Capability / Test engineering
A test strategy is an economic decision before it is an electrical one. The board that leaves the line has already been through 100% AOI and X-Ray on every hidden joint, so the question the test engineer answers is narrower and harder: which defects can still escape, and what is it worth to close each one.
Coverage matrix
Thirteen methods sit between the printer and the shipping carton. Each row states the question the method answers, the defect class it is blind to, and the cost and lead time it carries, so the strategy can be costed before it is chosen.
| Test type | What it proves | What it cannot see | Typical cost | Lead time to first run |
|---|---|---|---|---|
| Solder paste inspection | Deposited paste volume and alignment against the aperture, before any component is placed | Placement errors, and anything that happens during reflow | $0.10-0.30 per board | Inline, no added lead time |
| Pre-reflow optical | Component presence, orientation and value marking before the joint exists | Joint quality, because there is no joint yet | $0.15-0.35 per board | Inline, no added lead time |
| Post-reflow optical | Tombstoning, bridging, missing parts, wrong polarity and insufficient solder fillets | Anything under a BGA, a shielded can or a connector body | $0.20-0.50 per board | Inline, no added lead time |
| Two-dimensional X-Ray | Ball presence, bridging under a package, barrel fill on plated holes | Void depth, head-in-pillow, and electrical function | $1-5 per board | Inline, recipe setup 2-4 h |
| Three-dimensional X-Ray | Void percentage per ball and head-in-pillow through the package | Electrical function, and marginal wetting that is still contiguous | $5-20 per board | Inline, CT recipe 4-8 h |
| Bench visual inspection | Workmanship, labelling, mechanical fit and packaging | Features below 0.4 mm leaded or 0.35 mm leadless pitch | $0.10-0.25 per board | Inline, no added lead time |
| UV coating inspection | Coating coverage, keep-out compliance and edge definition | Adhesion, thickness uniformity and cure state under a shadow | $0.20-0.40 per board | Inline after the cure pass |
| In-circuit test | Continuity, shorts, wrong value, missing part, polarity and boundary-scan pin opens | Marginal joints that conduct under probe pressure, RF behaviour and drift | $0.50-2.00 per board after the fixture | 4-8 weeks for the fixture |
| Flying probe | The same electrical facts as in-circuit test, at any volume, without a fixture | High node counts economically, and joints that fail only under mechanical stress | $3-10 per board | Hours of programming, no fixture |
| Boundary scan | Interconnect continuity and IC pin opens through the scan chain | Nets outside the chain and every analogue section | $0.10-0.40 per board | 1-2 weeks for chain validation |
| Functional test, board level | Behaviour against written limits at one defined operating point | Marginal timing, temperature extremes and long-term drift | $1-10 per board | 1-2 weeks for the fixture and program |
| Functional test, system level | Whole-product behaviour with firmware, interfaces and enclosure | Wear-out mechanisms and any parameter that moves over months | $5-25 per board | 2-4 weeks including the harness |
| Thermal cycling and stress screen | Intermittent joints and thermal margin revealed by cycling | Random defects at low volume, and anything not stressed by the profile | $2-15 per board | 4-8 weeks to define the profile |
How the strategy is built
Sequence is chosen by cost per defect caught, not by prestige. Solder paste inspection costs cents and catches a printing fault before a hundred hidden joints are made from it. Optical inspection follows at under half a dollar and removes most visible faults before any fixture is loaded. Electrical test is the expensive gate, so it should only ever be asked to catch what the cheap gates physically cannot, which is why the coverage matrix names the blind spot of every method rather than its strengths.
A 1.0 mm test pad on a 2.54 mm grid costs nothing at layout and a revision later. We review every design for probe access on both sides, for a 3 mm clearance ring around tall components, and for via tenting that would stop a probe reaching the net. Without that review the fixture becomes a mechanical project: a cradle, a mating connector set and a cable tree that together can cost more than the probes they replace. Tooling holes and panel fiducials are checked in the same pass.
Every functional test ships with a limit table: the parameter, the nominal value, the lower and upper limit, the test condition and the instrument. A limit of plus or minus ten percent with no condition attached is not a test, it is a discussion. Where a limit cannot be met at the first article, the response is a written deviation against the named parameter rather than a widened band, because a widened band silently changes what the programme is buying.
Functional test log
Seven measured parameters from a motor-control programme, each against its own lower and upper limit at a stated condition. The log is attached to the serial number, so a field question can be answered from the record rather than from a re-test.
programme MTR-2251 - 4-layer motor controller, 24 V rail, 6 A load condition 25 C ambient, 24.0 V input, firmware 3.1.4, 6 A resistive load parameter limits measured result supply current, idle 18-26 mA 21.4 mA PASS supply current, 6 A 5.90-6.30 A 6.08 A PASS logic rail 3.234-3.366 V 3.301 V PASS hall offset -12 to +12 mV +4 mV PASS commutation freq 19.6-20.4 kHz 19.98 kHz PASS thermal rise, 10 min max 28 C 31 C FAIL retest heatsink seating corrected, 10 min soak repeated thermal rise, 10 min max 28 C 24 C PASS disposition rework pass 1 recorded against serial 0004187, limit unchanged coverage 97.4% of nets electrically verified, residual list attached RELEASED all limits met at the stated condition, 1 rework pass logged
Development deliverables
Test development is quoted as a fixed scope with three artefacts, so nothing about coverage or cost is left to a meeting after the first lot has been built.
TST-01 / Fixture
Probe plate layout, probe selection for the pad geometry, guide-pin and tooling-hole placement, and a mechanical cradle where the board has no probe access at all. The review runs at layout, so the changes that make a fixture cheap are made before the artwork is released rather than after.
TST-02 / Program
The program that drives the board, the limit table that decides pass or fail, and the boundary-scan chain where the design carries one. Every limit names its condition and its instrument, and the program writes a serialised result file rather than a red or green lamp, so a dispute can be settled from data.
TST-03 / Report
A per-net and per-defect accounting of what is verified, what is verified indirectly, and what is not verified at all. The residual list is the commercial part of the document: it names the defects the programme is choosing to carry, so the decision is made deliberately instead of being discovered from a field return.
Method economics
The three electrical methods differ less in what they measure than in how the cost is shaped. The table below is the comparison we put in front of a buyer when a volume band is still moving.
| Comparison | Flying probe | In-circuit test | Functional test |
|---|---|---|---|
| Fixture or setup | None; probe program only | Bed-of-nails fixture with a probe plate | Interface harness or cradle with a mating connector set |
| Setup cost | $0 NRE beyond programming hours | $2,000-50,000 depending on node count and side access | $1,000-10,000 for the fixture and program |
| Lead time to first run | Hours | 4-8 weeks | 1-2 weeks |
| Cost per board | $3-10 | $0.50-2.00 after the fixture is amortised | $1-10 board level, $5-25 system level |
| Cycle time | 1-10 minutes | 5-30 seconds | 30 seconds to 5 minutes |
| Volume band where it wins | 1-1,000 boards, and any high-mix low-volume programme | Above roughly 3,000 boards, and 10,000 plus without argument | Every band, because nothing else confirms behaviour |
| What it leaves open | Joint integrity under vibration, and RF behaviour above the probe bandwidth | Marginal joints that pass under probe pressure, and any parameter that drifts with temperature | Marginal timing and wear-out, which only cycling or field data reveal |
The volume figures above are planning bands, not a price list. A programme that mixes high node count with low volume is usually priced on flying probe even at 5,000 boards, because the fixture cannot be amortised across enough revisions.
Macro record
A loaded fixture, an instrument rack and a serialised result file. The physical arrangement matters because probe pressure, cable routing and grounding all change the measurement the limit table was written against.
Buyer questions
The crossover is normally between 1,000 and 3,000 boards, and it is arithmetic rather than opinion. A bed-of-nails fixture costs 2,000 to 50,000 dollars with a 4 to 8 week lead time, and in-circuit test then runs at roughly 0.50 to 2.00 dollars per board in 5 to 30 seconds. Flying probe needs no fixture, so it starts at zero NRE but costs about 3 to 10 dollars per board and takes 1 to 10 minutes. Below roughly 1,000 boards a year, flying probe plus functional test is almost always cheaper; above 10,000 boards the fixture is paid back inside the first production lot.
It can, but the cost moves from the fixture to the harness. Without probe access, the fixture has to reach the board through its own connectors, which means mating connectors, a cable tree and a mechanical cradle, and the unit-under-test socket often costs more than the probes it replaces. Functional test also depends on firmware that can drive the board into a known state and report a result, so the useful moment to raise test access is the layout review, where a 1.0 mm pad on a 2.54 mm grid costs nothing to add.
By escape risk and volume, in that order. The coverage report lists every defect class the programme can plausibly produce, the method that catches it, and what is left over after that method has run at its published limits. Cost and test time are then applied to the residual list, not to the whole board. A 200-unit instrumentation build and a 50,000-unit consumer build do not get the same strategy, and the report says so in writing rather than leaving it to habit.
Seven fields. A process engineer replies within one working day.
Seven fields. A process engineer replies within one working day.
Seven fields. A process engineer replies within one working day.