PCB Prototyping vs Production PCB

Sending the prototype files to a production fabricator without rewriting the notes does not promote the board. It asks a factory to repeat choices the prototype shop was free to make on the day. Every figure below is hypothetical teaching maths. None of it is a fabricator quote or a BrahmWorks result.

A fast prototype order often says "FR-4", a copper weight, a soldermask colour, and a finish, and leaves the rest to the shop's standard stack. That is acceptable when the question is "does this circuit start?" or "does this outline fit the printed shell?". It is not acceptable when the question is edge rate, a controlled-impedance pair, repeatable analogue noise, or a via structure the next shop must match.

Write, on the prototype itself, which conclusions are forbidden. If the stack was whatever the shop had, you did not learn the impedance. If the finish was a default rather than the finish you will ship, you did not learn how this paste and this finish solder. If electrical test was not bought, an open net can sit in the sample you treat as a golden unit. The hardware product development process allows a learning build to fake process details only when the record says so. A bare board with no such record is not a quiet success. It is an unread condition.

Name the material at least well enough that a second shop can buy it: resin system in ordinary language if you are not ready for a full callout, glass style if the impedance or the via aspect ratio needs it, and a glass-transition temperature if the assembly profile or the product's heat requires it. Name copper weight on outer and inner layers. Name the stack, with dielectric thicknesses, if any net is impedance-controlled. If no net is, say so, and do not pay for control you will not check.

Name the finish for the parts you actually place. A flat finish and a levelled finish behave differently under fine-pitch parts and under poorly printed paste. Name soldermask and legend only as far as they affect assembly or the user's reading of a label. A colour is rarely the expensive decision. An underspecified thickness or a mask on a pad you meant to solder is.

Name how the board is delivered. A single prototype outline, scored or mouse-bitten in a way the shop invented, is not a panel. Production wants rails or an array, fiducials the assembler agrees, tooling holes if the stencil and the pick-and-place need them, and a rule for how boards leave the array without tearing a trace. Put the revision on the copper or the legend in a place the assembler can read after placement. A filename is not a revision once the file has been copied.

Electrical test of the bare board catches opens and shorts against the netlist you sent. It does not catch a wrong schematic. Flying-probe test and a fixture are different costs and different volumes; choose knowingly, and do not describe an untested lot as tested. Impedance, if you specified it, needs a coupon or a method and a record. A note that says "controlled impedance" with no target and no measurement is decoration.

The first production article is a comparison of the board you received with the board you specified: stack, finish, thickness, hole sizes that matter, mask openings, and the revision mark. Keep that comparison. Later, when a radio or a regulator behaves differently, you will need to know whether the bare board changed. Assembly defects belong to assembly. Do not use a bare-board test to explain a tombstoned part, and do not use a functional fail to excuse an untested netlist.

Invented product: a small gateway with one differential pair that the designer intends to run quickly, a microcontroller, and a connector along one edge. Revision P is the prototype. Revision R is the first board the team wants a production shop to repeat. The schematic is unchanged. The notes are not.

Revision P, teaching note: "Four layers, FR-4, 1.6 mm, green mask, lead-free levelled finish, any stack." Ten boards, no electrical test, no impedance coupon. The question P may answer: does the circuit enumerate and pass traffic on the bench, with this silicon? The question it must not answer: will every later shop build the same edge behaviour on that pair?

Revision R, teaching note: a named four-layer stack with dielectric thicknesses; the differential pair called out with a target impedance and a tolerance; a flat finish because the teaching fine-pitch part on this board is the reason; electrical test included; an impedance coupon on the panel; an array with fiducials and a revision in copper. The question R may answer: can this shop repeat that specification? It still does not answer whether the product meets a regulatory limit. That is a different test, on a built unit.

Invented prices:

• Prototype lot, revision P: 10 boards for INR 12,000. No electrical test, stack not named.
• Production, revision R: a one-time charge of INR 15,000, then INR 7,000 for a panel of 10 boards, electrical test and a coupon included.

Apparent prototype unit: 12,000 / 10 = INR 1,200.

Cost of 100 boards of revision R: 15,000 + (10 × 7,000) = 15,000 + 70,000 = INR 85,000. Unit: 85,000 / 100 = INR 850.

The INR 850 is not "cheaper than" the INR 1,200 until you say they buy different things. The prototype unit did not include a named stack, a test, or a coupon. The production unit did, and it spread a one-time charge over 100 boards. Buy only one production panel and the sum is 15,000 + 7,000 = INR 22,000 for 10 boards, which is INR 2,200 each, dearer than the prototype and still a different object. Comparing INR 1,200 with INR 850 as if one shop had become more efficient is the mistake. At a teaching quantity of one panel they have not even become cheaper.

A stencil is not in either figure. Put a teaching stencil at INR 3,500 beside the conversation so nobody believes the bare-board unit is the assembled-board unit. 3,500 / 100 = INR 35 if you spread it over the same 100, and that INR 35 is assembly tooling, not PCB fabrication. Keep the buckets apart.

• The question this revision is allowed to answer is written, and so is the question it must not.
• Material, copper weights and thickness are named well enough for a second shop.
• Either the stack and impedance targets are specified with a measurement, or impedance is explicitly not required.
• Finish is chosen for the parts, not copied from the prototype default.
• Electrical test is either bought or recorded as not done.
• Array, fiducials and break-out method are agreed with the assembler.
• Revision is readable on the board after assembly.
• The first article is compared with the specification and the comparison is filed.
• Prototype conclusions that the stack did not support are listed as open.
• Bare-board cost is not presented as assembled-board cost.

You can send the plots. You should not send the notes unchanged if the prototype left stack, test, finish or panel to a shop default. Ask the production fabricator to build a specification, not a memory of another shop's week. Where the prototype was silent, the production file has to speak or explicitly waive.

When a net's edge rate and length make the interconnect part of the circuit, or when a standard you have chosen to follow sets a target. A slow pair across a short board may not need a controlled stack. Write the decision. Do not buy coupons "to be safe" and then ignore them, and do not skip them on a pair you are about to blame the silicon for.

Everything the prototype shop was allowed to invent: stack, material detail, test, finish if it was a default, array, fiducials, and the revision mark. The schematic changes only if the prototype answered its question by showing a circuit error. A production note that differs in silence, not in content, has not been written yet.

Only in the small sense that legend contrast, cleanliness and what the user must read can matter. It is almost never the cost driver next to layers, area, vias and test. Changing colour between prototype and production does not promote the board. Changing the unnamed stack does.

Bring the prototype note, the list of conclusions that note cannot support, and the production note you propose to send. A review can say whether revision R is a specification or still a prototype with a new filename.

Review the next stage with BrahmWorks