Counterfeit Component Detection — X-ray, Decapsulation, and Datasheet Forensics

When semiconductor allocation tightens — as it has consistently since 2021 — procurement teams reach for the open market. The open market is not a single market. It's a spectrum that runs from authorised franchised distributors at one end to unbranded reels in a back-room warehouse at the other. Counterfeits live everywhere along the middle of that spectrum, and they're not always cheaper.

The four kinds of counterfeit we encounter most often:

• Recycled — pulled from scrap boards, cleaned, re-marked, and re-sold. Often functional but with reduced reliability and unknown thermal history.
• Re-marked — a genuine part of a different speed grade, voltage, or temperature class, sanded down and laser-marked as the in-demand variant.
• Cloned — a fab-fresh die of a different (cheaper) design, packaged and marked to look like the target part. Functional behaviour can be wildly different.
• Empty — a hollow plastic body with no die inside. Rare, but we've seen it in shipments of high-value FPGAs.

"A counterfeit MCU that passes power-on self-test will still fail thermal cycling six weeks later in the field. The cost of the recall is two orders of magnitude larger than the cost of pre-screening." — Pioneer Horizon procurement QA

The cheapest screen catches the most obvious fakes. We run Tier 1 on every grey-market lot, sampling per AQL 1.0 (single sampling, level II). It takes our incoming inspector roughly 12 minutes per sample under a 50× stereo microscope.

What we look at

• Lead/pad condition — recycled parts show solder residue, plating wear, or non-uniform tinning. Genuine new parts have crisp, uniform leads with no wetting marks.
• Top-mark consistency — laser engraving depth, alignment to package edge, character font, and most importantly the relationship between the date code, lot code, and country-of-origin code. Inconsistencies across a single reel are diagnostic.
• Indent and pin-1 marks — mould tool marks should be sharp and uniform. Sanded re-marked parts often show a slight surface haze under raking light.
• Acetone wipe — a 30-second swab on a sacrificial unit. Surface re-marking dissolves; factory laser marking does not.

What this catches

Tier 1 catches roughly 60% of counterfeits in our experience. It catches almost all recycled parts and a meaningful fraction of re-marks. It does not catch competently re-marked parts at the high-end of the counterfeiter's craft, and it cannot detect cloned die.

Tier 2 adds 2D X-ray (and 2.5D oblique imaging for QFN/BGA) on a smaller sample — typically 5 units from any lot that passed Tier 1 with a sample size below 50. X-ray takes about 4 minutes per part on our setup; the analysis is the slow step.

What you can see in the X-ray

• Die size and position — a re-marked part will have a die that doesn't match the expected die size for that part number. Public die-size databases (Chipworks, TechInsights summaries, and our own internal records) make this comparison feasible.
• Lead-frame geometry — the lead-frame is part-number-specific. Cloned parts often use a generic lead-frame that doesn't match the original tooling.
• Bond wires — wire count, wire diameter, wire pattern. Genuine parts show the expected count and a deliberate routing pattern. Cloned parts show fewer wires or different geometry.
• Empty cavity — yes, this is rare but yes, we've seen it. X-ray catches it instantly.

When Tier 2 is required

We push to Tier 2 automatically on any line that:

1. Is single-sourced AND on allocation longer than 26 weeks (the temptation to substitute is highest here).
2. Has a unit price above $20 OR a BOM total exceeding $5,000 per board.
3. Is destined for an automotive, medical, or aerospace build.

For one customer's $400 FPGA, we found cloned die in 3 out of 50 sampled units — a 6% counterfeit rate from a "tier-1 distributor with paper documentation." That programme would not have survived field deployment.

Tier 3 is destructive. We chemically open the package (fuming nitric acid + sulphuric for plastic; mechanical for ceramic) and inspect the die under metallurgical microscope, then optionally under SEM for fine-feature comparison. This costs us roughly ₹3,500 per part and takes a working day end-to-end.

It is the only test that conclusively distinguishes between a genuine die and a clone of comparable size. We reserve Tier 3 for:

• Aerospace, defence, and medical programmes where field-failure liability is unbounded.
• Any lot where Tier 2 X-ray analysis returned ambiguous results.
• One-time vendor qualification for a new grey-market source.

What we record

• Die-shot photograph at 50×, 200×, and 1000×.
• Die markings (foundry code, mask revision, manufacturer logo).
• Cross-reference against a reference image from a known-good unit (we maintain a reference library for ~400 high-volume parts).

The pass/fail call is binary. If the die markings or visible structures don't match a known-good reference, the entire lot is quarantined and the vendor is recorded as suspect in our procurement database.

Counterfeit detection is a cost calculation. Every test catches more fakes; every test costs more. Here's how we draw the line in practice.

Always do Tier 1

On every grey-market lot, no exceptions. Cost: ~$5 per lot in labour. Catches the obvious stuff.

Add Tier 2 when…

• Unit price > $20 or
• BOM line is single-sourced or
• Programme is automotive, medical, or aerospace

Cost: ~$30 per sampled unit. Catches the smart fakes.

Add Tier 3 when…

• Unit price > $100 and field-failure cost > $1,000 per unit or
• Programme is safety-critical with no acceptable failure rate or
• Tier 2 returned ambiguous results

Cost: ~₹3,500 per part. Reserve for the cases where the alternative — a recalled product — would cost six figures.

The matrix isn't moral; it's economic. Counterfeit detection should cost less than the expected loss from undetected counterfeits. For most consumer programmes, Tier 1 is enough. For one safety-critical aerospace customer, every BOM line above $40 gets full Tier 2 + 1-in-200 Tier 3, and the line items still cost less than the field-recall reserve we'd otherwise have to hold.

Inspection is the last line of defence. The first line is supplier discipline. Five things we do — and recommend our customers do — to push counterfeits out of the supply chain before they ever reach incoming inspection:

1. Buy franchised whenever you can. Authorised franchised distributors (Avnet, Arrow, Future, etc.) have direct supply contracts with manufacturers. Counterfeits are statistically nil. The premium over grey-market is usually 5–15% — far less than the field-failure cost of a single bad lot.
2. Qualify alternate parts during design, not during shortage. If your BOM has a single sourced MCU, your design has a single point of failure. We help customers cross-reference and footprint-qualify a primary + two alternates on every line, so when allocation hits, we're not shopping the grey market under time pressure.
3. Build a vendor history database. Track every supplier you buy from, the parts you bought, the results of incoming inspection. Vendors with three bad lots get blacklisted; vendors with five years of clean lots get a faster sample-size reduction. Institutional memory beats individual judgement.
4. Demand chain-of-custody documentation. Original factory packaging (vacuum bag, MSL indicator card, foil bag), tube/reel labels matching the manufacturer's format, certificate of conformity. Forgery is possible but raises the bar.
5. Random-spot Tier 2 on low-risk franchised lots. Once a quarter, on a random franchised lot, run Tier 2 X-ray anyway. If the franchised channel ever drifts, you want to catch it before the field does.

For a structured walk-through of how we wire all of this into our incoming-inspection MES, see our per-board traceability article — the same audit trail that captures solder-paste batches captures component-lot inspection records.

If you're carrying a BOM you'd rather not lose sleep over, share it with our procurement team and we'll come back with a counterfeit-risk heatmap within five working days.