DFM: The Decisions That Don't Show Up on Drawings

Most teams treat Design for Manufacturing as a checklist — a review that happens once, late, against drawings that are mostly done. Hole spacing. Fillet radii. Draft angles. Minimum wall thickness. The checklist matters. It is not the part of DFM that decides whether first production hits yield.

The DFM decisions that matter most do not show up on any drawing. They are decisions about how the product will be built — tolerance philosophy, second sources, test fixtures, assembly sequence, calibration and provisioning, serviceability. Each is a cost-of-precision tradeoff. Each is usually under-specified. Each, if missed, shows up as a yield problem at PVT — where it costs ten times what it would have cost to decide at EVT.

This article names six of those decisions. It is a supporting piece to our pillar on From Prototype to Production.

DFM is a set of decisions, not a checklist

A drawing-level DFM review answers a narrow question: can this geometry be made? It is a buildability check. The broader question — can this product be built at the cost, volume, yield, and variance the business depends on — is never answered on a drawing. It is answered in upstream calls that most programs make implicitly, under schedule pressure, without writing anything down.

The checklist items — draft angles, hole-to-edge spacing, panelization efficiency, solder mask relief — are the output of a good process, not the process itself. A team that gets the six decisions below right passes a checklist review almost automatically. A team that only runs the checklist passes it and still misses yield.

Decision 1 — Tolerance philosophy

Every tolerance on every drawing is a cost decision. A shaft at ±0.05 mm costs real money to hold across a run. The same shaft at ±0.1 mm may cost a third as much, cycle faster, and open the supplier base by a factor of five. On most drawings we review, a third to a half of the tight tolerances are there because the engineer did not want to think about it — not because the function demands it.

The discipline is to classify features before the drawing is released. Functional tolerances — fit, motion, sealing — are tight and explicit, with a stack analysis behind them. Non-functional tolerances are loose on purpose, with a general-tolerance note a line inspector can use. A good mechanical drawing has five to ten tight callouts and the rest on general tolerance. A bad drawing has fifty tight callouts and a CM quoting at double the budget.

The same logic applies in electronics. A board that demands impedance control on every signal, 0.1% resistors across the BOM, and 3-mil trace-and-space where 5-mil works will quote expensive, yield lower, and narrow the fabricator list. Tolerance philosophy is the first DFM decision because every downstream decision is cheaper when the tolerances are honest.

Decision 2 — Second sources by design

“Qualified second source” is a phrase most BOM reviews contain and most programs cannot defend when the primary supplier misses a delivery. The failure mode: the second source was picked from a datasheet, not designed into the product. The pin assignment is subtly different. The thermal envelope does not match. The package is a QFN where the original was a QFP. The firmware assumes the primary’s boot sequence. The mechanical cutout is 0.3 mm too tight for the alternate connector.

Designing for a second source means architectural choices that keep alternates electrically, mechanically, and firmware-wise interchangeable. A common footprint that accepts two or three compatible parts. Pull-ups laid out for both. Firmware that detects the variant at boot. Mechanical cavities that clear the larger package envelope. The cost is a few hours of extra layout. The benefit is that when the primary goes on allocation, the line does not stop.

For every critical part, write down the second source, the validation status, and the firmware-or-mechanical delta. If you cannot write that down, you do not have a second source. You have a datasheet match.

Decision 3 — Line-operable test fixtures

End-of-line test is not a pass/fail check. It is the data-collection instrument that decides whether the line is in control. A fixture designed and operated by the engineer who built the prototype almost always fails at scale — not because the test logic is wrong, but because nobody else can operate it.

A line-operable fixture uses industrial connectors — no bench-grade BNCs or bare-wire jumpers. It presents a pass/fail result a line operator can act on without interpretation. It writes every result to a traceable record, keyed to the unit serial. It covers the defects that actually occur in volume. It runs at line speed — a thirty-second test on a ten-second takt is a bottleneck, not a test.

On AimRobotics — robotic dispensing tools running at Lockheed Martin, Aston Martin, BMW, and tier-1 electronics manufacturers across thirty-plus countries — the test fixtures were designed alongside the products, not after. Every unit ships with a test record validating motor performance, sensor calibration, communication integrity, and safety interlock before it leaves the line. That discipline is why the products work at 3 AM on a production floor, not just on a bench.

Decision 4 — Assembly sequence economy

Assembly time is a direct unit cost. A product that takes twelve minutes to assemble costs twice what the same product costs at six — in labor, line space, and throughput. Assembly sequence is a design decision.

Economical assembly has recognizable shape. Fasteners standardized to three sizes, not fifteen. Screw counts minimized; captive screws or self-tapping inserts cut assembly time roughly in half per joint. Sub-assemblies built in parallel on separate stations. No step requiring two operators. No step requiring a tool only one operator has.

An often-missed call is fastener direction. Top-entry screws are fast. Bottom-entry screws require flipping the unit, which doubles station time. On one program, changing eight bottom-entry fasteners to top-entry cut forty seconds off assembly — across a twenty-thousand-unit annual run, roughly a hundred and ten hours of line labor a year.

Walk the assembly sequence before tooling, with the CM in the room.

Decision 5 — Calibration and provisioning as part of the line

Every sensor-bearing product has a calibration step. Every firmware-bearing product has a provisioning step — serialization, firmware flash, bootloader lock, cryptographic identity. On prototypes, both happen on the engineer’s bench with a laptop. On a line, neither can.

The right architecture treats calibration and provisioning as station operations. End-of-line test flashes signed firmware, writes the unit serial, programs cryptographic identity, runs calibration, persists coefficients to traceable storage, and verifies — all in one pass, without engineering intervention.

This reaches back into electronics design. The MCU has to support programmable-at-production boot. A fixture-side programmer has to work without opening the enclosure. Calibration coefficients have to live in memory that survives firmware updates. Serial numbers have to come from a single source of truth. Firmware has to be signed; the signing process has to be part of the release workflow.

None of this is visible on a drawing. All of it has to be decided by DVT at the latest. A program that reaches PVT without a provisioning architecture ships its first production run by hand, with a laptop in the clean room.

Decision 6 — Serviceability from day one

Most hardware products will fail in the field. Serviceability is a DFM decision because the ability to service a unit is determined by choices made at EVT and DVT — not by a repair manual written after launch.

Can a field technician diagnose a failure without shipping the unit back? That means an accessible debug port, a status log that survives power cycle, and a self-test a non-engineer can trigger. Can a unit be opened and a failed sub-assembly replaced without destroying the housing? That is a fastener and adhesive decision. Can firmware be patched in the field without bricking the unit if the update is interrupted? That is a bootloader-and-partition decision.

The cost of skipping serviceability is paid in warranty reserves. A product that has to be returned, diagnosed, and shipped back has a warranty cost of two to five hundred euros per event in shipping and labor alone. A field-diagnosable architecture resolves most events with a firmware update or a mailed-out sub-assembly. Across ten thousand units at a 3% annual service rate, the difference is meaningful money — and a unit that comes back with a fault code, a log file, and a factory test record teaches the team something. A unit with none of that teaches nothing.

How these decisions get skipped

These decisions get skipped not because teams do not know they matter. They get skipped because each has a champion who is not in the room at the moment the decision is made, or no champion at all.

Tolerance philosophy gets skipped because the mechanical engineer owns the drawing and tight tolerances are the safer default when nobody pushes back on cost. Second sourcing gets skipped because procurement is downstream of the BOM freeze, not upstream of component selection. Test fixtures get skipped because the test engineer is hired after PVT starts, not before DVT. Assembly sequence economy gets skipped because the CM is engaged too late — after tooling is cut. Calibration and provisioning get skipped because they live in the seam between firmware and operations, and no one person owns both. Serviceability gets skipped because it is a post-launch concern in a pre-launch schedule.

The fix is organizational as much as technical. A program lead — internal or external — accountable for all six decisions, not just the ones inside their specialty, is the cheapest insurance a hardware program can buy. A good contract manufacturer will raise these decisions if invited early enough; we cover when at Choosing a Contract Manufacturer (From the Inside). The DFM gate itself is described in EVT, DVT, PVT: What Each Gate Actually Decides.

Programs that ship at yield are not the ones with the tightest drawings. They are the ones where the six decisions above were made early, deliberately, and in writing — where the drawings are the honest output of that discipline, not a substitute for it.