Blog · operations

PCB design review: the review the workspace cannot run

Does a design review inside the CAD tool cover what the PCB factory will check, and what has to be agreed at the handoff?

ChinaPCB Editorial 1814 words · 9 min read

The switching signals in this article

  1. 01 Two reviews, one name A design review approves intent; a CAM review tests files against drill tables, materials and machine limits. A pass in the first says nothing about the second.
  2. 02 The published asymmetry Sixteen capability lines publish an impedance tolerance to a percent. One line in 1,252 mentions a DFM service, and it sits inside a bundle of delivery promises.
  3. 03 Camera-ready is not buildable A release package identical to the one that was signed off can still carry a stackup the receiving plant does not laminate.
  4. 04 The unnamed review The manufacturability conversation usually happens as a CAM engineer's question on a live order, with no service level and no record you can approve.

A republished Altium article on design review and collaboration describes a workflow that most engineering teams would recognise: a shared workspace, comments attached to traces, copper planes and components, lifecycle states that move from Draft through In Review to Released, role-based permission over who may promote a design, and a packager that assembles the manufacturing outputs into a version-controlled release. It closes with the strongest claim in the piece, that once a release is approved “the contract manufacturer receives exactly the same package that was reviewed and signed off.”

That sentence is true, and it is where the article stops. A package being identical to the one your team approved is a statement about fidelity. It says nothing about whether the stackup can be laminated at that plant, whether the annular ring survives that drill table, or whether the impedance target has a coupon behind it. Those questions belong to a second review, run by different people, reading different documents, with the power to change different things.

Two reviews share one word

English uses “design review” for both. They are not the same meeting.

QuestionDesign review, in the workspaceCAM review, at the factory
Who runs itYour engineers, a reviewer outside the project, a hardware lead who signs the release.A CAM engineer and a process owner, on the plant that will run the order.
What it readsSchematic, layout, the design rules your team wrote, the BOM.Your Gerber or ODB++, drill file, fabrication drawing, stackup and quantity.
What it can changeTopology, part choice, layer count, net class. The design itself.Tooling, panelisation, etch compensation, drill sequence. Not your intent.
What it producesA resolved comment set and an approved revision that becomes immutable.A DFM report, a set of engineering questions, a process route and a delivery date.
What it cannot seeThe capability of the plant at your quantity, in your material, on the week you need it.Whether the intent was right. A buildable board can still be the wrong board.

The two fail in different directions. A missed design review usually produces a board that works, costs more, and cannot be tested at volume. A missed CAM review produces a board that matches your files exactly and still misses your specification: an annular ring below the plant’s minimum, a stackup nobody has in stock, an impedance target that appears in the notes but has no coupon and no measured result.

What the collaboration stack does automate

The features in the vendor article are real and worth using. An anchored comment is better than an annotated screenshot because it stays attached to the element it criticises. A lifecycle gate with role permissions is better than a verbal sign-off because it records who promoted the revision. An immutable Released state is better than a shared folder because it stops the quiet overwrite that turns two teams into two truths. The packager is better than a zip file assembled by hand at 23:00 before a shipment.

All of it is decision control. It answers one question well: did the right people approve this version, and is this the version they approved. It cannot answer the second question, because the second answer depends on a drill table, a laminate stock list and a lamination schedule that belong to the supplier.

There is a specific failure mode worth naming. A release gate is only as strong as what it blocks. If a design can pass Draft, In Review and Released while carrying a stackup the receiving plant does not build, then the gate approved a document rather than a product. The gate did its job; the job was narrower than the team assumed.

What 1,252 capability lines say about the second review

Every one of the 131 published factory profiles carries a capability table: item, value, source. Across the directory those tables hold 1,252 rows. All 131 profiles sit at the same evidence level, public-sources-reviewed, and none has been independently verified. That matters for how the counts below should be read: a capability line is what a factory chose to publish, not the whole of what it does.

Element of a manufacturability reviewCapability lines that describe it
Controlled impedance, with a tolerance15
Stack-up options offered3
A design or DFM review service1
Gerber, CAM or ODB++ review0

The asymmetry is the finding. The parameter a buyer can put a number on is published widely: Shennan Circuits at ±7 percent in mass production and ±5 percent on samples, Kinwong at ±7 percent on high-layer-count work, Glorysky at ±8 percent in 2026 and 2027 with a ±5 percent target for 2028, HannStarBoard Jiangyin at ±10 percent standard and +8 percent on advanced inner layers. The process that decides whether any of those tolerances survives production is mentioned once.

That one line belongs to Yuejie in Dongguan, and the row reads: “99% sample on-time delivery, 24-hour quick-turn, 2-hour quotation, free DFM and traceable manufacture.” Five commercial promises in a single cell. “Free” answers the price of the review, not its depth, and a promise bundled with four others is a service level rather than an engineering procedure.

Three other profiles make the same point from different angles:

  • JLC describes engineering support as standardised: a DFM check and a fixed option set rather than a conversation. For a standard four-layer order that is proportionate. For a 0.4 mm pitch BGA escape it is not, and the record itself tells you which kind of order the service was designed around.
  • Huaqiu is the clearest case of a check and a plant sitting in different places. The parent runs an EDA tool, a DFM checker, a component mall and the ordering front end; fabrication happens in subsidiaries, including the Jiujiang plant. The check you receive is software operated by the sales entity. That raises two questions no order form asks: which site builds, and does the checker use that site’s capability file.
  • Jingchengda states the whole problem in one instruction: ask which number the DFM check will use. On a display module with a fine-pitch bonding finger, that 5 µm is the difference between a yield and a rework loop.

The review that happens and never gets published

There is no shortage of manufacturability review in Chinese PCB production. There is a shortage of it in writing. What usually happens is a message from a CAM engineer on a live order: this pad is 0.15 mm under our etch compensation, do you want us to adjust, or do you want to revise the file. That message is the real design review of the manufacturing phase. It is also unlogged, unnamed, and carries no turnaround commitment.

The directory’s own editorial makes the same point from the buyer’s side: 96 of the 131 profiles close with a list of things to ask for, because the answers are not on the page. Victory Giant puts it directly — the dimensions that decide whether a stackup is buildable and at what yield are not published, and the answer should arrive as a process capability sheet with a revision date.

Six questions the workspace cannot answer

The list is short because only a few of these decisions are actually made at the plant.

  1. Which stackup will you build, layer by layer? Not the total thickness: the dielectric thickness and copper weight per layer, in the material you hold. Three capability lines describe stack-up at all, and CMK Wuxi is one of them, with 10-layer through-hole, 1-2-1 and 1-4-1 builds.
  2. What is your minimum annular ring for the drill table I sent, at my copper weight? The ring is a function of drill tolerance and layer registration, both plant-specific. You can compute it only after someone tells you both numbers.
  3. Will you build an impedance coupon, to which tolerance, and who signs the result? A published tolerance is a promise; a coupon with a measured value is evidence. Ask for the second before the panel is in the tank.
  4. How will you panelise, and does the array change my unit cost or my assembly yield? Panel choice is a fabrication decision with assembly consequences. A poor array is a cost line and sometimes a solder defect.
  5. Is my via filled and planarised, and where is that written? Via fill and any-layer structures are published capabilities; whether they apply to your via is a drawing question, and a drawing note beats an email.
  6. Who answers a technical question on a live order, and within what time? No profile in the directory publishes this. It is the single number that decides whether a question costs an hour or a day.

A review record you can approve

A usable manufacturability response has six fields, and a reply with fewer is a flag rather than a record:

  • the part and the revision it refers to;
  • the capability it was checked against, named as a document with a date;
  • the location of each finding, on a layer or a marked image;
  • the disposition offered: accept as is, waive with a reason, or change the design;
  • the owner of the decision and the date;
  • the file set reviewed, with a hash.

“No issues found” is only useful with the last two. Without a date and a file identity, the statement cannot be produced eighteen months later when a panel is in dispute and the question is which files were looked at.

One distinction is worth enforcing in the report itself: separate hard limits from cost suggestions. A fab will flag features it cannot build and features it would rather not build, and the two need different decisions. A waiver that arrives unlabelled becomes a quality claim later, and the argument is then about who agreed to what.

When a second pair of eyes earns its cost

Pay for an independent review when a failure would not be caught by electrical test and a respin costs more than ten times the review. In practice that means a first build at a new site, a fine-pitch BGA escape, an aspect ratio above 10:1, controlled impedance without a coupon, an unusual laminate, a rigid-flex stack, or a board that must fit an existing test fixture.

On a standard four-layer, 1.6 mm board at 6/6 mil, the plant’s CAM check is the review. A third-party read adds a day and rarely changes an outcome. The judgement is not about rigour in general; it is about which single decision, if wrong, would cost a panel, a week or a customer.

The fabrication topic follows the same package onto the production floor, the design topic covers the decisions that are settled before it is exported, and the RFQ checklist is the file list that makes a written answer possible.

Questions this article answers

Is a free DFM check the same thing as a design review?

No. A free DFM check compares your files with the supplier's rules and returns a pass or a list of flags. A design review asks whether the intent is right, and it can change the schematic. Only the second one produces a decision you approve and a record you can defend later.

Can a PCB factory change my design without asking?

It should not, and any adjustment it proposes should come back to you as a written waiver or an engineering change tied to one revision. The common failure is not a secret redesign; it is a verbal adjustment to a pad, a mask opening or a via that never reaches the drawing.

Does passing DRC mean the board can be manufactured?

No. DRC checks the rules you entered in the tool. DFM checks the same geometry against one plant's drill table, material stock, lamination schedule and machine limits. Two factories can return different answers on the same file and both can be correct.

How many factories in this directory publish a design review service?

One of 131 profiles mentions DFM at all, as one item inside a bundle of delivery promises, and 1,252 capability lines contain no mention of Gerber, CAM or ODB++ review. That is a statement about published records, and it is why the six questions below are worth asking in writing.

What should I send to get a useful manufacturability answer?

One package at one revision, with the fabrication drawing, the layer-by-layer stackup, an impedance table if any net is controlled, the quantity and the plant you expect to build it. The RFQ checklist covers the rest of the file set.