Part Lookup and BOM Screener

Preliminary screen from public data and population priors. Read the sources before relying on a number, and treat every prior as a reason to test, not a substitute for one. Full disclaimer.

1. What it does

For one part number or a whole bill of materials: finds the part in the public parts index and reports the best public number for each radiation effect with its source documents; decides the process technology class through a ladder of evidence and grades the confidence; attaches the latchup prior for that class and the total dose prior for the device family; judges each effect against your mission dose, margin and LET goal; lists the actions (covered, extend, test, protect, replace); and flags the rows where the unresolved technology class is what the decision turns on. The screen exports as CSV and Markdown, and as a hand off file for the SEE Analysis Workbook (Export for SEEA), which turns every screened line into part instances with their public data, technology decision, priors and suggested effects.

2. Inputs

• Part number. Orderable or generic; TI orderables resolve to TI generics through TI's catalog, ADI orderables through the rules the reliability harvest validated. A manufacturer helps the prefix lexicon when the part is not in the index.
• BOM text. One part per line; tab, comma or space separated columns; manufacturer, quantity (a bare integer) and reference designators (U1, Q3, U1;U2) are recognized in any order; a header line is skipped; lines starting with # are ignored.
• Mission dose and margin. Dose behind your shielding in krad(Si) and the radiation design margin; their product is the level a public pass has to reach.
• Latchup LET goal. The clean LET a public SEL result has to reach to count as covered.
• SEL prior population. Which EHSP stratum the latchup priors come from; modern parts by default.
• Per row overrides. Device type and technology class selectors inside each row's evidence panel; a choice re-screens the row.

3. Reading a row

• Match. Exact, or the generic or base part the record was found under; R marks a part with additional records in restricted sources.
• Technology class. The class, its confidence (high, medium, low, or set by you) and the rung that decided it; decision sensitive when the runner up class would change the latchup or dose picture.
• Public data. The headline numbers: TID pass, SEL clean LET or observed onset, SEU onset, SEB or SEGR LET, DD; the expanded row lists everything with the documents.
• SEL prior. Fraction of parts of that class that latch below LET 75, or n/a for classes without a pnpn path.
• TID. Public pass and fail doses, or the prior median failure dose of the device family.
• Verdict. One chip per effect: ok (covered or not applicable), test (data missing or short), issue (a recorded failure or observed latchup).

4. Calculation

• Index numbers. Best public value per effect (maximum pass, minimum fail or onset, maximum clean level), public document classes only, vendor claimed and aggregator values excluded.
• Ladder. User setting, stated process, manufacturer family or parametric architecture, prefix lexicon, device default; the first rung that answers wins, all rungs that answered are shown.
• Lexicon. Counts of technology class by manufacturer and prefix, prefix, and letters, learned from every database part with a process; the top class and its share.
• SEL prior. EHSP posterior for the class in the chosen stratum with its 90 percent interval.
• TID prior. TDHP lognormal (mu, sigma) for the device family, by technology when that stratum has at least eight parts; fraction failing by dose D is Phi((ln D minus mu) / sigma).
• Sensitivity. Runner up class from the lexicon distribution or the device default; flagged when the SEL prior moves by 15 points or crosses 5 and 20 percent, or the TID p10 moves by a factor of two.

5. Assumptions

• A public number for a part number applies to your part; grade, package, lot and date code can break that.
• The class priors describe populations, not the part in your hand.
• Manufacturer parametric data describes the silicon sold today.
• A prefix bucket is representative of the part; families that changed process, and rebadged parts, break that.

6. Implementation notes

• The index and lexicon are static files built from the database and manufacturer harvests; the page does no network calls beyond loading them.
• The parts index is about two megabytes; the first load takes a moment, then lookups are instant.
• A part number in the URL (?pn=LM124J) runs the lookup on load, so rows can be linked.
• The build refuses to publish if a restricted source marker appears in the output.

7. Limitations

• No Maxim process families (not on ADI's reliability pages), no fab or process for TI beyond the parametric architecture, and no coverage of parts the database has never seen except through the lexicon.
• Restricted records are counted, never shown.
• One headline number per effect; the sources hold the conditions.

8. References

Back to the tool · Parts Advisor · SEL Threshold Predictor · TID Priors · SEE Rates