Chapter 7 · Lesson 3 of 4 · about 14 minutes
Running cumulative and event tests
Cumulative damage shows up as a parameter or a function that changes with dose, and a final pass or fail is the least informative view of it. Measure each part before irradiation, measure the parameters the application depends on after every step, and record the elapsed time, because a device can recover or keep changing while it waits for the next measurement.
What you need to know
A cumulative test is a sequence of exposures with measurements between them, and those measurements are the result. Each part is measured before irradiation to establish its own baseline, and again after each dose step within a stated window. The window matters because MOS parts begin to anneal as soon as the beam stops, so the hour between beam-off and measurement is part of the test condition. TM1019 constrains that window. For MOS parts it also prescribes what follows: a room-temperature anneal and then the 168-hour rebound test at 100 °C under bias, which exposes interface-trap growth after the trapped holes have annealed. Bipolar parts follow the dose-rate provisions instead. The dose steps are chosen so that the parameter trajectory can be seen. Generally that means coarse steps through the first fraction of the required dose, finer steps through the region where failures are expected, and one step beyond it. The same steps are used for every part so that the trajectories can be compared. The bias during irradiation and during anneal, the temperature, the dose rate (chapter 6 lesson 2) and the elapsed time at each measurement are recorded with every point.
The parameters measured are the ones the circuit depends on, not simply the ones the datasheet lists. For U3, a bipolar amplifier, the application error budget is set by the input bias and offset currents, the offset voltage, the open-loop gain and the supply current. For a CMOS part it is the leakage current and the threshold shift. For U4 it is the current transfer ratio at the flight drive current and temperature. A part that passes every functional check but drifts out of the error budget at 45 krad has failed at 45 krad for this application. The population analysis of chapter 8 lesson 3 needs the whole trajectory of each part, because a final pass or fail hides where the crossing happened. Displacement-damage runs record the irradiation and storage temperature, and the proton dose delivered, with each fluence step, because displacement defects generally anneal at room temperature over days.
A single-event run records each event's initial signature and what restored normal operation. The recovery ladder is planned in advance and climbed one rung at a time: retry the operation, re-initialize the interface, reset the device, cycle its power. Each step is time-tagged, and the state after each step is verified by reading it (memory contents, configuration registers, output levels) rather than inferred from a command response. The interruption is measured from the event to validated function, which takes longer than the reset command. Persistent errors that survive a reset, and recoveries that fail, are kept as separate outcomes, since chapters 9 and 10 use them. The same apparent symptom (a bus that stops answering) has three different mission consequences depending on which rung clears it.
Classifying an event takes more than one observable. A supply-current step alone does not identify latchup. A functional interrupt can raise the current without a parasitic thyristor firing, and a limiter can mask a latchup. Current, voltage, functional state and the recovery step that cleared the event are recorded together. Events the record cannot classify are kept as unresolved rather than forced into the convenient category. Repeated read errors at one address are counted as one persistent event rather than many upsets.
Worked example 7.3. The U3 dose sequence and the U2 recovery ladder
U3, five parts, high dose rate. The steps are 10 krad(Si) to 30 krad, then 5 krad steps to 75 krad (10, 20, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75). The input bias current is measured within an hour of each step at 25 °C under the flight bias. The sequence ends with 24 h at room temperature and a final measurement; the 100 °C rebound test of TM1019 is a MOS provision and is not applied to this bipolar part. The circuit budget allows a bias current of 100 nA at the 100 kΩ source impedance of the application (a 10 mV error), and each part's trajectory is read against that limit. The failure dose used in chapter 8 lesson 3 lies between the last passing step and the first failing one. The record keeps both ends. The value quoted is the crossing interpolated on the measured trajectory between them, which is a convention stated in the record rather than a measurement.
U2, the recorder memory: after each interrupt the ladder is climbed and each rung is timed.
| Rung | Time budget | Verified by |
|---|---|---|
| detection | 0.5 s | the scrubber reports an unexpected pattern or the interface times out |
| isolation and reset | 2 s | controller asserts reset; rail current returns to the quiescent value |
| boot | 30 s | memory controller reports ready |
| reconfiguration | 10 s | mode registers read back as configured |
| reacquisition | 45 s | a known pattern written and read back without error; recording resumed |
The interruption, from the event to validated function, is 87.5 s, of which the reset command itself is 2 s. That 87.5 s is the downtime per event that chapter 9 lesson 3 multiplies by the interrupt rate. Measuring it as 2 s would understate the unavailability by a factor of 44.
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:
- When the parameters were measured relative to the exposure, and whether any anneal sequence followed, compared with the window and the sequence that TM1019 requires.
- The bias during the proton run, which is the operating mode chosen for the SEE test and may not be the bias that traps the most charge.
- Whether the dose arrived in steps with parametric measurements between them, as a cumulative test delivers it, or all at once with only a functional check at the end.
- The same question in the other direction: whether the dose the part accumulated during the proton run changed the SEE results themselves.
Exercise
For a cumulative test of your own, lay out the dose steps, and store the dose rate, bias and temperature history with every parameter measurement.
For each single-event mode that matters to the mission, write the recovery ladder, and record the signature before recovery, each action taken, the elapsed time and how the state was verified afterward.
Going deeper
Several mechanisms can move one parameter
Oxide-trapped charge, interface traps and displacement defects can each evolve at its own rate. The net drift of a parameter can therefore reverse direction, and a final measurement can hide an application failure that occurred at an intermediate step. Enough measurement points, with their times, are needed to tell the mechanisms apart, to put an uncertainty on the result and to support any extrapolation to another source or dose rate.
Keep each part's trajectory, and the fact that its failure dose is known only to lie between two steps, rather than replacing the set of curves with one average failure dose.
Classify with more than one observable
High current alone does not identify latchup, since a functional interrupt can change the supply current without any parasitic thyristor firing. Repeated read errors at one location may be a single persistent event rather than many independent upsets. Keep the unresolved classifications and the competing explanations in the record, and use further diagnostics or another experiment to separate the physical mechanism from the symptom the tester saw.
A latchup the part recovered from can still have caused latent damage. Recovery on the bench and long-term integrity of the part are separate claims.
The common mistake
A part that still works at the final dose may have been outside the application limits at an intermediate step.
An automatic reset can erase the initiating event and make different failure modes look the same.
Takeaway
- The measured dose response depends on how the dose was delivered and on when, and under what bias, the parameters were measured.
- The recovery ladder and its timing are what turn a device sensitivity into a downtime the system analysis can use.
Check yourself
Why is each part measured before its first exposure, when the datasheet already gives typical values?
Because the analysis follows each part's own trajectory from its own baseline. Without it, part-to-part spread that existed before irradiation is confused with radiation-induced change, and the interval-censored failure doses need the per-part record.
A U2 interrupt clears after a power cycle but not after a reset. How is it classified and why does it matter?
As a functional interrupt that requires a power cycle, kept separate from the reset-recoverable interrupts. The mission consequence differs (the downtime, what else is on the same rail, whether a power cycle is even available in flight), so the rates of the two classes are reported separately.
What does the rebound test at 100 °C look for?
Growth of interface traps after the trapped oxide charge has annealed. That growth can move a MOS parameter in the opposite direction from its post-irradiation value and, for some parts, can make the post-anneal state the worst case.
All five U3 parts meet the 100 nA bias-current limit at the final measurement, taken after 75 krad(Si) and the 24 h room-temperature anneal. A colleague reports the lot as passing 75 krad for the application. What can that final reading hide?
An application failure at an intermediate step. The net drift of a parameter can reverse direction as the mechanisms behind it evolve at different rates, so a part that drifted past 100 nA at 45 krad has failed at 45 krad for this application whatever it reads at the end, and a final pass or fail hides where the crossing happened. The tolerance bound of chapter 8 lesson 3 needs each part's whole trajectory, with its failure dose known to lie between the last passing step and the first failing one, which is why the record keeps every step and its measurement time rather than the end point alone.
Why does the U3 sequence take 10 krad(Si) steps to 30 krad and then 5 krad steps to 75 krad, rather than equal steps throughout?
The steps are chosen so that the parameter trajectory can be seen: generally coarse steps through the first fraction of the required dose, finer steps through the region where failures are expected, and one step beyond it. The required test dose for U3 is 30 krad(Si) (lesson 6.2), so the sequence spends its resolution above it, where the tolerance bound of chapter 8 lesson 3 needs the trajectories to failure. The failure dose of a part is known only to lie between the last passing step and the first failing one, so the step size in that region sets how well it is known, and the same steps are used for every part so that the trajectories can be compared.
The U2 reset command takes 2 s, and the interruption is logged as 87.5 s. A colleague calls the difference a tester artifact and uses 2 s as the device downtime. What is wrong with that?
The interruption is measured from the event to validated function, and each rung of the ladder is part of it: detection 0.5 s, isolation and reset 2 s, boot 30 s, reconfiguration 10 s and reacquisition 45 s, each verified by reading the state back rather than inferred from a command response. When the ladder is the flight-like recovery path (lesson 6.1), the 87.5 s is the downtime per event that chapter 9 lesson 3 multiplies by the interrupt rate, and using 2 s would understate the unavailability by a factor of 44. The reset command is only one rung, and the device is not available until the known pattern has been written and read back without error.
Further reading
- MIL-STD-883, Test Method 1019, "Ionizing radiation (total dose) test procedure."
- ESCC Basic Specification No. 22900, "Total dose steady-state irradiation test method."
- 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.
- 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.
- JEDEC JESD57A, "Test procedures for the measurement of single-event effects in semiconductor devices from heavy-ion irradiation," 2017.