Two assembled boards come off the same line looking identical. One works. The other has a ceramic capacitor with a hairline crack from depaneling, a BGA ball that never wetted, or a resistor pulled from the wrong reel. Electrical test is the only step that separates them before a customer finds out, and the two methods that dominate contract assembly — in-circuit test on a bed-of-nails fixture, and flying probe — differ by roughly $15,000 in upfront cost and a factor of thirty in cycle time. Choosing between them is volume arithmetic with a crossover you can compute before the design is released, plus a set of layout decisions you have to make while the board is still editable.

What Each Method Physically Does

In-circuit test presses the board onto a fixture holding hundreds of spring-loaded pogo pins, one per accessible net. Every node is contacted at once, so the tester measures resistances, capacitances, inductances, diode drops, and shorts across the whole board in a single pass, then applies power and checks regulator rails and device signatures. Cycle time is 20 to 60 seconds for a mid-complexity board.

Flying probe uses four to eight probes on independent gantries moving from test point to test point under program control. No fixture exists; the machine reads your CAD and Gerber data and executes a probing sequence serially. That serial execution is the entire tradeoff: a board an ICT fixture clears in 30 seconds takes a flying prober 4 to 15 minutes.

Coverage Differences That Matter

Both methods find the same fault families — opens, shorts, wrong value, missing part, reversed polarity. ICT does more of the powered work because holding every node simultaneously lets it isolate a device by guarding out its neighbors; flying probe leans harder on unpowered measurements and on capacitive open detection at package lead frames.

Neither sees inside a BGA. A ball that is cracked but still conducting, or a bridge under a package, is invisible because there is nothing to touch. That gap is filled by boundary scan and by optical and X-ray inspection, which your assembler should run alongside electrical test as part of a normal DFM review for electronics.

Where the Cost Crossover Lands

A dedicated ICT fixture for a board with 300 to 600 test points runs $8,000 to $25,000 including wiring, program, and debug on real boards, and takes four to eight weeks from released Gerbers. Per-board cost after that is $0.15 to $0.60. A flying probe program costs $500 to $2,500 in engineering NRE, is ready in days, and costs $3 to $15 per board.

Run the numbers with a $15,000 fixture, $0.40 per board on ICT, and $6.00 per board flying probe. The per-unit saving is $5.60, so the fixture pays back at roughly 2,700 boards. Below that, flying probe wins outright; above roughly 5,000 boards a year ICT wins on both cost and throughput.

Two things shift the crossover. Design churn destroys fixture economics, since every revision that moves test points means a re-wired fixture — which is why fixtures get ordered at the PVT stage rather than at EVT. And board complexity pushes flying probe cycle time up faster than it pushes fixture cost up, moving the crossover toward 1,500 units.

Boundary Scan Is the Third Leg

Boundary scan, defined by IEEE 1149.1 and usually called JTAG, moves test inside the silicon. Compliant devices carry a shift register cell at each pin, so a chain of scan-capable parts drives and senses the nets between them with no probe access at all — often the only way to verify interconnects between a BGA, an FPGA, and a memory device in each other's shadows. Getting it to work is a design decision: the JTAG signals have to reach a header, the chain order has to be documented, and a non-scan device in the middle breaks the chain unless bypassed. The same pass programs flash and CPLDs, folding device programming into test.

Functional Test Does Not Replace Either One

A functional test powers the finished assembly and exercises it the way a user would. It is the test customers care about and a terrible diagnostic: when it fails you know the board is bad, not which of 800 joints caused it. The workable structure is layered — optical inspection catches placement and solder shape, ICT or flying probe names the faulty reference designator, functional test confirms system behavior. Each layer needs its own jig, so a product heading into volume needs a production test fixture plan budgeted before the first mass build.

Design for Test: Decisions Made During Layout

  • Put a test pad on every net. Aim for 90 percent or better net access; below 80 percent the value of any probe-based test collapses.
  • Size pads at 0.9 mm (35 mil) diameter with 1.27 mm (50 mil) spacing. Sub-0.6 mm pads require fine-pitch pogo pins and cut fixture life.
  • Keep all test points on one side — a double-sided fixture roughly doubles cost — and add two non-plated tooling holes for registration.
  • Do not put test points under tall components, within 3 mm of the board edge, or on top of vias.

What to Put in the Test Spec for an EMS

Assemblers quote from what you give them, and a vague test line item turns into a change order. Specify the method by build stage, the required fault coverage percentage, who owns the fixture and where it is stored, pass/fail criteria and measurement limits, what happens to a failing board, and how test data gets logged and returned. Fixture ownership matters as much as mold ownership — write it into the assembly agreement, as you would when working with a contract manufacturer on any tooling asset.

Ask for yield data by fault type from the first production lot. A board failing 4 percent at ICT on the same reference designator every time is not a test problem; it is a first article inspection finding that arrived late, usually tracing to a footprint or a stencil aperture that should have surfaced during SMT assembly planning.

Getting the Test Strategy Right Before the Fixture Is Cut

Projects House builds test strategy into board design rather than bolting it on after layout: net access targets set during placement, a boundary scan chain that works, and a test spec an assembler can quote without ambiguity. Tell us your board complexity and annual volume through the contact form and we will tell you which side of the crossover you are on.