X-Ray Inspection of PCB Assemblies: What It Finds

Once a ball grid array is reflowed, nobody can see the joints underneath it. X-ray inspection is the standard answer, and it is also one of the most misunderstood processes in electronics manufacturing, because the image shows density rather than geometry and the operator has to interpret what the shadows mean.

Why X-Ray Is Used

Optical inspection, whether by eye or by camera, only sees what the light can reach. Any joint that is underneath a package, inside a barrel or behind a component is invisible to it, and modern assemblies put most of their interconnections in exactly those places.

The typical candidates are ball grid arrays, land grid arrays, chip scale packages, package on package stacks, quad flat no lead parts with their thermal pad, and through hole connectors where barrel fill has to be confirmed. Any of these can have an open, a short or a partially formed joint that passes a functional test at room temperature and fails later, and X-ray is the only cost effective way to see them in volume.

The value of the technique is not only defect detection. It is also process control. Voiding percentage, ball diameter, barrel fill and joint shape all drift as the paste, the stencil or the profile changes, and imaging a sample every shift gives an early warning long before the defect rate rises.

How the Image Is Formed

X-ray inspection works by transmission. A beam passes through the assembly, and the detector records how much of it survives. Denser material, meaning more solder or more copper, absorbs more and appears darker, while less dense material appears lighter. The result is a projection, not a photograph, and everything along the beam path is superimposed.

Most systems operate in two modes. A two dimensional transmission image gives a plan view of the whole board in a single pass, which is fast and is what most inline machines use. An oblique or angled view tilts the beam so that a joint can be seen from the side, which reveals the vertical shape of a fillet and can separate a joint from the copper underneath it. Some systems combine several angled views into a computed tomography reconstruction, which produces a slice through the board and gives the clearest picture of an individual joint, at the cost of much longer inspection time.

Resolution is quoted in micrometres of feature size, and it is a trade with throughput. A system that can resolve a 5 micrometre feature needs high magnification and a small field of view, so it inspects slowly. A production line usually runs at a resolution that is adequate for the smallest feature on the board and no better, which for a typical 0.4 mm pitch ball grid array means a resolution in the tens of micrometres.

What X-Ray Finds

The defects that X-ray is used for are specific, and knowing them makes the acceptance criteria easier to set.

• Opens and missing balls. A joint that never formed shows as a gap in the ball pattern, and a missing ball is obvious as an empty site.
• Bridging and shorts. Excess solder between adjacent balls or leads appears as a connection where there should be a gap.
• Insufficient or excess solder. A ball that is too small or too large relative to the pattern indicates a paste or stencil problem, and both are visible as a change in the density profile.
• Voids. Gas trapped during reflow leaves a lighter region inside the joint. Voiding is normal to some degree and becomes a defect only when it exceeds a specified percentage of the joint area or when it is positioned to affect the joint.
• Cracks. A crack through a ball or at the pad interface shows as a fine line in the density, though it is much easier to see in an angled view or in a slice than in a plan view.
• Head in pillow. The ball of the package sits on top of the paste but never coalesces with it, so the joint looks present in a plan view but has an insulation boundary inside. This is the defect that X-ray struggles with most, and it usually needs an angled view or a cross section to confirm.
• Through hole barrel fill. For connectors and press fit parts, the amount of solder inside the barrel can be measured from the image, which is much faster than cross sectioning.

Where the Limits Are

X-ray is a shadowgraph, and the shadow is the limit.

Where several dense features sit on top of each other, the image becomes unreadable. A package on package stack, a board with a heavy ground plane under a ball grid array, or a thick copper layer under a fine pitch device all reduce the contrast between the joint and its surroundings. The usual remedies are an angled view, a slice, or a change of inspection strategy rather than an increase in magnification.

Board thickness and material matter as well, because the beam has to penetrate the whole assembly. A thick metal core board or a large heatsink over the area of interest can make the joints underneath effectively invisible.

The third limit is interpretation. A plan view shows density, not the shape of the joint above and below the ball, so a joint that looks satisfactory from above can still be weakly attached at the pad. That is why many acceptance schemes rely on a combination of X-ray for the bulk of the population and destructive cross sectioning for the first article and for any confirmed suspect, which is the same technique the bare board fabricator uses to qualify plating and hole walls.

Voiding is the classic case of a criterion that is set by the customer rather than by physics. A void reduces the cross section of the joint, which affects its mechanical strength, but the effect is not linear and the industry has moved over the years on what an acceptable percentage is. Most specifications now express a limit as a percentage of the joint area, sometimes with an exception for a single large void, and some products accept substantially more voiding than others. The practical approach is to agree the figure in the assembly documentation and to measure it consistently, rather than to treat every visible void as a defect.

Inline, Offline and Sampling

There are two ways to deploy the technique, and the choice is economic rather than technical.

An inline machine sits in the assembly line after reflow and inspects every board at production speed. It catches defects at the point where they are made, gives immediate feedback to the process, and is the normal arrangement for high volume production of boards with area array packages. The cost is capital, floor space and the programming effort to create a reliable inspection program.

An offline machine is used for sampling, for first article inspection and for failure analysis. It costs less to operate, it allows longer inspection times and higher magnification, and it lets a technician look at an individual suspect joint without stopping the line. Many smaller assembly houses run an offline system and use it on a defined sample, which gives process visibility without the capital of an inline machine.

The sampling plan matters as much as the machine. A first article inspection of a new assembly should be thorough, including oblique views and cross sections of representative joints, because that is where the process is proven. In volume, a periodic sample with a defined acceptance rule is usually enough, provided the process is stable and the paste, stencil and profile are controlled. Where a defect is found, the sample size should be increased until the cause is identified and corrected.

Two supporting conditions make X-ray worth the effort. The first is a clean board: contamination, flux residue and stray solder balls confuse the image, so cleaning and handling matter. The second is documented acceptance criteria with pictures of acceptable and unacceptable joints, because an inspection program is only as good as the operator or algorithm that decides what a good joint looks like.

Working With the Assembly Partner

Three questions are worth settling before a build starts. Which joints are critical enough to require imaging, which is the same as asking where the design has hidden or fine pitch interconnections. What acceptance criteria apply, including the voiding limit and the treatment of head in pillow. And how a suspect joint will be resolved, whether by angled view, by slice, or by cross section.

It is also worth agreeing how the results will be reported. A pass or fail line item is not very useful; the value of X-ray comes from the numbers behind it, such as the voiding distribution, the ball diameter trend and the barrel fill measurement, because those are the signals that show a process drifting before it produces a defect. Where the board is complex enough to have a quality plan, those measurements belong in it as process indicators rather than only as pass or fail results.

FAQ

• Can X-ray see a short between two inner layers? It can see the density difference if the layers are separated enough to resolve, which is why angled views and computed tomography are used for inner layer faults.
• Is X-ray safe for the components? Yes for inspection purposes. The dose used for imaging is small and does not damage semiconductors, unlike the higher energy radiation used for deliberate testing.
• Does every board need X-ray? No. It is used where the joints cannot be seen optically, plus a sample for process control. A board with only leaded and visible parts is usually covered by optical inspection.
• How long does an X-ray inspection take? From a few seconds per board for a fast plan view to several minutes for a high resolution computed tomography scan of a single area.

Summary

X-ray inspection exists because the most important joints on a modern board are the ones nobody can see. It images by transmission, so it shows density rather than shape, and that single fact explains both what it is good at and where it struggles. Opens, bridges, missing balls, insufficient solder and barrel fill are all clearly visible; a crack through a ball or a head in pillow joint may need an angled view or a cross section to confirm.

Deployed inline it gives immediate process feedback and catches defects where they are made. Deployed offline it is a sampling and failure analysis tool, which is enough for many products. In either case the results are only useful if the acceptance criteria are documented, the sample size is defined, and the measurements are treated as process indicators rather than only as pass or fail. That combination is what turns an inspection step into a control on the assembly process, and it is what makes the data worth collecting on every build rather than only when something goes wrong.