HDI PCB Engineering Queries: Which Answers You Cannot Take Back

A quote confirmation does not start production

There is a gap in most HDI purchase orders that nobody plans for, and it sits in a place where buyers are not looking.

The sequence everyone expects is: send the data package, get a quote, approve the quote, issue the PO, and wait for boards. The sequence that actually happens inserts one more step between the PO and the first process operation. The fabricator's engineering group opens your files, compares them against what the line can physically do, and sends back a set of questions. Different suppliers call it an EQ, a technical query, a DFM query, a CAM question, or simply "a few points to confirm." The name varies; the function does not.

Nothing is cut, drilled, imaged or laminated until that exchange closes. Production starts when the answers come back, not when the PO is issued.

This matters for two separate reasons that buyers tend to collapse into one. The first is schedule: every day the EQ sits unanswered is a day the board is not being built, and whether that day comes out of the quoted lead time or gets added to it is usually undefined. The second reason is the one that costs more money and gets noticed later. Some of those questions are not questions. They are requests for permission, and the permission they are asking for is permanent.

An HDI build makes this sharper than a conventional two-layer board, for a structural reason. HDI designs carry more interdependent constraints — sequential lamination cycles, microvia structures that sit on top of or beside each other, aspect ratios tied to core thickness, trace geometry tied to copper thickness and to the dielectric that ended up between layers. Change one input and several others move with it. A conventional board can often absorb a small substitution without consequence. On an HDI build, the same substitution propagates, and the person who approved it in a one-line email reply may not have traced where it landed.

Why the reply is where authority changes hands

Consider the shape of a typical engineering query. It arrives as a numbered list. Each item describes something the fabricator found, states what they propose to do about it, and asks you to confirm. The tone is cooperative and the proposals are usually sensible. The format invites a short answer.

That format is the trap. A numbered list flattens items of very different consequence into a uniform shape, and the uniform shape invites a uniform reply: "All points agreed, please proceed." One sentence, sent in good faith, in the interest of not delaying the build.

What that sentence does is variable. For some items it changes nothing you would ever measure. For others it moves a design decision from your side of the table to theirs, and — this is the part that surfaces months later — it also moves the basis on which the board will be judged at incoming inspection. If your drawing said one thing, and an email reply agreed to something else, and the boards were built to the email, then at receipt you are holding boards that do not match your drawing and an agreement that says that is fine. Whether that is a problem depends entirely on whether anyone understood the consequence at the time.

So the useful question is not "should I agree to this?" Most EQ proposals are reasonable and agreeing is normally correct. The useful question is narrower and it can be applied item by item:

> Does this answer change the basis on which the finished board will be judged?

If yes, the buyer answers, in writing, in a form that updates the controlling documents. If no, the answer can be delegated — and delegated in advance, so it never generates a query at all.

Six kinds of query, three kinds of consequence

Engineering queries on HDI builds cluster into recognisable types. Sorting them by what the answer does — rather than by what the question is about — is what makes the list actionable.

1. Substitution. The fabricator does not hold the exact laminate, copper foil, prepreg or core thickness your stackup names, and proposes an equivalent. The word "equivalent" is doing a lot of work here. Equivalent in what respect: thermal, electrical, mechanical, or commercial availability? A substitution that preserves dielectric behaviour may change thermal performance, and vice versa. Consequence: high. This changes the physical board and usually changes the stackup drawing. It belongs to the buyer.

2. Rounding. A finished hole size, a slot width, an outline dimension or a copper weight does not land on a value the line produces, and they propose the nearest one they do. Often genuinely trivial. Occasionally not — if the dimension was driven by a connector, a press-fit part, a mounting interface or a mating assembly, the nearest available value may not fit. Consequence: depends entirely on whether the dimension had an external interface. Only the buyer knows that.

3. Re-solving. Your controlled-impedance target cannot be met with the trace geometry you drew on the stackup that materialised after substitution, so they propose adjusting trace width, spacing, or the layer the net sits on. This is the most consequential category and the one most often waved through, because it reads as a technical detail inside the fabricator's domain. It is not. Re-solved geometry changes your routing, and sometimes changes which nets are adjacent to which. Consequence: high, and propagates.

4. Conflict. Two of your own documents disagree — the drawing says one thing, the fabrication notes another, the Gerber a third — or your keep-out conflicts with something the process requires. The fabricator is not asking for permission here; they are asking you to resolve an ambiguity that only you can resolve. Consequence: this one cannot be delegated under any arrangement. Delegating it means the fabricator picks which of your documents to believe.

5. Process addition. They need to add something your data does not show: tooling holes, a process border, panel-level fiducials, test coupons, a different array format, a different depanelisation method. Mostly harmless to the circuit. Occasionally relevant — a process border changes panel utilisation and therefore price, and a depanelisation change affects board edges and mechanical fit. Consequence: low to medium, but check whether it touches price or edges.

6. Trade-off. Solder mask registration on fine-pitch pads cannot hold both the clearance and the dam width you drew; one has to give. Here the fabricator is telling you something true about physics and asking you to choose which consequence you prefer. Consequence: medium to high, and the choice depends on your assembly process, which they cannot see.

Reading that list, a pattern emerges. The categories that must stay with the buyer are the ones where the deciding information lives outside the fabricator's field of view — mating parts, enclosures, assembly process, the electrical intent behind an impedance number, which of your documents is authoritative. The categories that can be delegated are the ones where the fabricator has strictly more information than you do about the only thing that matters.

What can be pre-authorised, and how to bound it

The fastest EQ cycle is the one that never opens. If you tell the supplier in advance how to handle the predictable classes of question, those items do not come back as questions.

Pre-authorisation works when it is written as a bounded rule rather than as general goodwill. "Use your best judgement" is not pre-authorisation; it is an unbounded transfer of authority that will be read, correctly, as permission for anything. A usable pre-authorisation states the class of change, the limit, and the reporting obligation:

• Which class — hole size rounding, process border addition, fiducial placement, array format.
• Within what limit — expressed as a direction and a boundary. Rounding permitted in one direction only. Substitution permitted within a named property that must be preserved. No change permitted to any dimension carrying an interface callout.
• Reported how — the critical half, and the half that gets dropped. A pre-authorised change still has to appear somewhere you will see it, because the board you receive will have it. If it is never reported, you have authorised an invisible change and you will rediscover it during a second build with a different supplier who does things differently.
• Expiring when — pre-authorisation granted for one build is not pre-authorisation for the reorder, especially if the reorder runs on different material.

Equally important is the list that is explicitly not pre-authorised, stated as such: controlled-impedance geometry, any dimension tied to a mating part, laminate family changes, microvia structure changes, and any resolution of a conflict between buyer documents. Writing the exclusion list down is what makes the inclusion list safe.

Four ways to answer, four different levels of enforceability

The same decision, communicated four different ways, has four different standings when the boards arrive and something does not match. This is the part of the EQ process that buyers almost never think about at the time, and the part that decides the outcome of every subsequent dispute.

Verbal or chat confirmation. Fast, and the normal reflex when the project is late and the supplier contact is on the other end of a messaging app. It also leaves the thinnest record. A chat log does not travel with the order, is not visible to whoever performs incoming inspection, and typically does not survive a change of contact on either side. Worst of all, it is a reply without the question attached: six weeks later nobody can reconstruct what exactly was agreed.

Email reply. The most common form, and adequate when it is specific. The qualifier carries the weight. An email that says "points 1 to 6 agreed" is almost worthless as a record, because it does not contain the content of points 1 to 6 — it points at an attachment that may or may not be findable later. An email that restates each decision in its own words is a usable record. The cost is a few extra minutes of typing, and it is the highest-return few minutes in the whole transaction.

Revised data package. The buyer makes the change in the source design, re-exports, and resubmits with a new revision identifier. Slower, and the only form that keeps your design files and the physical board in agreement. This is the right response to any item in the substitution or re-solving categories, because those items change the design, and a design change that lives only in an email thread is a design that has two versions.

Specification or drawing addendum. A short signed document that records the decisions, references the EQ items, and attaches to the controlling specification. The appropriate form when the decision changes acceptance criteria rather than geometry — a tolerance that was agreed differently, a cosmetic allowance, a trade-off where the chosen side needs to be recorded so inspection does not reject to the original drawing.

The practical rule follows from asking who will need this record and when. Incoming inspection is performed by someone who was not in the EQ exchange, reading the drawing and the specification. If the decision is not in those two documents, inspection will apply the original requirement, and the boards will fail against a requirement you yourself agreed to waive. A decision that changes the acceptance basis has to end up in the document inspection reads. Everything else can live in email.

Whose clock is running

Lead time is quoted in days but almost never dated from an event. "15 days" from what — the quote, the PO, data approval, EQ closure? Each of those is a different calendar date, and on an HDI build with a couple of EQ rounds they can be a week or more apart.

Suppliers generally intend lead time to run from the point their engineering has everything needed to release the job to the floor. That is a defensible position: they cannot build while a question is open. Buyers generally assume it runs from the PO, because that is the event they performed. Both assumptions are held sincerely, neither is written down, and the gap surfaces as a schedule miss that both sides believe the other caused.

Three things are worth defining in the PO, and none of them require negotiating the lead time itself:

The start event. Name it explicitly. If lead time runs from EQ closure, say so — this is usually acceptable to a buyer who knows it in advance and plans around it, and it removes the argument entirely.

Who holds the clock during each round. The exchange has two sides. Time spent waiting for a buyer answer should sit with the buyer; time spent by the fabricator preparing the query, or re-querying something already answered, should sit with the fabricator. Without this split, every delay is attributed to whoever has less patience.

When the query has to arrive. This is the asymmetry nobody fixes. If the EQ arrives late in the quoted period, the buyer has lost schedule they could have protected by answering sooner. Requiring that queries be raised within a defined window after data receipt — whatever window the supplier is comfortable committing to — converts an open-ended risk into a bounded one. A supplier unwilling to commit to any window at all has told you something useful about how your job is queued.

A related and smaller point: re-queries. If an item is raised, answered, and then raised again in a later round because the first answer was not carried forward internally, that round is not a buyer delay. Saying so in the PO costs nothing and prevents a specific, common, and genuinely unfair attribution.

Moving the whole problem earlier

Everything above is remediation. The structural fix is to answer the predictable queries before they are asked, in the RFQ itself.

Most EQ items on a given design are foreseeable, and the evidence is that they repeat: the same board, quoted at three suppliers, tends to generate overlapping question sets, because the questions arise from the data rather than from the supplier. That makes them a property of your package, and properties of your package can be fixed once.

Practical moves, in rough order of return:

• State document precedence. When drawing, fabrication notes and Gerber disagree, which wins? One line in the fabrication notes eliminates the entire conflict category, which is the category that cannot be delegated and therefore costs the most round trips.
• Mark interface dimensions as such. Any dimension driven by a mating part, connector or enclosure should carry a note saying it is interface-critical and not available for rounding. This converts a query into a non-question.
• State impedance intent, not just the number. Which nets, what tolerance, verified by what method, and whether geometry may be adjusted to hold the target. Re-solving queries exist because the number alone does not say what you would trade to keep it.
• Declare your assembly process. Fine-pitch, bottom-terminated parts, double-sided reflow, selective solder. Trade-off queries about mask and pad geometry are answerable by the fabricator alone once they know what the board has to survive.
• Name the responder. One person with authority to answer EQ items, plus a named backup. A query routed to a general mailbox during a holiday week is a schedule risk created entirely by an org chart.
• Ask for the supplier's EQ pattern during quoting. A supplier who can describe the queries they typically raise on HDI builds is describing their process boundaries, which is more informative than a capability table and costs nothing to ask for.

The theme is consistent: the engineering query exists to extract information the fabricator needs and your data package did not contain. Every item you supply in advance is one round trip that does not happen, and one decision that does not get made by someone who cannot see your product.

Related reading

• HDI PCB manufacturing process — the process sequence the query is protecting.
• PCB manufacturing process guide — where EQ closure sits relative to the first irreversible operation.
• PCB fabrication specification sheet and binding quotes — writing the specification that an EQ answer has to update.
• PCB fabrication notes conflict checklist — removing the conflict category before it becomes a query.
• Incoming acceptance of unverifiable HDI attributes — what inspection can and cannot confirm about an agreed change.
• HDI PCB capabilities and request a quote.

FAQ

Is receiving an engineering query a bad sign about the supplier?

Generally the opposite. A query means someone compared your files against actual process limits before committing material. Suppliers who return no questions on a dense HDI build have either checked and found nothing — possible, if the package is complete — or have not checked, in which case the questions will arrive later as nonconformances. The query is the cheapest point in the whole order to discover a mismatch.

Can I just answer "use your standard process" to everything?

You can, and for the rounding and process-addition categories it is usually fine. For substitution, re-solving and trade-off items it transfers the decision without any record of what was decided, which means at incoming inspection the board is judged against a drawing it was not built to. The minimum viable version is to answer those items specifically and delegate the rest explicitly.

Does the EQ exchange change the price?

It can, and the PO should say what happens when it does. Substitutions, added process borders and changed array formats all touch cost. Agree in advance whether price is fixed at quote regardless of EQ outcome, or whether EQ-driven changes are repriced — and if repriced, whether you get the chance to decline. Discovering this after the boards ship is the expensive version.

Who should sign off on EQ answers internally?

Whoever owns the information the answer depends on, which is frequently not purchasing. Interface dimensions belong to mechanical. Impedance and layer assignment belong to electrical. Document conflicts belong to whoever owns the design release. Purchasing owns the transmission, the record and the clock. The common failure is purchasing answering a technical item to protect the schedule.

What if the supplier builds to something we never explicitly approved?

That is a different problem from the one in this article, and the response depends on whether the change appears in any record either side holds. It is also the reason the reporting obligation matters in pre-authorisation: an unreported change is indistinguishable from an unauthorised one at receipt. Both sides are then arguing about something neither can evidence.

Should EQ answers be treated as a change to the design revision?

For substitution and re-solving items, yes — those change the design, and keeping the revision in step is what prevents the second build from reverting. For rounding, process additions and most trade-offs, an addendum attached to the specification is sufficient. The test is whether a different supplier, given only your released files, would build the same board.