomputed for 11 values of Z/X and interpolated for the value you enter. Because Φ_eff at each LET is largest for some aspect ratio, taking the maximum across all ratios at each LET gives a rate that bounds every geometry; that geometry independent upper bound is Edmonds' as well and is reported alongside the best estimate.

Z/X	Meaning
0.01	Thin sheet; close to the cosine law worst case. The default when the geometry is unknown.
0.2	Moderate depth; a common best estimate for planar CMOS.
1.0	Cube; the isotropic limit.

The ratio moves the rate by a factor of two to five or more for the same Weibull. It is one of the largest single uncertainties in the calculation, and it is rarely known for a commercial part. Z/X = 0.01 is the conservative default; do not argue for a larger ratio without a die cross section or a vendor statement to point at.

Proton rate

σ_p is a function of proton energy because the mechanism is a nuclear reaction, not direct ionization. Parts with L_th above 15 MeV cm2/mg are treated as proton immune: 15 is roughly the highest LET a silicon recoil reaches. Without proton data the tool estimates σ_p from the heavy ion Weibull with a Bendel type convolution over reaction products, with the proton threshold energy tied to the heavy ion onset and a linear rise to saturation. Total rate is heavy ion plus proton, reported separately as well as summed. In LEO and during solar events the proton term can dominate for low threshold parts.

System total

Monte Carlo

10,000 iterations by default. Each iteration samples the uncertain quantities, runs the integration for every part, and sums. Reported: mean, and the 90, 95 and 99 percent levels of the rate distribution, with histograms and a survival probability plot over the mission. For full Weibull parts the sampled quantity is a beta distributed scale factor whose spread widens as L_th falls, reflecting how poorly the steep part of the spectrum is usually pinned. For partial and no data parts the sampling covers the missing parameters (below).

Bounding methods for thin data

No data. Each iteration first draws whether the part is susceptible at all, with a technology prior, then a threshold from the historical distribution, then a saturated cross section conditioned on that threshold:

W and s come from a library of empirically observed pairs. Default susceptibility priors, editable in the tool:

Technology	P(SEL susceptible)	Technology	P(SEL susceptible)
Bulk CMOS	70%	SiGe	25%
CMOS SOI	10%	FPGA	60%
BiCMOS	50%	Flash	55%
Bipolar	30%	DRAM	55%
GaAs	5%

Null result. The Poisson upper bound on the cross section from zero events at the test fluence, combined with an empirical power law in test LET:

These follow Ladbury's single point bounding approach [5]. They are bounds. If you have one non zero cross section point, the right description of the result is also a bound, not a characterization, and the report should say so.

4. Assumptions

• The sensitive volume is a single rectangular parallelepiped with a single critical charge. Multi volume devices with stepped cross sections are not represented by one Weibull.
• The heavy ion cross section curve was measured, or is being interpreted, as a function of effective LET for a thin planar volume. For FinFET and other non planar parts this fails, and the integral RPP model is known to be insufficient for them. The alpha law tool is one possible approach, not yet validated; its rate can be brought here as a number.
• Events are independent Poisson arrivals; the system total is a sum.
• The environment spectrum already includes the shielding you intend to claim.
• Weibull parameters correspond to the flight bias, temperature and operating mode, which for latchup means maximum voltage and temperature.
• The uncertainty models (beta scale factor, lognormal conditional cross section, the W and s library, the susceptibility priors) are calibrated to historical heavy ion data and may not represent a new process.

5. Implementation details

• CREME96 integral spectra are differentiated numerically with log log interpolation between points, which is appropriate because the integral spectrum is close to a power law over most of the LET range; the integral RPP folding (the exact chord length distribution, 90 by 90 directions per octant, 2000 point moment tables) is then applied across the 11 ratio tables and 120 LET bins, both for the built in environments and for an uploaded CREME96 file.
• The LET grid is log spaced from L_th to the top of the spectrum at 20 bins per decade; the integration variable is sigma, so grid density follows the cross section rise. The effective flux is taken at the log midpoint of each bin (since September 2026; taken at the lower edge of the bin, as the original code did, the sum ran 10 to 15 percent above the converged integral). With the exact tables the tool now reproduces the closed form I-HICUP rate for the same box to within one percent at the tabulated Z/X values and within a few percent between them.
• Precomputed effective flux tables cover Z/X from 0.01 to 1.0; the tool interpolates between them.
• Per part contributions are sorted by their share of the system rate at the mean, so the dominant driver is at the top of the table.
• The PDF report is generated client side with jsPDF and captures the chart canvases as drawn, so run the Monte Carlo before exporting or the plots come out empty. It contains: risk class from MTBE₉₀ (Low, Medium, High, Very High); environments, Z/X, duration and iteration count; mean and 90, 95 and 99 percent rates for heavy ion, proton and total; the parts table with parameters; per part contributions with bar charts; the distribution and survival plots; the susceptibility priors used; and method notes.

6. Limitations

No confidence interval produced here includes every source of model error. The important limitations are:

• Overall accuracy. Even with good cross section data and a well specified environment, a predicted rate is typically within a factor of two to ten of what flies. The confidence levels quantify the parameter uncertainty the model knows about, not the systematic errors it does not. For most screening work the 90 percent level is the number to carry; the 99 percent level is set by the tails of the assumed distributions rather than by anything measured.
• Threshold region. The rate integral is most sensitive where data are usually thinnest, just above onset. Sparse data there is the commonest cause of a wrong rate.
• Geometry. The RPP ratio is a factor of several and is usually a guess.
• Environment. GCR varies with the solar cycle and solar particle events are stochastic; the built in spectra are statistical descriptions, not forecasts. Assess both the quiet time rate (for total event count) and the event rate (for peak instantaneous risk and for sizing mitigation such as power cycling or safe mode).
• Estimated proton cross sections. The Bendel type estimate from heavy ion data can be off by an order of magnitude, particularly where the proton mechanism is not purely nuclear reaction driven. It is a screening number; if the proton term matters to the answer, get proton data.
• Bounding priors. The susceptibility priors and the conditional cross section distribution are drawn from historical databases weighted toward older technologies. For a new process or a specialized one, treat them as a starting point and adjust with what you know.

7. References

Back to the SEE Rate Tool