2026年11月6日
批次一致性怎么做：同一型号跨批次差异的来料检验（IQC）、留样与 PO 条款
同一型号第二批货为什么和第一批不一样？本文从来料、主控与固件版本、封装、测试工位四个变量拆解跨批次差异的来源，给出一套可执行的来料检验（IQC）方案框架：约定抽样、对比基准样、留样与批次追溯、不合格品处置，并附把批次一致性写进 PO 与验收条款的清单。
本文提供英文版本。
A purchase order names one model code. The second delivery carries the same code and label, yet the units inside behave differently: a drive that mounts on one host and hesitates on another, a write run that no longer matches the first lot, a carton finish shifted a shade. Buyers file this as a quality problem. More often it is a consistency problem — nothing failed outright, but an input to the build changed and nobody wrote it down.
This article is for OEM buyers and engineers who approve successive lots of one model: where lot-to-lot variation comes from, what an incoming quality control (IQC) plan must contain to catch it, and how to write batch consistency into the purchase order so acceptance has a definition. Chip-level authenticity is a separate exercise; here the subject is keeping lot two identical to lot one.
Why Two Lots of the Same Model Can Differ
Four inputs move between lots. Each is legitimate on its own; the problem is that they move silently.
- Incoming material. Die from another wafer lot, a different grade mix inside the same capacity point, a new PCB laminate, a new resin batch on a card body.
- Controller and firmware. The silicon revision steps forward, or firmware is updated for yield or host compatibility. Firmware is the most common silent change: it carries no visible part number.
- Packaging and assembly. Housing tool wear, anodising batch variation, magnet grade in a magnetic portable SSD, print and laminate.
- Test station. Another fixture, host platform or ambient temperature, or a pass threshold relaxed to clear a backlog. This one hides best, because the report still says pass.
None of the four is a defect, so batch consistency is a change-control target, not a zero-defect one. The question is not whether a unit is good, but whether it is the unit that was qualified.
What an IQC Plan Has to Contain
An IQC plan built on checking a few pieces produces arguments. Five elements make it enforceable.
1. An agreed sampling plan. Size per lot and per defect class, agreed in writing before the first shipment, applied identically at both ends, with the fate of a failing lot stated up front.
2. A defined measurement, not just a limit. Same tool, host, test file and ambient temperature on both sides. Agree the method before the limit, or the numbers are not comparable.
3. A benchmark sample. A signed, dated, sealed reference unit from the qualified lot, held by both parties.
4. Retention samples. Units pulled from every lot and stored under controlled conditions.
5. A batch record. Lot number linked to test data, firmware, material batch and test station, retrievable by whoever answers the claim.
An IQC Checklist by Product Type
Checkpoints differ by product because failure modes differ; each line needs an owner and a recorded result, not a tick box.
| Checkpoint | NV002 (PCIe 4.0, 5000MB/s, 256GB~2TB) | PS5005 (Type-C 3.0, 2000MB/s, 512GB~2TB) | TFP3310 (TF card, SD 3.0 SDR104, 100MB/s, 32GB~256GB) |
|---|---|---|---|
| Identity | Model, capacity, firmware string | Model, capacity, bundle contents | Model, capacity, class marking, print |
| Performance | Sequential read vs the 5000MB/s rating | Sequential read vs the 2000MB/s rating | Sequential read vs the 100MB/s rating, same reader |
| Physical | 2280 length, thickness, gold fingers | Housing finish, magnet pull, port alignment | Card thickness, contacts, resin finish |
| Lot record | Lot number, firmware, material batch, station | Lot number, firmware, magnet batch, station | Lot number, resin and print batch, station |
Benchmark Samples, Retention and the Batch Record
The benchmark sample settles any argument about how a unit used to look and behave. Draw it from the qualified lot, record its measured performance, seal it, and have both sides sign; reissue it when an approved change lands.
Retention should cover the period in which a claim can arrive — in practice the warranty term stated for that model. Retained units cost little; a disputed lot without physical evidence costs a great deal. The lot number is what makes them useful: resolving to test data, firmware and material batch lets a supplier say within a day which units are affected and what changed.
When a Lot Fails: Segregation and Disposition
A failing lot leaves released stock immediately: labelled as held, physically separated, blocked in the warehouse system. Returning held stock to good stock is how a consistency programme collapses, because traceability dies with it.
Disposition follows a fixed menu: accept with a written deviation signed by the buyer, sort and rework, return and replace, or scrap. Deviation redefines the accepted baseline for later lots, so it should stay rare.
Corrective action must change a recorded input — material batch, firmware, test station — not the inspection threshold. Threshold drift is the quiet failure: a relaxed limit clears one lot, then hides the next four.
Writing Batch Consistency into the PO and Acceptance Clauses
A specification describes a unit; a purchase order has to describe a lot. Six clauses close the gap.
- Reference to the approved sample. Acceptance against the signed benchmark and its recorded data, not a catalogue description.
- Change notification before shipment. Any change to controller revision, firmware, memory source, housing, magnet grade or test station declared and approved before the lot ships.
- Lot identification. Lot number on the unit label, carton and shipping documents.
- Sampling and acceptance criteria. The agreed plan, method and disposition options, in an annex rather than an email thread.
- Retention obligation. Who keeps samples, how long, under what conditions.
- Traceability data per shipment. Test summary, firmware and material batch shipped with the lot, so IQC can be cross-checked.
These clauses matter more than the unit specification: the specification was qualified once, the clauses keep the tenth shipment comparable to the first. NV002, PS5005 and TFP3310 are each built and tested against a fixed internal baseline, available to buyers as a benchmark sample set.
Frequently Asked Questions
How many units should be pulled from each lot?
There is no universal number. Size belongs to the sampling plan agreed between buyer and supplier, per lot and per defect class, and should differ between a cosmetic check and a performance check. What matters is that both sides pull the same quantity by the same method.
Is a sealed benchmark sample enough on its own?
No. A sample without recorded measurements proves appearance only. Pair it with the figures taken when it was qualified, and with the firmware and material batch behind it.
Do retention samples have to last the whole warranty period?
They should cover at least the period in which a claim can be raised, usually the warranty term stated for that model. Storage conditions matter as much as duration.
Conclusion
Batch consistency is built from unglamorous parts: a sampling plan both sides accept, a method that does not move, a sealed reference with data behind it, retained units, a lot number that resolves to a record, and a purchase order that demands notice before a change ships.
Shenzhen LingC Industrial Co., Ltd. supplies storage programmes for OEM and private-label buyers and will give selection advice for your target scenario, plus a proposed IQC checkpoint list for the model you are qualifying. Contact Cherie at Cherie@lingctech.com.
本文亦发布于 lingctech.com 博客。查看线上原文
