space-rha.comShort courseChapter 3 · RHA requirements

Chapter 3 · Lesson 4 of 6 · about 10 minutes

Writing the SEE criteria

The environment sets how often particles arrive. The mission duration and the number of susceptible parts set how many opportunities for an event accumulate. The application and the consequence the project can accept decide what the system has to tolerate.

LEO alone does not set the LET requirement.

Try it: the screening LET on the SEL LET kernel

The numerical kernel of the Space-RHA SEL LET tool, running in the page. Change the environment, the number of susceptible parts, the mission length and the risk target, then solve.

More opportunities

With the response of each device held fixed, more susceptible devices and a longer operating life mean more chances for a failure over the mission. Device count is a convenient stand-in for complexity, but the technology, the shared resources, the operating modes and the recovery scheme matter as well.

Two missions in the same LEO

Compare a small, short mission with a complex, long-lived one, both in the same 550 km, 53° LEO and therefore with the same spectrum at the part. The SEL test LET requirement tool in the reading list estimates a screening LET for each from a historical SEL population and a criterion that the user selects explicitly. Since the environment is the same, the difference between the two answers comes entirely from the mission.

Connect rates, windows and recovery

The expected number of events over an interval follows from the assumed rate per device, the device count and the duty cycle. The service lost follows from that count and the time each recovery takes. Working the numbers through for one function shows how a per-device rate becomes a consequence that the project can judge.

Arithmetic. When the contributions are constant and independent, the expected count is the rate per device multiplied by the device count, the active fraction and the interval. When the rate varies over the interval, the expected count is the integral of the rate history over that interval instead.

Recovery credit. The arithmetic above assumes that every modeled event interrupts the service and that every successful recovery takes the same fixed time. Real hardware generally departs from both assumptions. How much of the fault space the detection actually covers, what a reset does to the rest of the system, and which events share a cause all have to come from test and analysis.

Screening requirement from MEAL. A screening LET can be derived from an expected-count criterion (how many events the mission can tolerate) or from a strict zero-SEL probability criterion. The two are different requirements and should be kept apart. Both depend on the historical population used as the prior and on how well that population represents the parts in question.

3D scene

More parts, more opportunities

The susceptible population and the criteria that connect it to a screening LET.

  1. A small, short mission: five susceptible parts for a tenth of a year in a 550 km, 53° orbit. With the spectrum at the part fixed, the number of opportunities for a latchup is set by the count and the time, and the worked example puts the strict screening LET for this case at 7 MeV·cm²/mg.

  2. Fifty susceptible parts for fifteen years in the same orbit: the environment did not change, the number of opportunities did, and the strict screening LET rises to 56 MeV·cm²/mg. Device count is a convenient stand-in for complexity; the technology, the shared resources, the operating modes and the recovery scheme matter as well.

  3. The expected-count criterion (fewer than one expected latchup over the mission in 99 % of the prior's draws) gives roughly half the LET of the strict criterion (99 % probability of zero latchups at the 90 % confidence rate). Both depend on the historical population used as the prior. Which one the project adopts must be written into the requirement, together with the orbit, the count, the duration and the acceptable probability.

The original space-rha.com environment and mission renderer (NASA imagery and spacecraft models, credits in the view), with the course scenes drawn by the same renderer. Trails, clocks and counts are illustrative; the scene computes no dose or rate.

Worked example: the screening LET for a constellation satellite

The question. A 550 km, 53° constellation satellite carries commercial CMOS parts whose latchup thresholds are unknown. How high must the heavy-ion screening LET be, and how does it move with the number of susceptible parts and the mission length?

Inputs. The Space-RHA SEL LET tool kernel, with its historical prior on LET thresholds (EHSP 2026-09), the CREME96 environment behind 100 mil of aluminum, an RPP depth-to-side ratio of 0.2, 1,500 Monte Carlo trials, a required probability of no latchup of 99 % over the mission, and a 90 % confidence level on the system rate.

Susceptible partsMissionStrict criterion: screening LETExpected-count criterion of the paper
50.1 year74
515 years3117
500.1 year167
5015 years5629
203 years3114
20, in free space / GEO3 years5935

Steps.

  1. Hold the orbit fixed and read down the table. Five parts for a tenth of a year need screening only to LET 7 MeV·cm²/mg under the strict criterion; fifty parts for fifteen years need 56. The environment did not change; the number of opportunities did.
  2. Compare the two criteria. The paper's expected-count criterion (fewer than one expected latchup over the mission in 99 % of the prior's draws) gives roughly half the LET of the strict criterion (99 % probability of zero latchups at the 90 % confidence rate). Which one the project adopts is a policy choice that must be written into the requirement.
  3. Move to free space. For the same twenty parts over three years, the screening LET rises from 31 to 59, because the full galactic spectrum reaches the parts (chapter 1, lesson 2).

Result. "LEO" alone does not set the LET requirement. The orbit, the number of susceptible parts, the mission length, the acceptable probability and the criterion all enter, and the requirement should state each of them. The explorer above runs the same kernel for any combination; the SEL LET tool on the site is the full version.

Exercise

Write a rate-and-recovery requirement for one defined function, conditioned on the operating window in which an event matters.

A mission-average availability figure cannot verify every operating window, so state the window the requirement covers.

Going deeper

What a test screen establishes

An LET screen bounds the response of the part at the conditions tested. It does not give a probability for the whole mission, and a run with zero events still leaves a finite upper bound on the cross section.

Zero-event example. For a Poisson count with a known exposure and full detection, the one-sided upper bound on the cross section at confidence level C is -ln(1-C)/fluence. That bound applies to the event definition and the beam condition that were tested. It is one point, and it is not an LET response curve.

Coverage. The screen should specify the voltage, the temperature, the angle, the ion species and its range, the fluence and the number of samples. Proton effects, whether by direct ionization or through nuclear reactions, and any other field that applies need their own evidence. An effective-LET cosine rule is generally not a substitute for evidence on the mechanism itself.

Takeaway

Derive the SEE requirement from MEAL and the complexity of the system, not from the orbit alone.

Check yourself

A requirement says "no SEL to an LET of 75 MeV·cm²/mg". What has to be added for a test to verify it?

The ion range at the sensitive depth, the fluence per LET and the confidence it supports, the bias and temperature (maximum voltage and 125 °C are usual), the event definition (what current, for how long), and the die or package preparation. Without them "no SEL to 75" is a label, not a verifiable requirement.

Why are screening, rate characterization and system verification kept as separate requirements?

They answer different questions with different tests: whether a mechanism exists under bounding conditions, how often it occurs in the mission environment, and whether the function survives it with the mitigation in place. One test rarely answers all three, and a requirement that mixes them cannot be closed cleanly.

A reviewer argues that the 550 km, 53° orbit is benign, so the screening LET of 7 MeV·cm²/mg that covered a five-part, tenth-of-a-year mission is enough for a fifty-part, fifteen-year satellite in the same orbit. What is missing from that argument?

The spectrum at the part is the same for both missions, so the environment is not what separates them; the number of opportunities is. With fifty susceptible parts for fifteen years the worked example puts the strict screening LET at 56 MeV·cm²/mg, against 7 for the small, short mission, and the same twenty parts over three years move from 31 in this orbit to 59 in free space. The orbit, the device count, the mission length, the acceptable probability and the criterion all enter the requirement, and each should be stated in it.

The SEL LET tool gives two screening LETs for the same mission, 31 MeV·cm²/mg under the strict criterion and 14 under the expected-count criterion. Which one belongs in the requirement?

Whichever the project adopts as policy, written into the requirement together with the orbit, the device count, the duration and the acceptable probability. The expected-count criterion (fewer than one expected latchup over the mission in 99 % of the prior's draws) and the strict criterion (99 % probability of zero latchups at the 90 % confidence rate) are different requirements and should be kept apart; the first generally gives roughly half the LET of the second. Both depend on the historical population used as the prior and on how well that population represents the parts in question, which is a judgment the requirement should also record.

An availability figure for a function assumes that every event interrupts the service and that every recovery takes the same fixed time. What does real hardware generally do, and where does the missing information come from?

Real hardware generally departs from both assumptions: the detection may cover only part of the fault space, a reset can disturb the rest of the system, and some events share a cause, so neither the count of interruptions nor the time per recovery is a constant. How much of the fault space the detection covers, what a reset does to the rest of the system and which events share a cause have to come from test and analysis, not from the arithmetic. A mission-average availability figure also cannot verify every operating window, so the requirement should state the window in which an event matters.

Further reading