Chapter 11 · Lesson 2 of 3 · about 12 minutes
Residual risk and configuration identity
Each option open to the project reduces the susceptibility, contains the consequence or collects stronger evidence, and each has a cost. The analysis says what remains uncertain and what happens if an assumption is wrong. The project's responsible decision maker weighs that residual risk against the mission's needs, resources and alternatives.
What you need to know
The radiation analysis informs a decision, and the decision belongs to the mission. What the analyst delivers is a set of options with their consequences and evidence: what each option costs in money, schedule, mass and performance, what residual risk it leaves and how well that residual is known, and what more testing would tell the project before it chooses. Likelihood and consequence are reported separately because they are uncertain in different ways. The likelihood of a U1 event is a rate with a known factor between the best estimate and the worst case. The consequence depends on the architecture and on whether the recovery path works, and that is either a demonstrated fact or a gap. A residual risk statement gives the probability band, the consequence, the basis and the uncertainty in a form the authority can accept, reduce or reject. A small modeled rate does not rule out an event in the critical window, and a wide uncertainty by itself does not make a risk high.
Before more testing is paid for, ask whether the result could change the choice. A test is worth its cost when it could; it is not worth its cost when it would only refine an estimate that already supports the choice. A second heavy-ion run of U1 at angles would narrow the factor of 3.7 between the geometry cases. If the project can accept both cases, the run changes nothing, and the beam time is better spent on the U2 proton SEFI gap, where the result could change the choice.
The evidence applies to one particular design, and designs change. The assessed configuration therefore has to be identified in enough detail that the flight configuration can be compared with it. That means part numbers with manufacturer, lot and date code, and die revision where known; board revisions; firmware and bitstream identifiers; the environment and transport versions; the response data and the analysis scripts with their settings; and the protection settings (trip currents, scrub intervals, watchdog timeouts). A manifest ties each report to those identifiers, and each identifier has an owner who hears about supplier and design changes. A file named final, or a hash of its bytes, identifies a document. It says nothing about whether the hardware on the bench matches that document. Reproducibility also covers the chain of transformations (spectrum normalization, live-time correction, fit constraints, geometry case), because a software default that changed between the analysis and the review changes the number without changing the data.
Worked example 11.2. The U1 options and the residual risk statement
| Option | Cost | What it resolves | What it leaves |
|---|---|---|---|
| Keep U1, add the verified current limiter and an independent recovery supply | medium | consequence of a high-current event | whether repeated trips cause latent damage |
| Replace U1 with a part tested to LET 75 without high-current events | high, schedule | likelihood | lot-to-lot SEL variability of the replacement |
| Buy a second heavy-ion run of U1 at the flight voltage and temperature | low | bias dependence of the threshold | consequence unchanged |
The residual risk statement for option 1, as presented to the project, reads as follows. A high-current event on U1 in three years has a probability between 0.14 % (best estimate, X/5) and 0.50 % (worst case, X/100). The figure comes from a heavy-ion curve with an onset near LET 21 at 125 °C and maximum input, integrated over the CREME96 solar-minimum galactic spectrum behind 100 mil of aluminum. Solar heavy ions are not included and would add to the figure during an event week. With the limiter and the independent recovery supply, demonstrated on the bench with the output capacitor present, the consequence is an interruption of the recorder, recovered by the ladder of chapter 7, rather than a loss of the function. The demonstration did not cover repeated events, and latent damage after a trip remains an open item, with a post-trip inspection planned for the next bench session. The project records its acceptance of this residual, open item included, together with the scope of the acceptance: this lot, this voltage, this orbit, three years. The manifest entry for the decision lists the U1 lot and date code, the board revision with the limiter settings, the analysis script version and geometry case, the CREME96 spectrum identifiers, and the owner of each.
Exercise
Write an option comparison and a recommendation with its scope, stating the assumptions, the unresolved concerns and who decides.
Create an evidence manifest that links the flight configuration identifiers to the raw data, the model versions and the reviewed analysis products.
Going deeper
Risk acceptance has a scope and an authority
The decision record names the hardware and mission baseline, the conditions under which the risk is accepted, the rationale, the authority and the changes that would reopen it. A small modeled rate does not rule out an event in a critical window, and a wide uncertainty does not by itself mean a high failure probability. The technical risk analysis is one record; the project's acceptance of the risk is another.
A smooth probability curve does not grant approval, and neither does a familiar mission class.
Reproducibility includes the chain of transformations
A saved final rate is not enough if the spectrum normalization, the dead-time correction or the fit constraints cannot be reconstructed. A change in a software default can alter the result without any change to the device data, so the model versions, the unit conversions and the numerical settings are controlled along with the hardware. A hash identifies the bytes of a file. It says nothing about whether the measurement was valid or whether it applies to the flight case.
Keep a frozen copy of the evidence, and keep with it a record of why that copy represents the assessed system.
The common mistake
A matching part name or file hash does not show that the construction is the same, that the tested sample represents the flight lot, or that the assumptions still hold.
Takeaway
- A residual risk stated with its band, its consequence and its gaps lets the project authority make a choice that can be reviewed.
- Configuration control keeps the assurance argument valid and reproducible as the design changes.
Check yourself
The project asks whether the U1 risk is "high". What should the analyst answer?
A probability band (0.14 % to 0.50 % over three years), the consequence with and without the recovery path, the basis and the gaps. Whether that is "high" is the project's judgment against its own risk tolerance; the analysis supplies the numbers for that judgment.
A supplier changes the U2 die revision after the test. Which parts of the record identify the change and which claims reopen?
The die revision field in the manifest, owned by procurement, flags the change. The SEU, SEFI and proton claims reopen, because the single-event response generally follows the mask set; the U2 TID claim reopens only if the process changed. The mitigation design itself is unchanged, but its sizing (interleaving, scrub interval) is rechecked against the new cluster distribution.
Why is a hash of the analysis script not sufficient for reproducibility?
A hash identifies the bytes of the script and nothing else. The inputs and settings (the spectrum files, the live-time corrections, the fit constraints and the geometry case) have to be recorded with it, and the physical configuration they describe has to be the flight one.
A second heavy-ion run of U1 at angles is offered at low cost, and the project asks whether it is worth buying. How does the course decide?
By asking whether the result could change the choice. The angular run would narrow the factor of 3.7 between the geometry cases, the band that the residual risk statement reports as 0.14 % to 0.50 % over three years. If the project can accept both ends of that band, the run only refines an estimate that already supports the choice, and the beam time is better spent on the U2 proton SEFI gap, where the result could change the choice. A test is worth its cost when the answer could move the decision, not when it would make a number more precise.
One reviewer reads the worst-case 0.50 % over three years and calls a U1 event not credible for the mission. Another reads the factor of 3.7 between the cases and calls the risk high. What is wrong with each?
A small modeled rate does not rule out an event in the critical window, and a wide uncertainty by itself does not make a risk high. Likelihood and consequence are reported separately because they are uncertain in different ways: the likelihood is a rate with a known factor between the best estimate and the worst case, while the consequence depends on whether the recovery path works, which is a demonstrated fact or a gap. The residual risk statement gives the band, the consequence with the limiter and the independent recovery supply (an interruption rather than a loss of the function), the basis and the open item on latent damage, and the authority accepts, reduces or rejects that within a stated scope.
Why does the manifest record the limiter trip current and the scrub interval next to the part lot and date code?
Because the evidence applies to one particular configuration, and the protection settings are part of it. The bench demonstration of U1 was made with a specific limiter setting and the output capacitor present, and the uncorrectable-word claim was sized for a specific scrub interval (lesson 10.1); a firmware image that changes either changes the number without changing the part. Each identifier has an owner who hears about supplier and design changes, so that the flight configuration can be compared with the assessed one before the evidence is credited. A file named final identifies a document, not the hardware on the bench.
Further reading
- ECSS-Q-ST-60-15C, "Space product assurance: radiation hardness assurance, EEE components," 2012.
- NASA, "Avionics Radiation Hardness Assurance (RHA) Guidelines," NASA/TM-20210018053, 2021. NTRS 20210018053.
- K. A. LaBel et al., "Emerging radiation hardness assurance (RHA) issues: a NASA approach for space flight programs," IEEE Trans. Nucl. Sci., vol. 45, no. 6, pp. 2727-2736, 1998. doi:10.1109/23.736521.