Chapter 5 · Lesson 5 of 6 · about 12 minutes
Priorities, assemblies and hidden parts
Not every BOM line gets the same test plan. Ask what could stop an important function, how uncertain that path is and which piece of information could change the design decision. Sometimes a supplier answer settles more than another beam run, and sometimes replacing the part early avoids a long campaign.
Work the consequential unknown first.
There is rarely enough beam time or schedule to give every BOM line the same test plan, and most lines do not need it. For each part, ask what could stop an important function, how well that path is understood and what piece of information would change the design decision. Sometimes a question to the supplier settles more than another beam run would, and sometimes replacing the part early avoids a long campaign altogether.
A part you know little about is not the same as a part that is likely to fail. Low information and high failure probability are different problems, and they call for different actions.
Trade actions by the question each one resolves.
The candidate actions are of different kinds: get the report author to clarify a result, verify the part identity, analyze the circuit, test a representative lot, extend an existing test to a wider operating range, verify a mitigation, or replace the part. Compare them by what each one would resolve, and only then by beam access, lead time, cost and schedule, because the most expensive experiment is often not the one that settles the open question. Lower-priority concerns stay on the list with an owner and a date; closing them quietly is how they get forgotten.
A box label can hide many vulnerable parts.
A module can contain processors, memories, regulators, oscillators and interface devices, each with its own response. Ask what is inside and what the test actually exercised. A board that kept running through a demonstration has shown you one configuration and one workload. It has not shown you the response of every part inside it or every software state it can be in.
A working output can hide retries, resets, corrected errors and margin that is quietly being used up.
Review what the integrated test could observe.
Get the internal BOM, or a controlled description from the supplier, where one is available. Record the board revision, the firmware, the configuration, the workload, what was monitored and how the board was recovered. Keep the evidence for the integrated function separate from the evidence for the parts inside it. Error correction, filtering and power management can hide events from the monitor, and they can also change the radiation and thermal conditions the internal parts see.
When the supplier will not disclose the internal BOM, write that down as an uncertainty and work it with the supplier, the designer and the project.
Exercise on the demonstration board
Pick one part on the demonstration board and lay out three actions for it: a question to the manufacturer about identity, a test that extends the coverage, and a design mitigation. Compare what each one would settle and what it would cost. Then select M1, the compute module, and list the internal parts, operating modes and recovery paths that its fictional board-level demonstration did not exercise.
Going deeper
Separate what you do not know from the risk to the mission.
A list of unknowns is not a probability distribution, and it does not give you an expected mission loss. Priority comes from the consequence, the exposure, how much is known about the susceptibility and whether a new result could change what the engineers do. Where there is no physical model of the mechanism, another count from the beam generally does not reduce the uncertainty that is controlling the decision.
Treat the system boundary as part of the experiment.
What a board test can observe depends on how faults propagate inside the module, what detects and corrects them, how many retries the firmware allows and what the external instrumentation records. To get a component cross section out of a protected output you need a written account of what the monitor could and could not see. A board-level TID result likewise depends on the dose distribution inside the module, the bias on each internal part, the parameters that were watched and the variation between internal parts, so a claim that production boards will behave the same way needs an argument that the tested configuration represents them.
Keep the differences between internal parts and their shared dependencies in view. Multiplying nominally independent part probabilities together assumes an independence the module generally does not have.
The common mistake
An open item gets deferred only by an explicit project disposition, with the reasoning recorded. A board that passed does not make every part inside it immune, and part-level data do not by themselves verify the system.
Takeaway
- Spend the effort on the evidence that could change an engineering decision with a real consequence.
- A board-level result supports the configuration and the behavior that were exercised, and no more.
Check yourself
The review ranks the parts by how little is known about each and sends the least-documented ones to the beam first. What is wrong with that ranking?
A part you know little about is not the same as a part that is likely to fail; low information and high failure probability are different problems and call for different actions. Priority comes from the consequence, the exposure, how much is known about the susceptibility and whether a new result could change what the engineers do, and a list of unknowns is not a probability distribution and gives no expected mission loss. Sometimes a question to the supplier settles more than another beam run, and sometimes replacing the part early avoids a long campaign altogether.
Why does the course compare candidate actions first by the question each one would resolve and only afterward by beam access, lead time, cost and schedule?
Because the most expensive experiment is often not the one that settles the open question. The candidate actions are of different kinds (getting the report author to clarify a result, verifying the part identity, analyzing the circuit, testing a representative lot, extending an existing test to a wider operating range, verifying a mitigation, replacing the part), and where there is no physical model of the mechanism, another count from the beam generally does not reduce the uncertainty that is controlling the decision. Lower-priority concerns stay on the list with an owner and a date; closing them quietly is how they get forgotten.
M1, the compute module on the demonstration board, ran one workload through a board-level exposure without a visible failure, and the supplier will not disclose the internal BOM. What has the demonstration shown, and what do you write down?
It has shown one configuration and one workload, not the response of every part inside or every software state the module can be in. A working output can hide retries, resets, corrected errors and margin that is quietly being used up, and error correction, filtering and power management can hide events from the monitor while changing the radiation and thermal conditions the internal parts see. Record the board revision, the firmware, the configuration, the workload, what was monitored and how the board was recovered, keep the evidence for the integrated function separate from the evidence for the parts inside, and write the undisclosed BOM down as an uncertainty to be worked with the supplier, the designer and the project. To get a component cross section out of a protected output you need a written account of what the monitor could and could not see.
Why can the failure probabilities of the parts inside a module not simply be multiplied together to give the module's?
Multiplying nominally independent part probabilities assumes an independence the module generally does not have: shared rails, clocks, resets and fault detection couple the parts, and a board-level result depends on the dose distribution inside the module, the bias on each internal part, the parameters that were watched and the variation between internal parts. Keep the differences between the internal parts and their shared dependencies in view, and a claim that production boards will behave as the tested one did needs an argument that the tested configuration represents them.
For Q1, the power MOSFET on the demonstration board, lay out the three actions the exercise asks for and say what each one would settle.
A question to the manufacturer about identity settles whether the die and process behind the heavy-ion report are the ones that will fly. A test that extends the coverage, a drain-voltage step at the actuator's gate condition and a reference LET with range into the device (lesson 7.4), settles the bias question that the lower-voltage run left open. A design mitigation that keeps the operating point, with its switching transients, inside a derated safe operating area settles the question by avoidance without claiming a rate (lesson 5.2), provided the derating rests on data for this device. Compare the three by what each resolves, and only then by beam access, lead time, cost and schedule.
Further reading
- NASA Avionics Radiation Hardness Assurance Guidelines (2021), §3
- G. R. Allen, "Predicting On-Orbit Performance of Devices," RADECS 2025 Short Course, Slides 4–5 and notes: mission consequence and multidisciplinary RHA
- LaBel, Johnston, Barth, Reed and Barnes (1998), Emerging RHA Issues, §III