space-rha.comShort courseChapter 10 · Mitigation and system behavior

Chapter 10 · Lesson 4 of 4 · about 13 minutes

Retained margin and proving the implementation

TID and displacement damage move parameters and reduce gain until the circuit no longer meets its function, and no code or reset brings that margin back. The remedies are a tolerant part, an operating point with headroom, calibration, a redundant channel or local shielding, each evaluated against the element that is degrading.

What you need to know

Cumulative damage moves parameters, and no code or reset moves them back. The mitigation for dose and displacement damage is margin in the circuit. That can be a part whose parameters stay inside the budget at the end-of-life dose, an operating point with headroom against the drift, a calibration that removes the drift, a redundant channel that is switched in (and that has been accumulating the same dose while it waited), or shielding that lowers the dose at the part. Each of these is evaluated the same way: carry the measured parameter trajectories (chapter 7 lesson 3) into the circuit analysis at end of life, with temperature, load and aging, and read the result against the requirement on the function. The application limit is generally reached before the datasheet limit. A part can pass every vendor criterion and still fail the circuit.

Shielding is evaluated with the transported field rather than a rule of thumb. For electron-dominated orbits a few millimeters of aluminum at the part remove most of the dose. For proton-dominated orbits the return per gram diminishes quickly, and the secondary particles from a high-atomic-number shield can add to what the shield removes (chapter 1 lesson 5 and chapter 2 lesson 3). For galactic ions, shielding is nearly irrelevant to the single-event rate. Shielding against displacement damage follows the proton spectrum and the NIEL of the device material.

A technique becomes evidence only when it is verified on the implementation. Fault injection works through the modeled states and recovery paths systematically. It reaches every register the injector can address, and it misses the physical modes the model does not contain. A beam test reaches the physical modes, including the ones nobody modeled, but it samples them at whatever rate the beam produces rather than systematically, and a high flux produces overlapping faults and dead time that change how the mitigation under test behaves. The two methods complement each other and are combined with timing, interface and operational tests. Every coverage claim should state its denominator: which modes, states and observables it covered, and which gaps it left. The gaps direct the next analysis or test. A change to the netlist, bitstream, firmware image, reset settings or operating mode reopens the verification of whatever it touched.

Figure 10.4. The current transfer ratio of U4 against 50 MeV proton fluence (a teaching curve with a half-value at 1.0 × 10¹¹ per cm²), with the five-year sun-synchronous trapped-proton fluence above 10 MeV marked. The receiver needs a CTR of at least 0.5 of the initial 1.2.
Current transfer ratio of U4 against 50 MeV proton fluence (teaching curve) 109 1010 1011 1012 0 0.2 0.4 0.6 0.8 1 proton fluence, cm⁻² CTR relative to the unirradiated value 5 years SSO, protons above 10 MeV minimum CTR the receiver needs exp(−ln2 × Φ / Φ½)

Worked example 10.4. Two end-of-life calculations: U3's offset error and U4's link margin

U3 is fed from a 100 kΩ source impedance. Its input bias current, measured through the TID sequence of chapter 7, reaches 60 nA at 30 krad(Si) at high dose rate and 120 nA at low dose rate (the ELDRS factor of chapter 8 lesson 3). The offset error at the input is I_b × R: 6 mV at high dose rate and 12 mV at low dose rate, against a budget of 10 mV. The high-dose-rate figure meets the budget. The low-dose-rate figure, which is the one that applies in flight, does not. The options are a lower source impedance (which the sensor may not allow), a calibration that measures and removes the offset in flight, an auto-zero architecture, or a part with lower bias current. Each of these is a circuit decision; the radiation analysis informs it but does not make it.

U4's current transfer ratio follows the teaching curve CTR/CTR₀ = exp(−ln 2 × Φ/Φ½) with Φ½ = 1.0 × 10¹¹ protons per cm² at 50 MeV. Over five years in the sun-synchronous orbit the trapped-proton fluence above 10 MeV is 2.6 × 10¹⁰ per cm² (chapter 1 lesson 4; a 10 MeV-equivalent conversion through NIEL would adjust it and is left as the exercise). The CTR falls to 83 % of its initial value, 1.00 against the 0.5 the receiver needs, so the link keeps a margin of 2.00 at end of life. A solar-event fluence of the kind in chapter 9 consumes part of that margin. If the margin were used up, the drive current would have to rise in inverse proportion to the CTR, at the cost of the LED's own lifetime and thermal budget. Both checks go into the end-of-life analysis, together with the temperature dependence of the CTR, which the teaching curve omits.

Exercise

Work out the end-of-life performance of the circuit, and whether each mitigation is still available at end of life, using representative degradation histories and operating conditions.

State the verification coverage by failure mode, configuration and observable, and let the gaps that remain decide the next analysis or test.

Going deeper

History and interacting mechanisms matter

Dose rate, oxide trapping, interface states and annealing all change the parameter trajectories. Displacement defects change carrier lifetime and transport in ways that depend on the device structure, and degradation can also change the transient response or the SEE tolerance of a part. Crediting combined effects needs evidence for the technology, the test sequence and the flight history, because extrapolating each effect on its own can miss an interaction.

An unpowered spare channel can still have accumulated dose while it waited.

Coverage depends on the faults that were tried

Recovery from a chosen set of injected faults bounds only the faults, states and observations that set represented. A beam samples the physical response distribution, with its own limits on what the source can produce and what the instrumentation can see. Neither method on its own establishes exhaustive coverage of a complex device. The report should say which states were assumed equivalent, which modes were missed, and how the detector, the actuator and the protected function depend on one another.

A changed netlist, bitstream, firmware image, reset setting or operating mode has to be verified again wherever it touches the assurance argument.

The common mistake

A dose margin does not by itself cover dose-rate effects, the tail of the part population, displacement damage or a change in SEE response.

A finite campaign cannot show that every state and failure mode of a complex system was exercised.

Takeaway

  • Mitigation against cumulative damage protects the performance budget and the physical structures that deliver the function; no reset recovers either.
  • Verification on the implementation is what turns a mitigation from an intention on a block diagram into evidence about the required function.

Check yourself

Why does the circuit analysis use the low-dose-rate bias current rather than the high-dose-rate value that was measured on more parts?

Because the flight dose arrives at a low rate, and bipolar parts generally degrade more per rad at low rates (ELDRS). The high-dose-rate value is the optimistic one. The application argument has to use the low-dose-rate characterization or an enhancement factor.

A redundant U3 channel is kept unpowered as a spare. Does it arrive at the switch-over with fresh parameters?

Generally not. Dose accumulates in the unpowered part, and for some technologies the unpowered bias state traps as much charge or more. The spare is analyzed at the dose it has received by the time it is switched in.

What does a beam test of the implemented mitigation add that fault injection cannot?

The physical modes the model did not contain, such as transients in the voter, upsets in the configuration of the injector itself, or a SEFI that stops the scrubber. The beam samples those modes at its own rate rather than systematically, so the two methods are used together.

U3 meets every datasheet limit at 30 krad(Si) in the test report, and the parts engineer closes the dose question. What does the worked example show?

The application limit is generally reached before the datasheet limit. U3's input bias current of 120 nA at 30 krad(Si) at low dose rate, through the 100 kΩ source impedance, gives an offset error of 12 mV against a budget of 10 mV, so the circuit fails while the part passes every vendor criterion. The measured parameter trajectories have to be carried into the circuit analysis at end of life, with temperature, load and aging, and read against the requirement on the function. The options that follow (a lower source impedance, in-flight calibration, an auto-zero architecture, a part with lower bias current) are circuit decisions that the radiation analysis informs but does not make.

A project proposes extra shielding to bring the U3 offset error inside its budget and, at the same time, to reduce the U2 upset rate. Where does the rule of thumb about shielding hold, and where does it fail?

Shielding is evaluated with the transported field rather than a rule of thumb. For electron-dominated orbits a few millimeters of aluminum at the part remove most of the dose, so for the dose at U3 the proposal may help. For proton-dominated orbits the return per gram diminishes quickly, and the secondaries from a high-atomic-number shield can add to what the shield removes (chapter 1 lesson 5 and chapter 2 lesson 3). For galactic ions, shielding is nearly irrelevant to the single-event rate, so the heavy-ion share of U2's upsets would barely move.

U4 keeps a CTR margin of 2.00 at the end of five years. Why does the analysis not stop there?

Because the margin is computed against the trapped-proton fluence alone, 2.6 × 10¹⁰ per cm² above 10 MeV, and a solar-event fluence of the kind in chapter 9 consumes part of it. The teaching curve also omits the temperature dependence of the CTR, and the 10 MeV-equivalent conversion through NIEL would adjust the fluence. If the margin were used up, the drive current would have to rise in inverse proportion to the CTR, at the cost of the LED's own lifetime and thermal budget, so both checks go into the end-of-life analysis rather than the single figure of 1.00 against the 0.5 the receiver needs.

Further reading

  • J. R. Schwank, M. R. Shaneyfelt and P. E. Dodd, "Radiation hardness assurance testing of microelectronic devices and integrated circuits: radiation environments, physical mechanisms, and foundations for hardness assurance," IEEE Trans. Nucl. Sci., vol. 60, no. 3, pp. 2074-2100, 2013.
  • J. R. Srour, C. J. Marshall and P. W. Marshall, "Review of displacement damage effects in silicon devices," IEEE Trans. Nucl. Sci., vol. 50, no. 3, pp. 653-670, 2003.
  • R. L. Pease, R. D. Schrimpf and D. M. Fleetwood, "ELDRS in bipolar linear circuits: a review," IEEE Trans. Nucl. Sci., vol. 56, no. 4, pp. 1894-1908, 2009.
  • G. P. Summers, E. A. Burke, P. Shapiro, S. R. Messenger and R. J. Walters, "Damage correlations in semiconductors exposed to gamma, electron and proton radiations," IEEE Trans. Nucl. Sci., vol. 40, no. 6, pp. 1372-1379, 1993.
  • NASA, "Avionics Radiation Hardness Assurance (RHA) Guidelines," NASA/TM-20210018053, 2021. NTRS 20210018053.
  • Space-RHA, "NIEL tool," space-rha.com/niel-tool.