How to Design a PCB for Mass Production

Mass production, in this article, means the fabricator and the assembler can repeat the board from the files. No one chooses a special part for each unit, and no joint is repaired as a habit before the board is called good. It does not mean a particular quantity. The pilot, the yield and the contents of a release package are a separate subject, set out in How to Take a Hardware Prototype to Production. What follows is the board design that makes such a package buildable. Every dimension and every rupee is hypothetical teaching maths. None of it is a fabricator's rule, a quotation, or a BrahmWorks result. If your supplier's capability sheet disagrees, the sheet wins.

The footprint belongs to the manufacturer part number on the bill of materials, including that part's recommended pad and its tolerance. It does not belong to a similar part that happened to be on the prototype. A module with a different centre pad, or a connector whose latch is mirrored, is a new board even when the courtyard looks familiar. If you allow an alternative, its land pattern has to match, or it is not an alternative. "Electrically similar" is not a footprint.

Polarised parts should share an orientation the machine and the inspector can learn. A diode rotated to make the schematic look tidy, beside another diode rotated the other way, is a placement defect you have drawn on purpose. Mark pin 1 in a way that still exists after the body is fitted. Silkscreen that disappears under the component tells the person who is no longer looking, and does not tell the camera or the operator who is.

Leave a courtyard the nozzle can enter, and room for a neighbour's fillet. A layout that is dense in the viewer and impossible to rework will still be reworked when the first lot fails, with heat and with damage to pads. Density you cannot inspect is not a cost reduction. It is a decision to find faults later, one board at a time, with a iron you claimed to have left behind.

The fabricator builds a panel, not the single outline in your CAD. Rails, fiducials, tooling holes and the way boards are separated are part of the design, even when a process engineer draws the panel itself. If you leave all of that to "whoever panels it", you still own the connectors they crack and the traces they run through a score.

V-score and routed tabs stress the board in different places. Parts and copper need to stay off a score, and a heavy part next to a tab is a cracked joint after the board is pushed out. Choose the method you are designing for, and put the keep-out in the file. Changing the method after placement is frozen moves that keep-out into components you can no longer shift.

The stencil sets how much paste each pad gets. A large thermal pad under a QFN or a module is not a small signal pad scaled up. One unsplit opening can leave too much paste, float the part, or lose paste down open vias in the pad. Tenting those vias, plugging them, moving them out of the pad, or splitting the opening are options to agree with the assembler. They are not a single recipe this article can hand you, because the paste, the thickness and the oven are theirs. The prototype that was soldered with an iron never asked the question. The iron and the technician were the process, and they will not be on the line.

A production test needs pads a fixture can hit, on a spacing that fixture can hold, on a side it can reach. Pads under a shield, under a cell, or on a via that has been tented shut are not test points, however clearly they are labelled in the schematic. Programming is the same constraint. If the only way to load firmware is a header you must fit and then remove, or a connector the housing covers before the test, then the assembled product is not the thing you programmed. Decide where access lives, and whether the header is fitted, deleted, or replaced by pads. Write that decision on the assembly drawing so the line does not invent a fourth option.

Inspection comes before a functional test can mean anything. Silkscreen over a pad hides the joint and contaminates it. A hidden pad under a thermal land has no fillet to look at. X-ray or a boundary test may be the right method. It should be chosen, and it will appear on the quote. Discovering it when the quote arrives means the layout has already spent the cost.

The teaching board is a four-layer industrial controller. The first five units were built by hand. The processor is a QFN with a central pad. A technician tinned the pad, placed the part and heated it until it sat down. All five ran. Nothing about that result is false. It is also not a result about reflow.

The assembler uses paste and an oven. The central pad has six open vias into a plane. Paste moves down the vias. What remains is uneven. On the first panel, one board shows a void under the pad and an open corner pin. The hand-built units could not have warned you. They were a different joining process, and their success has no bearing on the stencil.

The change, which has to be agreed with that assembler rather than asserted from a blog, is the via treatment and the opening in the stencil. An instruction that says "inspect the QFN" does not put paste back on the pad. Until the via and the stencil match the process you are paying for, the open pin will keep being recorded as a placement error. Placement may be perfect. The solder left.

Invented geometry, so the counting method is visible. This is not a panel offer and not a price.

• Board outline: 62 mm by 38 mm.
• Gap between boards: 2 mm.
• Usable window, rails already removed: 240 mm by 160 mm.
• A panel price used only so a division has an input: INR 4,800, assumed per panel and not per area.

Along the 240 mm side, three boards need 3 × 62 + 2 × 2 = 186 + 4 = 190 mm, which fits. Four boards need 4 × 62 + 3 × 2 = 248 + 6 = 254 mm, which does not. Along the 160 mm side, four boards need 4 × 38 + 3 × 2 = 152 + 6 = 158 mm, which fits. Five would not. The count is 3 × 4 = 12.

Area of those boards: 62 × 38 = 2,356 mm², and 12 × 2,356 = 28,272 mm². The window is 240 × 160 = 38,400 mm². The boards use 28,272 ÷ 38,400 = 0.736 of it, which is 73.6 percent. The rest is gap and the unused strip. It is not automatically waste you can delete. Some of it is the gap the router needs.

Turn the board to 38 mm by 62 mm. Six across the 240 mm side need 6 × 38 + 5 × 2 = 228 + 10 = 238 mm, which fits. Two along the other side need 2 × 62 + 2 = 126 mm, which fits, and three would need 3 × 62 + 2 × 2 = 190 mm, which is more than 160 mm. The count is 6 × 2 = 12. Rotation, on these numbers, buys nothing.

Now suppose 4 mm of the long side is not doing any mechanical or electrical work, so the outline may become 58 mm by 38 mm, and the enclosure and the connectors still agree. Four across: 4 × 58 + 3 × 2 = 232 + 6 = 238 mm. Four along the short side stay at 158 mm, as before. The count is 16.

Area: 58 × 38 = 2,204 mm², and 16 × 2,204 = 35,264 mm². As a share of the window, 35,264 ÷ 38,400 = 0.918, or 91.8 percent.

If, and only if, the fabricator really charges INR 4,800 for the panel regardless of how many boards are on it, the bare-board share is 4,800 ÷ 12 = INR 400, or 4,800 ÷ 16 = INR 300. The difference is INR 100 a board. Across an invented 1,000 boards that is INR 1,00,000. The difference disappears if the price is by area, if the smaller outline hits a connector or a fixture pad, or if assembly and test cost more than the bare board you just saved. Panel arithmetic is a reason to ask a mechanical question. It is not permission to shrink an outline that the housing still needs. Repeat the count with the fabricator's real window, their real method of separation, and a price that says what it includes.

• Every land pattern matches the part number you will buy, and any alternative truly fits it.
• Polarity and pin 1 are consistent, and they remain visible after the part is placed.
• The method of separating boards is chosen, and parts and copper stay clear of the score or the tab.
• Fiducials and tooling features match the panel you expect to build.
• Large pads have a via plan and a stencil opening agreed with the assembler.
• Test pads and programming access match the fixture, and they are still reachable in the housing.
• Silkscreen is not printed on pads.
• Joints you cannot see have an inspection method, and that method is on the quote.
• The outline has been checked against the enclosure, not only against the panel window.
• The files carry a revision the assembler can write on the build record.

Only when the prototype was built by the same kind of process. A hand-soldered board can carry footprints, open vias and connectors that a line will not accept. Moving from an iron to an oven is a design change even when the schematic is untouched. Keeping the Gerbers because the five units ran keeps the iron in the file. Review the stencil, the panel and the test access as their own release, with their own faults found on a panel, not on a favourite unit.

Enough to separate a bad joint or a wrong part from a bad design, for the faults you are willing to catch on each unit. Supplies, the programming connection, and the nets that fail in ways a whole-board functional test cannot isolate are the usual set. A fixture that hits every net can cost more than the escapes it would catch, and it can force an outline the product does not have room for. Choose from the faults you have seen and the faults that are unsafe to ship, not from a desire to touch every net once.

Draw the constraints: the outline, the keep-out from each edge, where a score is forbidden, and which side the fixture uses. Let the fabricator propose the panel if they will stand behind separation. Then review what they propose against those constraints. A panel you draw in isolation, for a sheet size they do not use, is how a careful 12-up becomes an 8-up after the order is placed, with no one going back to the outline that caused it.

The bill of materials with manufacturer part numbers, the assembly drawing, the notes for paste and solder mask, the separation method, the test and programming method, and the revision. A zip of copper layers asks the assembler to guess polarity, finish and which pads exist only for test. Those guesses become the product you receive, and they will not match the product you simulated. The extra documents are the design. The Gerbers are one view of it.

On UltraFast EV charging hardware, BrahmWorks reviews the enclosure and the electronics together, including how cables are routed and how components are reached. A production board inherits the same rule at the scale of one outline. The edge of the board, the connector that passes through a wall, and the access a fixture needs are one design. Releasing Gerbers with no housing section, or a housing with no idea of the fixture, leaves that clash for a pilot to discover. No yield, cost, time or test result from that charging work is stated here.

Bring the fabrication outputs, the bill of materials and the solder process you intend to use. BrahmWorks can mark the gaps a hand-built set of boards will not show, while changing them is still a layout rather than a scrapped panel.

Review the next stage with BrahmWorks