Per-Board Serialisation: From Solder Paste Batch to Box Build, Audit-Ready

Every board that comes off our lines carries a unique serial in the form of a 2D Data Matrix code, laser-marked on a copper-free silk pad before solder paste is printed. The serial is permanent — it survives reflow, conformal coating, and box-build assembly. Within our MES that serial is the primary key that pulls up every event in the board's manufacturing history.

What a serial query returns in under five minutes

• Bare PCB lot, fab vendor, lot date, IPC-A-600 acceptance record.
• Solder paste batch, manufacturer lot, refrigeration history, time-out-of-fridge before use.
• Stencil ID, last clean cycle, last tension check.
• SPI (Solder Paste Inspection) result per pad (or per joint, depending on programme) — area, volume, offset.
• Component reels consumed, manufacturer lot codes, incoming inspection result, splice events.
• Pick-and-place machine, programme version, feeder positions, placement-rate exception events.
• Reflow profile actual — peak temperature, time above liquidus, ramp rates — captured by the on-board profiler.
• AOI result, X-ray result for BGAs, ICT result with margin per net, functional test result.
• Operator IDs at each manual station with cert currency on the date.
• Box-build station events, torque-driver records, final QC sign-off, dispatch date.

"The first time we ran a full trace on a customer's field failure — a board that had been in service for 14 months — we pulled the reflow profile, the paste lot, the X-ray scan, and the ICT margins on every signal within four minutes. The root cause was identifiable that day. Without that trace, the investigation would have run for weeks." — Pioneer Horizon traceability engineer

That five-minute trace is the deliverable our customers value most when something goes wrong. It is also what an IATF Stage 2 auditor or AS9100 auditor will demand — and it is what failed audits typically lack.

Many EMS operations record traceability at the lot level — "this lot of 200 boards was built on line 3 between 09:15 and 11:42 on 18 July with paste batch P-2025-0712." That's better than nothing, but it conflates 200 boards into one record. When one of them fails in the field, you don't have the data to distinguish a paste-volume outlier on serial #134 from a process-mean drift.

Per-board events we record

• SPI volumes per pad (typically 60–800 pads per board, all stored).
• Reflow profile per board — the on-board profiler ties the thermocouple curve to the serial captured at the conveyor entry.
• AOI pass/fail per joint with image link for any flagged joint.
• X-ray BGA void measurement per BGA per board for safety-critical builds.
• ICT result per net per board.
• Functional test pass/fail per test step per board.

What this costs in storage

A complex board with 600 SPI pads, two BGAs, a 200-net ICT, and 40 functional test steps generates roughly 30–50 KB of structured data per board. At 50,000 boards/month, that's 1.5–2.5 GB/month — trivial by modern storage standards. The cost is overwhelmingly in the engineering effort to wire the data sources cleanly into a single schema, not in storage.

Schema discipline

One table per measurement type, foreign-keyed to a master Board table. The master Board row has the serial, the lot association, the programme ID, the dispatch destination. Every event table has board_serial, event_timestamp, station_id, operator_id, instrument_id, value, pass_flag. Get this right on day one — restructuring later is months of work.

What per-board enables analytically

Once the data is per-board, you can do things the per-lot model can't: serialise SPI volume distributions to see process drift within a single shift, correlate AOI false-positive rates with stencil cleans, find the operator whose station's first-pass yield runs 0.3% below the team mean across 5,000 boards. Quality improvement becomes data-driven instead of folklore-driven.

The serial is useless if you can't read it five years from now in a field-return depot. Marking method, location, and contrast all matter. We have evolved a standard that survives our typical product life — typically 7–10 years on industrial boards, longer on aerospace.

Marking method

• Laser marking on a copper-free silk pad — our default. 5×5mm Data Matrix, 0.25mm cell size, ECC 200. Laser is fibre, 20W, ablating the silk layer to expose the substrate. Survives reflow, conformal coating, and decade-scale field wear.
• Inkjet on bare PCB — backup option for very low-cost programmes. Survives reflow if properly cured. Less robust against solvents.
• Laser direct on copper — for products without silk in the marking zone. Annular marking against a darker oxide; readable but lower contrast.

Location rules

• Visible from the top face after assembly, not obscured by the tallest component within 10mm.
• Not under a heatsink, daughter-card, or shield can.
• At least 5mm from any board edge — protects against board-routing damage.
• Coordinated with the customer's documentation so the field tech knows where to look.

Readability validation

We sample-read every 250th board with a fixed handheld reader at the dispatch station. If any single board fails to decode on first attempt, we audit the previous 250 and adjust the laser parameters before continuing. ISO/IEC 15415 grade B or better is our standard; grade C is the rejection threshold.

Conformal coating interaction

Acrylic and silicone coatings preserve readability. Polyurethane and parylene can reduce contrast — for products in those coating families we either mask the Data Matrix during coating or specify a higher laser-ablation depth pre-coat. Validation is part of FAI for any new coating choice.

Traceability data is only useful if it's accessible — quickly, by the people who need it, in formats their auditors will accept. We have built a small library of standing queries that cover the questions auditors actually ask. The queries return in seconds; without them, the same questions would take hours.

Standing audit queries

1. Serial → full birth certificate — every event, every record, one PDF, under five minutes.
2. Paste lot → all boards consumed — list of serials, lots, dispatch destinations. Used for paste-related field-failure trace-backs.
3. Component lot → all boards consumed — same shape, used for counterfeit or recall scenarios.
4. Operator → all boards touched in the last 90 days — used during operator-cert investigation or quarterly review.
5. Instrument → all measurements during a given window — used when an instrument is found out of calibration retroactively.
6. Programme → first-pass yield by station by week — quality-trend report.
7. Customer → all serials dispatched, by date range — RMA reconciliation.

Retention periods

• Automotive (IATF) programmes — production life + 15 years per IATF clause requirements. We default to 20.
• Aerospace (AS9100D) — typically contractual; common contracts demand life of part + 10 years. Some defence programmes require longer with no fixed end.
• Medical (IEC 60601 / ISO 13485) — life of device + 5 years minimum; many customers specify life + 15.
• Consumer / industrial general — 7 years post-dispatch is our minimum.

Storage tier strategy

First 18 months: hot storage, queryable in seconds. Months 19–60: warm storage, queryable in under 30 minutes. Beyond 60 months: cold storage, queryable within 24 hours. The cost ratio across tiers is roughly 1 : 0.1 : 0.01. For a typical 10-year retention footprint we run about 8% of total data in hot storage and the cost is comfortably below 0.5% of total quality cost.

For the upstream chain — how component-lot data enters the system — see counterfeit detection.

Last quarter we received an RMA from a Tier-1 automotive customer: a single board returned from a field-failure event 11 months after dispatch. The board had failed intermittent communication on a CAN bus, and field diagnostics suggested a marginal solder joint on a transceiver pin. The customer wanted root cause within five working days. We had it within ninety minutes of the board arriving at our dock.

The walk-through

1. Minute 0 — board arrives, dispatch barcode is scanned, the master birth-certificate query runs automatically and prints to the investigating engineer's screen.
2. Minute 5 — birth certificate shows: built 14 March, paste batch P-2025-0307, SPI on the transceiver pin in question was 89% volume vs 100% target — a yellow flag at the time but inside acceptance band.
3. Minute 12 — reflow profile per-board captured. Profile shows peak temperature on this board was 244°C (target 246°C, low-side acceptance 242°C). Combined with the 89% paste volume, this is a marginal joint by intersecting metrics.
4. Minute 22 — AOI image of the joint pulled. Pad shows slightly insufficient wetting angle on the pin in question — within acceptance, but visually at the lower bound.
5. Minute 35 — X-ray of the joint (we keep X-ray images for safety-critical BGA pins but in this case the pin was a QFN). No X-ray for QFN pins. Decap arranged.
6. Minute 60 — same paste-batch + reflow-window combination queried across the rest of the production. 47 other boards from the same shift fell in the marginal intersection. Customer is notified preemptively; field-monitoring extended on those serials.
7. Minute 90 — root-cause documented: combined paste-volume and reflow-temperature drift produced marginal joints on a small subset of boards. Containment plan in place; corrective action initiated on reflow profile setpoint and SPI acceptance limit.

What this looks like without traceability

Without per-board records, the same investigation would have taken weeks. The customer would have received a "we're investigating" letter, the 47 other boards in the marginal cohort would have shipped silently, and any one of them could have produced a second RMA before root cause emerged. The traceability investment paid for itself in this single incident.

If you're building a traceability backbone or considering an MES upgrade, our manufacturing systems team can scope a pilot in two weeks.