AOI, X-ray, Flying Probe and ICT: What Each Test Actually Catches

Blog › Technical Guides › Test & Inspection

Technical Guides · Test & Inspection

AOI, X-ray, Flying Probe and ICT: What Each Test Actually Catches

Four test methods, four different kinds of blindness. Each one is a sensor with a specific blind spot — so the useful question is not “do you test?” but “what can this method not see, and what covers that?”

September 2026·11 min read·O-Leading Technical Team

Ask a supplier “do you test?” and the answer is always yes. Ask which test sees which defect, and the conversation becomes useful: a 100% optical inspection can miss every joint under a BGA; a 100% electrical test can pass a board whose plated barrel is already cracked; and a perfect X-ray image says nothing about the ionic residue still sitting between two traces.

This guide separates the four methods buyers ask about most — AOI, X-ray, flying probe and ICT — by what each one physically senses, what that lets it catch, and what stays invisible. It covers bare boards and assemblies, because the same four names mean different machines and different programmes depending on the stage.

01The Question That Picks the Method

Not “how accurate is it?” but “what is it blind to?”

Every inspection method trades one kind of visibility for another. Optical inspection needs line of sight. X-ray needs a density difference. Flying probe and ICT need a place to put a probe. Once you state the constraint, the choice usually makes itself.

02AOI: The Surface, the Fillet and the Placement

The fastest, cheapest, most programmable set of eyes in the factory — with one hard limit: it needs to see the surface.

Automated optical inspection compares what is on the board with what the CAD and the inspection programme say should be there. Because it is optical, it is excellent at everything visible and useless at everything hidden.

What AOI reliably catches

• Missing, extra, misaligned or reversed components, and wrong part numbers through marking verification
• Solder bridges between leads and pads, insufficient or excess solder, poor fillets
• Lifted leads, tombstoning and (with 3D AOI) coplanarity problems that a 2D system reads as a shadow
• Paste deposition problems when an SPI step is used ahead of placement
• On bare boards: inner-layer and outer-layer imaging defects — opens, shorts, nicks and copper residue left by etching

Where it stops

AOI cannot see a joint underneath a BGA, a CSP, a QFN or a connector body. It also cannot judge the inside of a plated barrel, and it cannot tell you whether the copper under the solder mask is the thickness you ordered. Those three limits are exactly why the next three methods exist. A typical 2D system also produces false calls on shiny or shadowed joints, so the value of the output depends on how the programme is tuned and how the operator dispositions calls.

03X-ray: Everything Under the Package

The only one of the four that looks inside a soldered joint — and the one whose images are easiest to over-read.

X-ray imaging passes radiation through the assembly and records what survives. Solder, copper and voids attenuate differently, so a BGA ball shows up as a disc with darker regions where it is hollow. That is the whole basis of the method, and also its limitation: it sees density, not correctness.

What X-ray reliably catches

• Voids inside BGA, CSP and QFN joints, and their size and position within the ball
• Bridging and shorts hidden under a package body, where no optical system can reach
• Missing, displaced or collapsed balls, and solder-ball integrity after rework
• Head-in-pillow conditions, read as a waist or discontinuity in the joint
• Through-hole barrel fill on PTH connectors, when the view angle and technique are set up for it

The void number, read properly

The figure most often quoted for BGA voiding is a ceiling of 25% of the ball’s projected area. The number is real, but treating it as a pass mark on its own is a mistake: a void sitting at the pad interface behaves very differently from the same volume sitting in the middle of the ball. Acceptance criteria come from IPC-A-610, with IPC-7095 used as the guidance for image acquisition and analysis — which is why a serious X-ray report says how the image was acquired, not only what percentage it measured.

Where it stops

X-ray cannot verify a component value, cannot read a marking or a polarity dot reliably, and cannot see surface solder defects that an optical system would catch instantly. On a dense double-sided board, a 2D system produces a shadowgram in which features from both sides overlap; separating them needs an angled or 3D technique. And the classic trap: an X-ray image that shows nothing is not evidence that nothing is wrong — it is evidence that nothing was visible in that view.

04Flying Probe: The Bare Board and the Small Batch

No fixture, no tooling cost, no minimum volume — and no way to be fast.

A flying probe tester moves a small number of heads across the board and touches one net at a time. Because there is nothing to build, it can start from CAD data within hours, which makes it the default for bare-board electrical test on prototypes, first articles and small batches — and the reason engineering changes do not stop a build.

What it reliably catches

• Opens and shorts on bare boards, net by net, to the level of test specified in the purchase order
• Continuity on loaded assemblies through accessible test points, without a dedicated fixture
• Impedance checks where the fabricator runs coupon structures, using TDR as described in IPC-TM-650 method 2.5.5.7
• Isolated defects that a fixture would average away: single-net opens introduced by a scratched trace

Where it stops

Two limits matter commercially. The first is time: probing net by net takes minutes per board, so on a 50,000-piece order the per-board test time becomes the dominant cost — which is the point at which ICT is cheaper, not better. The second is access: probe pads must exist and be reachable, and on a dense assembly the probe may simply not fit between components. Design decides this. A board with generous test points is cheap to test; a board without them is expensive to test or impossible to test fully.

05ICT and Bed of Nails: The Volume Workhorse

Hundreds of probes touch at once, so the board is tested in seconds — as long as someone paid for the fixture.

In-circuit test presses a board onto a bed of spring-loaded pins, one per test node, and measures the whole board in parallel. It verifies continuity, detects components that are missing, reversed or of the wrong value, and can be extended with powered measurements and functional checks.

What it reliably catches

• Opens, shorts and wrong-value or wrongly oriented components across every accessible node
• Assembly errors that optical inspection cannot judge: a 10 kΩ resistor where the BOM calls for 1 kΩ
• Coverage of otherwise unreachable nodes, when boundary scan is used alongside the nails

Where it stops

The fixture is the whole story. It costs money and lead time before the first board is tested, it must be stored and maintained, and a design change can obsolete it. It also cannot judge joint quality: an electrically continuous BGA ball that is structurally weak passes ICT exactly as a good one does. And where the board gives no probe access — fine-pitch packages, dense double-sided assemblies, boards with conformal coating — the nails simply cannot reach.

06The Gap No Electrical Test Closes

A board that passes continuity can still be a board that fails in the field.

All four methods above answer one of two questions: does it look right or is it connected. Neither is the same as will it survive ten years of thermal cycling. A plated barrel that is thin at the knee, a via with a cracked wall, resin recession at the resin-to-copper interface — these can measure as electrically perfect today and fail months later.

Three families of test close that gap, and they are sampled or destructive rather than 100%:

07Which Test Runs at Which Step

The same four names appear at two different points in the flow, on two different products.

08What to Put in Your RFQ

Test strategy is cheapest to fix in the quote, and most expensive to fix after the first shipment.

09Frequently Asked Questions

Related reading: Testing & inspection · Quality control · PCB assembly · Capabilities · Request a quote