Chapter 10 · Lesson 3 of 4 · about 13 minutes
Latchup protection and analog transients
A current limiter can interrupt a latchup before damage, but only if detection, switching and rail discharge all happen before the damaging energy has been delivered. Burnout and gate rupture do their damage during the initiating event itself, so protection claims are made per mechanism rather than for high-current events in general.
What you need to know
Latchup protection is a race between the parasitic thyristor and the circuit that starves it. Once triggered, the thyristor holds as long as the supply can deliver the holding current, and the damage depends on the energy delivered into the small region that carries that current. Metal fuses, junctions overheat, and parts that recovered on the bench have shown degraded leakage and later failure. A current limiter that is credited as protection has to be measured: its trip threshold against the latchup current, its total latency from the current step to the open switch, the time for the rail to decay below the holding voltage, and the restart. The energy reaching the fault comes from two sources, and the local one is the one usually forgotten. The output capacitor of the regulator discharges into the latched path however fast the upstream switch opens. A limiter with a 10 µs latency can therefore be irrelevant to a part with 47 µF on its rail. The capacitor empties into the fault over a few hundred microseconds, and the switch, open or not, is upstream of it. Protection claims are made by mechanism. A limiter addresses latchup; it says nothing about burnout or gate rupture, which complete within the ion's own time scale and are addressed by derating (chapter 7 lesson 4).
A single-event transient is a pulse at a pin, and what it does is decided downstream. The same 1 V, 2 µs pulse is harmless on a slow housekeeping line and fatal on a comparator that latches a shutdown, or on a reference that feeds an ADC in the middle of a conversion. The receiving function defines the threshold, the amplitude, duration and timing at which a pulse crosses a functional limit, and the mitigation is chosen against that threshold. A low-pass filter attenuates short pulses in proportion to their width over its time constant, at the price of slowing the wanted signal by the same time constant. Time qualification in logic or software ignores a state that does not persist. Range checks and voting of redundant sensors reject an outlier. A safe-state design makes the shutdown require a persistent condition. A filter can also stretch a disturbance it does not remove, so the test measures the signal at the receiving function, under the real load and bandwidth, with the range of pulse shapes the beam produced rather than one representative waveform.
Regulators need particular attention because their transients reach everything on the rail. A dropout of a few microseconds on a shared 3.3 V rail can reset several devices at once, and the regulator's own transient response, its output capacitance and the brown-out thresholds of its loads decide what happens. The measured amplitude and duration distributions from the beam, mapped through the tolerances of the loads, give the rate of consequential transients. Adding up the raw transient cross sections of unrelated parts gives a number that means nothing.
Worked example 10.3. The energy behind a U1 latchup, and the U3 transient at the comparator
U1's output rail carries 47 µF at 3.3 V and feeds a CMOS part whose latchup current is 0.8 A. The current limiter upstream trips at 0.4 A with a 10 µs latency.
| Source of energy into the latched path | Value | Note |
|---|---|---|
| Upstream supply until the limiter opens | 26 µJ | 0.8 A × 3.3 V × 10 µs |
| Local output capacitor | 256 µJ | ½ C V² with 47 µF at 3.3 V; discharges in about 194 µs at the latchup current |
The capacitor delivers 10 times the energy the supply delivers during the latency, and it does so whether or not the limiter trips. The limiter alone is therefore not the protection argument for U1's loads. The argument has to be one of three things: a demonstration that the capacitor's discharge is survivable (a test with the capacitor in place, followed by leakage and functional inspection), a layout that places the bulk capacitance upstream of the switch, so that the switch isolates the latched load from the stored charge and only the small decoupling the load needs for its own operation remains downstream (sized so that its ½ C V² stays well below the energy the part has been shown to survive), or a smaller capacitor with a faster regulator. For U3, the beam produced transients of up to 1 V and 2 µs at the output that feeds a comparator with a 0.25 V threshold. A first-order filter with τ = 20 µs passes a peak of 1 V × (1 − e^(−2/20)) = 0.10 V, below the threshold with a margin of 2.6. The wanted signal, a step that previously settled in microseconds, now reaches 90 % after 46 µs, and the control loop has to tolerate that delay. Longer transients in the beam record, the tail of the duration distribution, pass the filter with less attenuation and are checked one by one against the threshold.
Exercise
Identify the destructive mode and the circuit the protection has to cover, and record the operating conditions and the fault energy under which the protection was tested, because the claim holds only inside them.
For the transients, trace each sensitive interface to its receiver, define the functional threshold there, and verify the proposed mitigation with the range of pulse shapes the beam produced.
Going deeper
A safe operating region needs the right evidence
High-field destructive effects depend on the ion species, its range, the bias and the device structure, so two ions at the same LET do not generally carry the same risk of damage. Rate models for SEB and SEGR are particularly uncertain. Avoiding the exposure, conservative operating limits or a different part is often easier to defend than a failure-rate prediction that looks precise.
A current limiter protects against latchup; it does not in general protect against burnout or gate rupture.
Propagation depends on state and timing
Electrical attenuation, logical masking and the timing of the capture window each change whether a transient has a consequence. A pulse on a regulator can reset many devices at once, and a transient in an ADC or a clock can corrupt state that several functions depend on. Nonlinear feedback and protection loops can amplify or prolong a disturbance. A coupled circuit simulation of all this is only as good as its validation against the measured behavior of the source and of the receiving circuit.
The same reasoning rules out adding the raw transient cross sections of unrelated parts and calling the sum a system interruption rate.
The common mistake
A clean supply current trace by itself does not show that the internal structures escaped damage, or that every destructive mode was covered.
A waveform measured with a different load or a different monitoring bandwidth may not bound what happens in flight.
Takeaway
- Latchup protection is credited on measured timing and on the energy that reaches the fault; a current-limit setting by itself shows neither.
- A transient mitigation has to protect the receiving function under its real operating state, load and timing.
Check yourself
A limiter trips in 10 µs. Why is that not sufficient evidence that U1's loads survive a latchup?
Because the 47 µF output capacitor delivers 256 µJ into the latched path on its own, far more than the supply delivers in 10 µs. Survival has to be demonstrated with the capacitor present, or the bulk capacitance has to be moved to the supply side of the switch, leaving downstream only what the load needs for decoupling.
A 20 µs filter is proposed for every analog output. What does the transient record have to show before it is credited?
The amplitude and duration distribution of the beam transients. A pulse longer than the time constant passes nearly unattenuated, so the tail of the duration distribution, mapped through the filter, has to stay below each receiver's threshold. The slowed wanted signal also has to meet its timing.
Why are transient cross sections of unrelated parts not added into a system interruption rate?
Because a transient matters only when it crosses the threshold of its own receiver. The rate of consequential events is computed per path, through the tolerance of the receiving function, and only those rates are summed for a function.
The U1 limiter trips at 0.4 A, half the 0.8 A latchup current of the load, and the review takes that as proof that the limiter will act. What else has to be measured before the limiter is credited?
The trip threshold is one of four measured items. The total latency from the current step to the open switch, the time for the rail to decay below the holding voltage, and the restart all have to be measured, because the thyristor holds as long as the supply can deliver the holding current and the damage depends on the energy delivered meanwhile. The claim is also made by mechanism: a limiter addresses latchup and says nothing about burnout or gate rupture, which complete within the ion's own time scale and are addressed by derating (chapter 7 lesson 4).
U1's beam record shows output dropouts of a few microseconds, and a colleague dismisses them because the regulator output is back within microseconds. What decides whether they matter?
The loads. A dropout of a few microseconds on a shared 3.3 V rail can reset several devices at once, and the regulator's own transient response, its output capacitance and the brown-out thresholds of its loads decide what happens. The measured amplitude and duration distributions from the beam are mapped through the tolerances of the loads to give the rate of consequential transients, and the signal is measured at the receiving function under the real load and bandwidth, because a waveform taken with a different load may not bound what happens in flight.
After a latchup test the supply current of U1's load returned to its pre-event value, and the report concludes that the part is undamaged. Why is that conclusion not supported?
A clean supply current trace by itself does not show that the internal structures escaped damage. Metal fuses and junctions overheat in the small region that carried the holding current, and parts that recovered on the bench have shown degraded leakage and later failure. The demonstration has to be made with the 47 µF output capacitor in place, since it delivers 256 µJ into the fault whether or not the limiter trips, and followed by leakage and functional inspection.
Further reading
- A. H. Johnston, "The influence of VLSI technology evolution on radiation-induced latchup in space systems," IEEE Trans. Nucl. Sci., vol. 43, no. 2, pp. 505-521, 1996.
- H. N. Becker, T. F. Miyahira and A. H. Johnston, "Latent damage in CMOS devices from single-event latchup," IEEE Trans. Nucl. Sci., vol. 49, no. 6, pp. 3009-3015, 2002.
- A. H. Johnston, G. M. Swift, T. F. Miyahira and L. D. Edmonds, "A model for single-event transients in comparators," IEEE Trans. Nucl. Sci., vol. 47, no. 6, pp. 2624-2633, 2000.
- F. W. Sexton, "Destructive single-event effects in semiconductor devices and ICs," IEEE Trans. Nucl. Sci., vol. 50, no. 3, pp. 603-621, 2003.
- NASA, "Avionics Radiation Hardness Assurance (RHA) Guidelines," NASA/TM-20210018053, 2021. NTRS 20210018053.