space-rha.comShort courseChapter 5 · BOM scrub and parts review

Chapter 5 · Lesson 4 of 6 · about 12 minutes

Reading cumulative and single-event evidence

A dose test may report output voltage when your circuit depends on leakage, offset or startup. An optocoupler can still switch and yet have lost the transfer ratio the receiver needs. Compare the measured degradation with the limit the application can tolerate at end of life, with the margin policy applied.

Check the parameter the circuit actually needs.

A dose report may give output voltage against dose when what your circuit depends on is leakage, offset or whether the part starts up. An optocoupler can still pass a signal after exposure and yet have lost the transfer ratio the receiver needs. Compare the measured degradation with the limit the application can tolerate, at the end-of-life conditions and with the project's margin policy applied.

A proton exposure delivers both ionizing dose and displacement damage. When you read a proton test, keep the two contributions separate, because they scale to the flight environment differently and generally degrade different parameters.

Review the lot, the rate, the bias, the timing and the endpoint.

For a TID report, check whether the samples represent the lot you will buy, what bias was applied during irradiation, the dose rate, the dose steps, what annealing was done and under what conditions the parameters were measured, and, where the technology makes it relevant, whether low-dose-rate sensitivity (ELDRS) was assessed. For a displacement damage report, keep the material, the particle energies, the NIEL or equivalence assumptions used to relate them, the injection level during operation and the annealing. Historical data from an older lot can guide the plan, but they generally do not bound a new lot.

A measurement taken at the board level can miss the internal parameter that is drifting, or the surrounding circuit can compensate for the drift until it runs out of range.

Know which events the test could see.

An upset, a functional interrupt, a latchup and a burnout each need their own monitoring to be seen and their own evidence that the part recovers. Read the device state and the operating conditions during the run, and not only the particle and the fluence. A run with no events is useful evidence about the modes that run could detect. It says nothing about the modes it could not.

A destructive mode with a serious consequence needs to be looked at even in LEO.

Inspect the range, the operating corners and the recovery.

For a heavy-ion report, record the ion species, the energy, the LET and the residual range at the sensitive structure, the angle, the fluence, the flux and how the device was monitored. For a proton report, record the energy and note whether the mechanism could be direct ionization or nuclear secondaries. In both cases check the supply voltage, the temperature, the workload, what was done to recover the part and how events were classified. For a power device, the evidence has to come from the relevant drain and gate bias and from the same structure.

A proton screen on its own, or a single nominal LET value, cannot establish a general heavy-ion bound.

Flow chart

Dispositioning one part and one mechanism with evidence

The identity test depends on the effect: die revision and process for single events, lot, date code and dose-rate basis for cumulative effects. A bound argued from a more severe tested condition closes the question with conditions attached. A missing record becomes an action with an owner; it says nothing about susceptibility either way. Walk the three teaching cases through the chart.

  1. A manufacturer heavy-ion report exists for the same die revision and process, tested at maximum input and 125 °C to LET 68 with the range stated, and the counts, the fluence and the recovery steps are in it. The question closes with the report cited; a die revision or a bias change reopens it.

  2. A workshop paper on the same part number and the same die revision, confirmed with the manufacturer, but tested in a different mode and pattern with no multiplicity data, and no physical argument bounds the multiplicity from it. The result is an action with an owner: a targeted heavy-ion run in the flight mode with the multiplicity recorded.

  3. A prior-mission report on the same part number, from a different lot, and bipolar dose response is lot-specific. The question stays open until a lot-specific test of the flight parts at high and low dose rate closes it, or it is accepted with the numbers of lesson 8.3 in the record.

Dispositioning one part and one mechanism with evidence yes no yes no yes no yes no yes no One part, one radiation question Evidence exists forthis part(manufacturer, archive,database, priormission)? Same silicon? SEE: dierevision and process.TID and DD: lot anddate code, dose-ratebasis Conditions cover theflight use: bias,temperature, mode, LETrange and range atdepth, energies, doserate? Endpoints are the onesthe circuit needs, withcounts, fluence andrecovery recorded? A bound can be argued(a more severecondition tested, aneffect-specificphysical argument)? Closed: cite the report, record theapplicability argument and thereopening triggers (lot, revision,bias, mission change) Closed with conditions: the bound,its basis and its limits in therecord Action with an owner: targeted test(chapter 6), analysis, supplierdata, design mitigation, orreplacement Open until the action closes it, oraccepted by the project as adocumented residual risk with itsscope (lot, conditions, mission) A missing record is a finding, not a susceptibility and not an immunity

Exercise on the demonstration board

On the demonstration board, compare the TID record of U3, the bipolar amplifier, with the displacement damage record of U4, the optocoupler. Each is missing a different fact; name it for each. Then select Q1, the power MOSFET, compare the drain bias of the reported heavy-ion test with the drain and gate conditions of the actuator switch, and say what the zero-event result bounds and where that bound stops.

A question to carry through the course

When, if ever, can the dose applied to a part during a proton SEE test be used as a proxy for a TID test?

Things to weigh here:

  • What the proton SEE report actually records: the energy and the fluence, whether anyone converted them to a dose through the stopping power in silicon, and the bias and temperature the part was held at during the run.
  • Whether any parameter was measured at all, and whether it was measured before, during or only after the exposure.
  • How many parts were exposed and from which lot, set against the samples and lot identity a TID lot test would need.
  • The displacement damage that arrived with the dose, which matters for a bipolar or optical part and generally matters much less for a MOS part.

Going deeper

Separate population inference from mechanism scaling.

A small sample can miss a failure distribution that is mixed or has a bad tail. A tolerance bound needs a population model you can justify and samples that represent the lot, and a larger design margin does not supply either. On the displacement damage side, NIEL equivalence between particles depends on the material and on the endpoint. Combined TID and DD effects, annealing of the defects and the operating history can all limit how directly a reported fluence transfers to your case.

Appendix E and the TID and DD mechanism models are there to test these assumptions before you fit a distribution because it is convenient.

Do not confuse a statistical limit with missing coverage.

If the counts are Poisson and the monitoring caught every event at the condition that was run, zero events gives a one-sided upper limit on the cross section of −ln(1−C)/fluence at confidence C. That limit holds for the LET, angle, range, device state, temperature and sample that were tested. Moving it to any other condition takes an argument, and correlated events or a mode the monitoring could not see need their own treatment.

The common mistake

This chapter does not hand you a universal sample count, a dose-rate correction or a lot-to-lot transfer factor, because the right value generally depends on the part and the mechanism. And a rate is only as good as the response data behind it: the cross section has to apply across the local particle field, and not only at the one condition the test happened to run.

Takeaway

  • A passed TID or DD test applies to the lot it sampled, the exposure history it followed and the parameter it measured.
  • Zero observed events gives a limit under the conditions of that test, which is a different thing from immunity.

Check yourself

The displacement damage record of U4, the optocoupler, reports a CTR loss after exposure at a single proton energy on two demonstration samples. Why does the course ask you to keep the ionizing and displacement contributions apart, and what else is the record missing?

A proton exposure delivers both ionizing dose and displacement damage, and the two scale to the flight environment differently and generally degrade different parameters, so a CTR measured after a proton run mixes two contributions unless they are separated. The record also leaves the energy dependence unestablished from one energy, measures the CTR at one LED current and one post-exposure time, and has two samples against the population the flight lot needs; the material, the particle energies, the NIEL or equivalence assumptions, the injection level during operation and the annealing all have to be kept. The endpoint is the minimum transfer ratio the receiver needs at end of life, not whether the part still switches.

A prior-mission TID report for U3, the bipolar amplifier, covers the same part number but a different lot. Why does that not close the question, and in what two ways can it be closed?

For cumulative effects the identity test is the lot, the date code and the dose-rate basis, not the part number, and bipolar dose response is lot-specific; historical data from an older lot can guide the plan but generally do not bound a new lot. The question stays open until a lot-specific test of the flight parts at high and low dose rate closes it (lesson 6.2), or the project accepts it with the numbers of lesson 8.3 in the record; acceptance is a controlled decision by the responsible authority, and it adds no evidence.

The heavy-ion report for Q1, the power MOSFET, recorded no destructive events at a drain bias lower than the actuator switch uses. What does the zero-event result bound, and where does the bound stop?

If the counts are Poisson and the monitoring caught every event at the condition that was run, zero events gives a one-sided upper limit on the cross section of −ln(1−C)/fluence at confidence C, and that limit holds for the LET, angle, range, device state, temperature and sample that were tested. For a power device the evidence has to come from the relevant drain and gate bias and from the same structure, so the bound stops at the tested drain voltage and gate setting; the intended drain and gate conditions were not covered, and moving the limit to them takes an argument that the physics generally does not support in that direction, since the higher bias is the more severe condition. What closes the question is range- and bias-resolved destructive-SEE evidence, a drain-voltage step at a reference LET (lesson 7.4), compared with the real switching envelope.

A dose report for U1 gives the output voltage against dose and the part passes, while the circuit depends on its startup and its leakage. What does the report leave open?

A dose report may give output voltage when what the circuit depends on is leakage, offset or whether the part starts up; U1's record on the demonstration board measured the output after a gamma exposure and left startup and leakage uncharacterized. Compare the measured degradation with the limit the application can tolerate at end-of-life conditions with the margin policy applied, and check whether the samples represent the lot, the bias during irradiation, the dose rate, the dose steps, the annealing and the measurement conditions. A measurement at the board level can also miss an internal parameter that is drifting while the surrounding circuit compensates for it until it runs out of range.

U3, the bipolar amplifier, and U2, the CMOS memory, went through the same proton SEE run. When the run is weighed as dose evidence, why does the displacement damage that arrived with the dose count differently for the two?

A proton exposure delivers displacement damage along with the ionizing dose. For a bipolar or optical part that matters, because bulk defects reduce the gain through a path separate from the surface ionization (lesson 4.6), so a parameter measured after the run mixes two contributions that scale differently to flight; for a MOS part it generally matters much less. That is one of the things to weigh, alongside what the report records of the energy, fluence, bias and temperature, what was measured and when, and the samples and lot against what a TID lot test would need; the course carries the question forward rather than settling it here.

Further reading