Design for Manufacturing: 10 mistakes that kill IoT product yields

Your prototype works perfectly. You order 5,000 units. 30% fail QC. Production timeline slips by 4 months. Sounds familiar? These are the 10 most common DfM mistakes we see — and how to avoid them.

1. Designing PCBs only for hand assembly

The prototype is hand-soldered with 0402 components and 0.4mm pitch BGAs. The production line uses automated pick-and-place. Problem: your design has no fiducial markers, the panel doesn't have v-cuts, components are placed too close to the edge for stencil printing.

Fix: involve a manufacturing engineer in the schematic capture phase. Add fiducials, courtyards, manufacturing test points from version 0.

2. Specifying components that have one supplier

That special MEMS sensor you chose is brilliant. But there is exactly one manufacturer worldwide, with a 26-week lead time, and they're discontinuing it next year.

Fix: for every critical component, identify at least one drop-in alternative. Run the BOM through a parts-management tool (SiliconExpert, Octopart) to flag end-of-life and single-source risks.

3. Ignoring panel optimisation

Your PCB is 95×95mm. The fab uses 300×400mm panels. You'd fit 8 boards per panel — but with a 2mm rotation you'd fit 9. Over 10,000 units that's 1,250 boards saved.

Fix: design with panel optimisation in mind. Talk to your fab about their preferred panel sizes.

4. No firmware programming station strategy

Each board is hand-programmed via USB during assembly. At 30 seconds per board, 10,000 units = 83 hours just for programming. Add JTAG troubleshooting to that.

Fix: design for parallel programming. Use programming pogo-pin beds. Pre-flash chips before SMT. Or use over-the-air bootloaders at first power-on.

5. No automated test fixture (ATE)

The factory tests each board with a multimeter and a Python script on a laptop. Tests are inconsistent, take 5 minutes per board, and fail in undetectable ways.

Fix: design and build a dedicated ATE fixture early. It pays for itself in the first batch. Define pass/fail criteria explicitly with measurable thresholds.

6. Mechanical tolerances too tight

The enclosure design assumes ±0.05mm tolerance on every dimension. Real injection moulding gives ±0.15mm at best, and changes with temperature. Result: 20% of cases don't close properly.

Fix: design with realistic tolerances. Use clearance fits and snap-fit guides. Consult your mould-maker before finalising the CAD.

7. Forgetting EMC/EMI

Your prototype passes FCC pre-screening at the lab. Production units fail. The difference: long flying leads on the prototype that act as antennas were replaced by short PCB traces in production.

Fix: EMC engineering reviews at every revision. Pre-comply early. Have a Plan B (ferrites, shielding cans) ready in the design.

8. No incoming quality control (IQC) plan

You assume your component suppliers send qualified parts. They don't always. A bad batch of capacitors causes 15% of your boards to fail in customer hands.

Fix: establish IQC criteria for critical components. Sample testing on every shipment. Establish supplier scorecards.

9. No serialisation strategy

Each board needs a unique ID for cloud provisioning, traceability, warranty. You discover this 2 weeks before pilot production. Improvised label printing leads to duplicates, illegible labels, traceability failures.

Fix: design serialisation into firmware. Use UID stamping, QR codes, RFID. Plan barcode scanning at every assembly station.

10. Treating the contract manufacturer as a black box

You ship Gerbers and BOM, expect 5,000 working units 8 weeks later. Reality: dozens of small DfM questions, component substitutions, lead-time issues need active answers. Without engineering support, the CM makes guesses.

Fix: assign dedicated engineering support to the CM. Daily standups during NPI. Have an engineer onsite for the first build.