space-rha.comShort courseChapter 7 · Test execution and observation

Chapter 7 · Lesson 2 of 4 · about 12 minutes

Dosimetry and live time

The machine setting and the fluence at the active region are different quantities. Confirm the beam position, field size and coverage at the device itself, and keep the monitor reading, its calibration and the material the units refer to, so that the delivered exposure can be reconstructed after the campaign.

What you need to know

The dosimeter sits somewhere other than the sensitive volume. A facility reports particle fluence from a monitor (a scintillator, a Faraday cup, a set of detectors around the beam) that has been calibrated against the field at the device plane. Whether that field is uniform depends on the setup. A sample placed near the edge of the collimated field, or a large board with devices at several positions, sees a fluence that differs from what the monitor reports. The record therefore keeps the beam profile or uniformity figure, the position and size of the device in the field, the calibration of the monitor and the date of that calibration. Proton facilities built for radiotherapy often report dose in water or monitor units. Converting that to proton fluence at the device needs the energy at the device plane after any degrader and the stopping power at that energy, and the conversion is written out once with its factors so that it can be checked.

Geometry enters twice in a tilted run. The facility quotes fluence normal to the beam. The device presents a projected area reduced by cos θ, so the fluence through the device plane is the beam fluence times cos θ. The path through a thin sensitive volume grows by 1/cos θ, and the effective LET is quoted as LET/cos θ. A test that applies the cosine to the fluence and quotes the effective LET has used the thin-slab convention. A test that does both and then also takes the facility's "effective fluence" has corrected twice. The plan names one convention, and it is applied once, at a point the record identifies.

Live time is the fraction of the beam-on interval in which the device was in its intended state and the acquisition could record an event. Every reset, every readout that blinds the monitor, every buffer overflow and every interval in which the beam was on while the board was being reconfigured takes something off it. The fluence in a cross section is integrated over the live intervals only. This correction is valid for a stationary single-event process when the device state is known. It cannot restore events lost to a mechanism that the setup could not observe, and it cannot be applied to a burst that saturated the record, because the burst itself was not observed. The flux is chosen so that the setup can keep up with the expected event rate. A SEFI-prone device at a high flux produces interrupts faster than it recovers, and the run then measures the recovery time instead of the cross section. For a pulsed beam the spill structure is recorded as well, because pileup depends on the instantaneous flux rather than on the run average.

The uncertainty in the exposure has more than one part. Monitor calibration and field gradients are common to every point taken in a setup, so they shift the whole curve rather than scatter the points. Energy spread, positioning and a beam that drifts during the run are partly random. The dosimetry uncertainty of a heavy-ion facility is generally quoted near 10 %, and in the budget of chapter 8 lesson 4 it is kept separate from the Poisson uncertainty of the counts.

Worked example 7.2. Tilt normalization and live fluence for a U2 run

A U2 run with krypton (surface LET 30.9) is tilted to reach a higher effective LET. The facility delivers 10⁷ ions per cm² normal to the beam.

Anglecos θFluence through the device planeEffective LET from 30.9 MeV·cm²/mgPath through 50 µm of overburden
0°1.0001.0 × 10⁷30.950 µm
30°0.8668.7 × 10⁶35.658 µm
45°0.7077.1 × 10⁶43.671 µm
60°0.5005.0 × 10⁶61.7100 µm

The run at 45° is logged with 7.1 × 10⁶ ions per cm² and an effective LET of 43.6. The ion, energy and angle stay in the record next to those derived numbers. In a separate run at normal incidence, 600 s at 1.0 × 10⁴ ions per cm² per second, U2 interrupted 4 times and was reset, blind for 15 s each time. The delivered fluence is 6.0 × 10⁶ ions per cm² and the live fluence 5.4 × 10⁶ (90.0 %). The SEU cross section uses the live fluence. The SEFI count of 4 uses the fluence up to each interrupt, which is the same live fluence to within the last interval. The resets themselves are the SEFI events, and they are not corrected away.

Exercise

For one run of your own, write down the monitor calibration, the position and size of the device in the field, the local coverage and the exact normalization applied to the exposure, with the point at which it was applied.

Store the beam status, the device state and the acquisition readiness on one time base, and mark the intervals that no correction can recover.

Going deeper

Where the exposure uncertainty comes from

The exposure uncertainty can include the monitor calibration, field gradients, positioning, energy spread and changes during a run that nobody measured. Several of these are common to every point taken in one setup, so they shift the whole curve together. A target stack also changes the particle energy, the LET and the secondaries, so one percentage on the incident fluence does not always describe the uncertainty at the sensitive region.

Archive the beam maps and the calibration records, and transport the beam to the active region when a surface measurement does not represent where the interaction happens.

When the live-time correction holds

For a single-event process whose rate is steady while the device is in its intended state, the exposure lost while the device was blind can be taken out of the denominator, provided the state of the detector and of the device is known throughout. Pileup that was not seen, a recovery that depends on the device state, or a failure correlated with the beam can each defeat a simple dead-time correction.

Keep the measurement of recovery time apart from the measurement of exposure in the sensitive state. Both are useful, and they answer different engineering questions.

The common mistake

Gy(water), rad(Si), monitor units and particle fluence are four different quantities, and the conversion between them has to be written out.

A live-time percentage corrects for the intervals in which you know the device was blind. It cannot recover events lost to a detection failure you did not know about.

Takeaway

  • A cross section or a dose-response curve is only as traceable as the exposure in its denominator.
  • A bias in that denominator passes straight through into the rate prediction, however precise the prediction looks.

Check yourself

A proton facility reports the exposure as 20 Gy(water). What is needed to turn it into a fluence at the device?

The proton energy at the device plane (after the degraders and the beam line), the stopping power of water at that energy to convert the dose to a fluence, and a check that the device was inside the uniform field. The dose in silicon is a separate conversion that uses the silicon stopping power.

A run plan says "fluence corrected for tilt" and the facility software also reports "effective fluence". What is the risk?

A double correction, because both apply cos θ to the normal fluence. The plan names one convention and the point at which it is applied, and the record keeps the raw beam fluence and the angle so that either quantity can be reconstructed.

Why does a high flux ruin a SEFI measurement even though it shortens the run?

Because interrupts arrive faster than the device recovers. The live fraction collapses, the record contains overlapping recoveries, and what gets measured is the distribution of recovery times under pileup rather than the interrupt cross section.

In the normal-incidence U2 run, the SEU cross section uses the live fluence of 5.4 × 10⁶ ions per cm². A colleague argues that the 4 interrupts should likewise be set aside as dead-time artifacts. What is wrong with that?

The resets are the SEFI events. The live-time correction removes from the denominator the 15 s intervals in which the device was blind; it does not remove the events that caused the blindness, which are the numerator of the SEFI cross section. That count uses the fluence up to each interrupt, which is the same live fluence to within the last interval, so the SEFI result is 4 events against about 5.4 × 10⁶ ions per cm². Setting the resets aside would turn a measured interrupt cross section into a clean run that never happened.

During a U2 run a burst of upsets overflows the acquisition buffer for part of the exposure. The team proposes to treat the overflow like a reset and remove that interval from the live fluence. Why does the lesson not allow it?

The live-time correction holds for a stationary single-event process in which the device state is known and the acquisition could have recorded an event; it removes exposure from intervals known to be blind, on the understanding that the events outside them were seen. A burst that saturated the record was itself not observed: the count during it is unknown, so no adjustment to the denominator restores it, and the interval is marked as one that no correction can recover. The flux is chosen so that the setup keeps up with the expected event rate, and a run that overflowed is generally repeated at a lower flux rather than corrected.

A fit to the U2 cross-section curve weights each point by its Poisson error alone. What does that treatment leave out, and why does it matter for the result?

The exposure uncertainty. Monitor calibration and field gradients are common to every point taken in one setup, so they shift the whole curve rather than scatter the points, and the per-point Poisson errors say nothing about that shift. The dosimetry uncertainty of a heavy-ion facility is generally quoted near 10 %, and chapter 8 lesson 4 keeps it as a separate term in the budget, since a shift in the denominator passes straight through into the rate prediction however precise the counts look. Energy spread, positioning and a beam that drifts during the run are partly random and are treated on their own.

Further reading

  • JEDEC JESD57A, "Test procedures for the measurement of single-event effects in semiconductor devices from heavy-ion irradiation," 2017.
  • ESCC Basic Specification No. 25100, "Single event effects test method and guidelines."
  • ASTM F1192, "Standard guide for the measurement of single event phenomena (SEP) induced by heavy ion irradiation of semiconductor devices."
  • R. Koga, "Single-event effect ground test issues," IEEE Trans. Nucl. Sci., vol. 43, no. 2, pp. 661-670, 1996.