What is a PCB inspection?
A PCB inspection checks a bare or assembled circuit board for defects before it moves on. On an assembly line it includes SPI after printing, AOI after placement and reflow, AXI for hidden joints and electrical tests such as in-circuit and functional test. Visual inspection catches placement and solder faults early, before they are buried under further assembly.
What is a solder bridge?
A solder bridge is unwanted solder joining two leads or pads that should be electrically separate. It usually comes from excess paste, stencil misregistration or fine-pitch pads. On the line, 2D AOI sees the bridge as a continuous bright band between leads, and the deep-learning layer separates true bridges from reflections and silkscreen that rule-based AOI often calls by mistake.
What is tombstoning in PCB assembly?
Tombstoning is a chip component standing on one end after reflow, soldered to only one pad. It happens when one pad wets before the other or carries more paste, so surface tension lifts the part. Angled cameras and 3D AOI show the raised body clearly, and the model learns tombstones across package sizes from 0201 upwards without separate rules per size.
How do you check for bad solder joints?
You check solder joints by imaging each fillet under multi-angle lighting and comparing its shape and volume with the acceptance criteria. Colour dome lighting shows fillet slope, 3D AOI measures volume and height, and X-ray covers hidden joints such as BGAs. AI classification sorts each call into bridge, insufficient, excess, lifted lead or false call, so operators only re-check real candidates.
How does dead pixel detection work on a panel line?
Dead pixel detection works by driving the panel to uniform test patterns and imaging it with a camera that has several sensor pixels per display pixel. A dead sub-pixel shows as a dark point on white or colour fields; a stuck one shows as a bright point on black. The system records the pixel coordinate, colour channel and pattern for each fault.
How does mura detection work?
Mura detection works by imaging the panel on grey and colour test patterns, removing lens and sensor non-uniformity with flat-field correction, then finding low-contrast luminance or colour gradients. Deep learning helps because mura shapes vary: clouds, bands, spots and edge glow. The model grades each region against your limit samples instead of a single global threshold.
What is wafer defect inspection?
Wafer defect inspection finds and classifies scratches, particles, pattern defects and residue on wafers between process steps. Patterned-wafer tools use bright-field and dark-field optics to flag defect coordinates; review and classification then decide which are killer defects. Opsio works at the classification stage, labelling each flagged defect so engineers review the ones that affect yield.
What is 3D AOI inspection?
3D AOI inspection measures height on the board as well as the 2D image, usually with phase-shift structured light projected from several angles. It adds solder volume, component coplanarity and lifted-lead height. The deep-learning layer uses the height map together with the colour image, which helps on reflective fillets and dark components where 2D images are ambiguous.
What is the difference between AOI and AXI inspection?
AOI uses visible light and sees only what is on the surface; AXI (automated X-ray inspection) sees through components to hidden joints. AOI covers placement, polarity, visible fillets and bridges at line speed. AXI is used for BGAs, QFNs and bottom-terminated parts where joints are hidden. AI classification works on both image types, but this page focuses on optical inspection.
Can AOI inspect through-hole components?
Yes, AOI can inspect through-hole components, usually from the solder side after wave or selective soldering. It checks pin protrusion, fillet shape, bridges between pins and missing solder. Through-hole joints vary more than SMT joints, which produces many rule-based false calls, so a trained model that has seen your real good and bad joints usually helps most here.
What does AOI programming involve?
AOI programming means building a library of inspection rules for every package, footprint and board, then tuning thresholds until false calls are acceptable. Each new package or revision needs more tuning. A deep-learning layer reduces that work: it learns from labelled AOI images and operator verdicts, so new packages are added by labelling examples rather than rewriting rules.
What does an AOI false call mean?
An AOI false call is a joint or component the AOI flags as defective that an operator then judges acceptable. False calls come from tight thresholds, reflections, silkscreen, supplier changes and board colour variation. They cost operator re-verification time and train operators to approve calls quickly, which raises the risk of escapes. Reducing false calls is the main economic case on SMT lines.
What is the difference between wafer inspection and metrology?
Wafer inspection looks for defects anywhere on the wafer; metrology measures specific dimensions such as film thickness, critical dimension and overlay at chosen sites. Inspection answers what is wrong and where; metrology answers whether a feature is within specification. AI classification adds most value on inspection output, where thousands of flagged defects need sorting.
What is the IPC-A-610 standard?
IPC-A-610 is the industry acceptance standard for electronic assemblies, defining what target, acceptable and defect conditions look like for solder joints and components. It sets three product classes: Class 1 general, Class 2 dedicated service and Class 3 high performance. An inspection system can apply class-specific thresholds; it does not certify your product or your operators.
How should a solder joint look?
A good solder joint shows a smooth, concave fillet that wets both the pad and the termination, with the lead outline visible. IPC-A-610 sets minimum fillet height, side overhang and end-joint width by component type and class. Dull, grainy, convex or non-wetted joints are indicators to check. These criteria become the labels the model learns.
