space-rha.comShort courseChapter 9 · Rates and mission impact

Chapter 9 · Lesson 1 of 4 · about 29 minutes

Environment times response

A part is vulnerable only where the mission supplies particles that can trigger its measured response. The rate comes from the LET bins where the flux and the response overlap. A large saturation cross section alone does not set the rate, and a high test LET alone does not show that the part is immune in the mission.

What you need to know

A rate is the overlap of two curves: the differential flux of particles against LET after transport to the part, and the response of the part against LET. Where the flux is large, at low LET near the onset of the Weibull curve, the response is generally small. Where the response has saturated, the flux has fallen by orders of magnitude. The product peaks in between, and the rate is its integral. The figure shows this for U2 in a sun-synchronous orbit. The contribution per bin is largest just above the onset, near LET 2 to 5, and the bins above LET 30, where the test spent most of its beam time, contribute almost nothing. For the test plan this means the low-LET runs decide the rate. For the uncertainty budget it means the onset region is where the counts are needed.

The simplest integral, σ(L) times the differential flux summed over LET, treats the measured normal-incidence cross section as if it applied to particles from every direction. It is a useful first look and a check on the units, but the standard tools compute something else. The rectangular-parallelepiped method models each sensitive volume as a box of lateral size X and depth Z. A particle deposits charge in proportion to its LET and its path length through the box, so an ion below the normal-incidence threshold still upsets the cell if it crosses the box along a long enough chord. The path lengths of isotropic particles through a box follow a known distribution (Bradford gave it in closed form; the calculator samples it). The integral RPP method of Pickel and Petersen reads the Weibull curve as the distribution of thresholds over many such boxes and integrates each threshold over the chord distribution and the spectrum. CREME96 implements this method. The course calculator reproduces it with a Monte Carlo chord distribution so that you can change the geometry.

The depth of the box is the one parameter the data do not give. In a thin box, grazing paths are long compared with the depth and the rate rises above the normal-incidence shortcut. In a cube they are not. A common practice reports two numbers: a best estimate with the depth at a fifth of the lateral size (X/5), and a worst case with the depth at a hundredth (X/100). The ratio between them is the directional uncertainty of the rate. For a soft part whose onset lies below most of the spectrum the ratio is small (1.08 for U2). For a part whose onset falls where the spectrum drops steeply the ratio is large (3.7 for U1). Angular test data constrain the depth. Without them, report both numbers.

Check the units before anything else. The spectra used here are omnidirectional integral fluxes (particles per cm² per day above an LET, with the 4π solid angle integrated in), in silicon, behind 100 mil of aluminum, Z = 2 to 92, solar minimum. The cross section is per device in cm² at normal incidence. The product is events per device per day. A spectrum per steradian, a cross section in µm² per bit, a dose-rate reference in a different material, or a differential spectrum multiplied by an integral response will each produce a smooth and wrong curve. The response is already an effective area, so do not divide it by the die area again. Keep the SEU, SEFI and destructive responses as separate integrals, because each has its own consequence.

Figure 9.1. The heavy-ion flux per LET bin in a sun-synchronous orbit and its product with the U2 response: the rate comes from the bins just above the onset. Normal-incidence shortcut; the IRPP integral shifts weight further toward low LET.
Where the heavy-ion SEU rate of U2 comes from (sun-synchronous orbit) 1 10 100 10-6 10-5 10-4 0.001 0.01 0.1 1 LET in silicon, MeV·cm²/mg per LET bin, per device per day Weibull onset LET at 25 % of saturation ions in the bin, per cm² per day σ(L) × ions in the bin: events per day
Try it: an IRPP rate on the CREME96 LET spectra

Pick an orbit and a response (or type your own Weibull parameters), compare the best estimate and worst case, and add a proton cross section where the orbit has trapped protons. The figure shows which LET bins carry the rate.

Figure 9.1b. The fraction of isotropic paths longer than a given length through a square box of depth Z = X/5 and Z = X/100, in units of the lateral size X; the thin-slab law (Z/x)² holds until the path approaches the lateral size. The thin box has far more long paths, so its rate is higher for a part whose onset lies above most of the spectrum.
Fraction of isotropic paths longer than a given length through a square sensitive volume 0.01 0.1 1 10-4 0.001 0.01 0.1 1 path length in units of the lateral size X fraction of entering particles with a longer path Z = X/5 Z = X/100 infinite slab, (Z/x)²

Worked example 9.1. The heavy-ion upset rate of U2 in four orbits

Orbit (CREME96, 100 mil Al, solar minimum)Ions above LET 1, per cm² per dayabove LET 10Shortcut σ × flux, per dayIRPP X/5, per dayIRPP X/100, per day
LEO 400 km, 51.6°, quiet290.0290.130.370.42
Sun-synchronous 600 km, 97.8°590.2060.541.041.13
MEO 20,200 km, 55°1640.8031.973.353.56
Free space / GEO, solar minimum1710.9202.143.583.79

For U2 (σ_sat 0.396 cm² per device, onset 1.0), the IRPP best estimate in a sun-synchronous orbit is 1.04 upsets per device per day. That is 1.9 times the normal-incidence shortcut, because grazing paths push ions near the onset over threshold. The worst case is only 1.08 times the best estimate, because most of the spectrum is already above the onset. The geosynchronous value is 3.4 times the sun-synchronous value, since the geomagnetic cutoff removes the low-rigidity ions in low Earth orbit. For U1 (onset near 21) the same orbit gives a best estimate of 1.3 × 10⁻⁶ and a worst case of 4.6 × 10⁻⁶ per day, a ratio of 3.7. For the hard part, the sensitive-volume depth is the dominant uncertainty in the rate.

The four teaching parts in this chapter

Four constructed parts run through chapters 6 to 12 so that each chapter's work can be followed on the same hardware: U1, a shared regulator that powers sensing, logic and the recovery controller; U2, a 1 Gbit recorder memory under SEC-DED error correction and periodic scrubbing; U3, a precision analog chain with an accuracy allocation; and U4, an optical interface that must keep its signal margin. Their response curves are stated in the build (constructed from a declared true curve with Poisson sampling), so that every number in these chapters can be traced; they are teaching cases, not commercial devices.

PartFunctionThis chapter's step
U1 · shared powerA shared regulator powers sensing, logic and the recovery controller.Keep the recoverable interruptions and the irreversible damage as separate counts. A device event rate on its own is not a rate of functional loss.
U2 · recorder memoryA memory stores science data under ECC and periodic scrubbing.Fold each response with the spectrum at the part, then bring in the device count, the time at risk and the critical windows. The probability of an upset is not the probability of mission loss.
U3 · precision analogA precision analog chain has to hold its accuracy and timing allocation in the application.Keep the performance retained after dose separate from the SET rate. Whether a transient matters depends on the receiving threshold, the capture window and what the circuit is doing at the time.
U4 · optical interfaceAn optical interface has to keep enough signal margin through the mission.Turn the cumulative response into the link margin that remains. Report the transient rate and the functional-interruption rate as different claims.

Exercise

For a rate you have computed, write down the input spectrum, the normalization of the response, the transport case and the integration method, and keep the SEU, SEFI and destructive outputs as separate numbers.

Give every imported spectrum, response curve and result a note of its units and its normalization (per bit or per device, differential or integral, which reference material), so that a later reader can check the product.

Going deeper

The response already contains geometry assumptions

A response measured at several angles needs a flux integral that is resolved in angle too. Any LET-only shortcut builds in assumptions about the sensitive-volume geometry, the direction of the particle and how the charge is collected. Nonplanar structures and charge shared between neighboring nodes can break the simple slab picture. The scalar integral in this lesson shows where the weight falls; it does not reproduce every rate engine.

An integral spectrum multiplied by a differential response gives a wrong answer unless one of them is transformed first.

Tool input conventions are part of the model

One engine may expect square micrometers per bit while another accepts square centimeters per device. These are input conventions, not disagreements about the physics. A change of reference material changes the stopping power and therefore the LET coordinate. Where one particle upsets several bits, the count has to say whether it is events or bit errors, and how correlated words and the detection logic were handled.

The measured σ is already an effective area, so do not divide it by the geometric area of the die again.

The common mistake

A simplified integral cannot tell you anything about a mode that was not tested, and it gives a wrong number when the angular conventions of the spectrum and the response do not match.

A check of the units catches conversion errors. It cannot show that the response applies to the mission conditions or that every mode was covered.

Takeaway

  • The LET bins that contribute most of the rate show which measurements and which environment assumptions matter most to the trade.
  • A mistake in the units or in the event definition can dominate the result and still leave a smooth, plausible plot.

Check yourself

The test spent three of seven runs above LET 30. What did those runs contribute to the U2 rate?

Almost nothing to the rate itself; the flux above LET 30 is about ten thousand times lower than above LET 1. They did fix σ_sat, which scales the whole curve, and they showed that the response saturates. The rate is decided by the runs near the onset.

Why is the worst-case to best-estimate ratio 1.1 for U2 and 3.7 for U1?

The ratio measures how many ions below the normal-incidence onset can upset the cell along long paths. For U2 the onset is at LET 1, below most of the spectrum, so few ions are added. For U1 the onset is above LET 20, where the flux falls steeply, so the thin box with its many long chords adds a large share.

A spreadsheet multiplies the integral flux above L0 by σ_sat. What does it compute?

A rough upper estimate that counts every ion above the onset as if the response were already saturated there. It ignores the rise of the curve and the path-length distribution, so it serves as a check on magnitude rather than as a rate.

A rate calculation multiplies a differential LET spectrum by a per-bit cross section in µm² and produces a smooth curve of events per day. Why does a plausible-looking curve not mean the number is right, and what is checked first?

The spectra used here are omnidirectional integral fluxes (particles per cm² per day above an LET, with the 4π solid angle integrated in), in silicon, behind 100 mil of aluminum; the response is per device in cm² at normal incidence; the product is events per device per day. A spectrum per steradian, a cross section in µm² per bit, a dose-rate reference in a different material, or a differential spectrum multiplied by an integral response each produce a smooth and wrong curve. Every imported spectrum, response and result carries a note of its units and normalization, and since the measured σ is already an effective area it is not divided by the die area again. A check of the units catches conversion errors; it cannot show that the response applies to the mission conditions or that every mode was covered.

The IRPP best estimate for U2 in a sun-synchronous orbit is 1.04 upsets per day, 1.9 times the normal-incidence shortcut of 0.54. Where do the extra events come from, and which parameter controls how many there are?

The shortcut treats the normal-incidence cross section as if it applied to particles from every direction. In the box model a particle deposits charge in proportion to its LET and its path length, so an ion below the normal-incidence threshold still upsets the cell if it crosses the box along a long enough chord, and for U2 those grazing paths push ions near the onset over threshold, where the flux is largest. The depth of the box is the parameter, and it is the one the data do not give; a thin box has far more long paths than a cube, which is why a best estimate at X/5 and a worst case at X/100 are both reported unless angular test data constrain the depth.

A reviewer notes that the geosynchronous heavy-ion rate of U2 is 3.4 times the sun-synchronous rate and concludes that a part qualified for GEO is covered for any low Earth orbit. What is missing from that argument?

The 3.4 compares the galactic heavy-ion term alone, in which the geomagnetic cutoff removes the low-rigidity ions in low Earth orbit, for one shielding (100 mil of aluminum) and solar minimum. It says nothing about trapped protons, which lesson 9.2 shows supply about 139 of the 140 upsets per day that U2 sees in the sun-synchronous orbit, a term the geosynchronous rate does not contain. For a soft part in low Earth orbit the proton term is generally the rate, so the orbit with the lower heavy-ion rate can have the higher total.

Further reading

  • J. C. Pickel, "Single-event effects rate prediction," IEEE Trans. Nucl. Sci., vol. 43, no. 2, pp. 483-495, 1996.
  • E. L. Petersen, J. C. Pickel, J. H. Adams and E. C. Smith, "Rate prediction for single event effects: a critique," IEEE Trans. Nucl. Sci., vol. 39, no. 6, pp. 1577-1599, 1992.
  • J. N. Bradford, "A distribution function for ion track lengths in rectangular volumes," J. Appl. Phys., vol. 50, no. 6, pp. 3799-3801, 1979.
  • A. J. Tylka et al., "CREME96: a revision of the Cosmic Ray Effects on Micro-Electronics code," IEEE Trans. Nucl. Sci., vol. 44, no. 6, pp. 2150-2160, 1997.
  • R. A. Reed et al., "Physical processes and applications of the Monte Carlo radiative energy deposition (MRED) code," IEEE Trans. Nucl. Sci., vol. 60, no. 3, pp. 1441-1461, 2013.
  • D. L. Hansen et al., "Correlation of prediction to on-orbit SEU performance for a commercial 0.25-µm CMOS SRAM," IEEE Trans. Nucl. Sci., vol. 54, no. 6, pp. 2525-2533, 2007. doi:10.1109/TNS.2007.908787.
  • Space-RHA, "SEE rate tool," space-rha.com/see-tool.html (the CREME96 LET spectra used here).