Total Dose Hardness Priors

Preliminary analysis aid. Check the result against your own analysis and, where needed, a test on the intended part. Full disclaimer.

1. What it shows

For each of 22 device families, the fraction of part types in the site's radiation test corpus that had a parameter reported outside its datasheet limit by a given total dose, as a survival curve against dose with bootstrap bands, at the standard test rate, an intermediate rate and the low rate; the same split by grade, test era and process; how often part types were flagged as enhanced at low dose rate and the distribution of the enhancement factor; and for optocouplers the CTR retention against 1 MeV neutron equivalent fluence with a fitted damage constant and a calculator. Release TDHP 2026-09, built from 6,639 records on 3,772 part type and dose rate combinations.

2. Inputs

• Device family. Families come from a keyword mapping of the device type field. Pick the one your part's function belongs to; the part type count is shown in the list.
• Dose rate class. Standard (1 rad/s and above, plus records with no rate stated), intermediate (0.01 to 1 rad/s), low (10 mrad/s and below, including elevated temperature tests). For a space mission the low rate curve is the relevant one for bipolar and BiCMOS linear parts; for CMOS the standard rate is the accepted test.
• Subset. COTS, Mil and Hi-Rel, or rad hard and QML, available for the standard rate only.
• Mission dose and design margin. Their product is the specification dose the headline number is read at.
• Optocoupler inputs. Particle, proton energy, fluence, LED type if known, and the minimum CTR fraction the circuit tolerates. Neutron input is taken as 1 MeV equivalent directly.

3. Calculation

• Observations. One per part type per dose rate class: highest dose passed, lowest dose with any parameter out of limit. Failures recorded at 0 krad and values below 0.5 krad are dropped as pre irradiation out of spec; doses above 3000 krad are right censored at 3000.
• Curve. Turnbull nonparametric maximum likelihood survival on a fixed dose grid; log interpolated for readout; curves stop at 1000 krad because the last grid cell holds the censored remainder.
• Headline fraction. One minus the survival at the specification dose. The band is the 5 to 95 percent bootstrap over part types (100 resamples), computed for family level curves.
• Lognormal summary. Interval censored maximum likelihood, shrunk toward the all-family fit with weight n/(n + 8); shown as an overlay and quoted as a median. Headline numbers do not use it.
• ELDRS. Susceptibility fraction with a Jeffreys Beta prior among part types with a recorded determination. Enhancement factor is the standard rate failure dose over the low rate failure dose for part types with both; lower bounds for part types that failed at low rate only are listed separately.
• CTR. Retention statements converted to 1 MeV neutron equivalent fluence by silicon NIEL and fitted to CTR/CTR₀ = 1/(1 + K Φ_eq) with lognormal K; median and 10th and 90th percentile curves; the calculator inverts the law for the fluence at which the retention reaches your minimum.

4. Assumptions

• "Out of limit" means what the test author judged against the datasheet or SMD limits. For a derated application the curves are conservative.
• The corpus is what people chose to test and publish. Failure fractions are upper bounds for a random catalog part and fair estimates for a part somebody intended to fly.
• Records with no dose rate stated are treated as standard rate tests.
• Publication year stands in for design era.
• Silicon NIEL scales GaAs and AlGaAs LED damage with energy well enough for a prior; it is not the right NIEL for the LED.

5. Implementation notes

• The published asset contains fitted parameters, grid survival values, bands and aggregate counts. Internal records are counted as internal and never identified; the anonymous optocoupler points carry only a fluence and a fraction.
• Families with fewer than five part types are omitted; the comparison chart requires ten.
• The build script, the family mapping, the dose rate parser and the fits are in one Python file and re-run against the database when the corpus grows; the release name changes when they do.
• The lot acceptance report generator evaluates test data on its own; these priors are for the decision before the test.

6. Limitations

• First parameter out of limit only. How far parameters drift, which a derated design needs, is a separate extraction not yet run.
• The optocoupler prior rests on 19 statements with a two decade spread in the damage constant; it says test, not how the test will come out.
• The ELDRS susceptibility fraction is biased upward by who gets tested at low rate; the enhancement factors are not.
• The intermediate rate class is a mixed population. The family mapper is keyword based and misfiles some parts.
• Grade, era and process subsets carry point estimates without bands.

7. References

Back to the priors · TID Lot Acceptance Report Generator · Parts Radiation Advisor