X-Ray Slice And CT Inspection For PCB Quality
An X-ray slice is a cross-sectional image of a board taken from a rotational scan, and it is the only non-destructive way to look inside a plated hole, a via or a solder joint. Where a two-dimensional image collapses the whole thickness of the board into one shadow, a slice isolates a single plane and shows it at the resolution of the detector.
This article explains how a slice is produced, what computed tomography adds, how voiding is read in a solder joint, and where the acceptance criteria come from.
Why A Top-Down Image Is Not Enough
A conventional X-ray image is a transmission shadow. Everything in the beam path contributes, so a via under a ball grid array appears as a grey disc whose greyness is the sum of the copper in the barrel, the copper in the pads above and below, and any void inside. A thirty percent void in one barrel and a twenty percent void in a differently shaped barrel can produce the same grey level, so the method cannot size a defect, only suggest that something differs from the reference board.
The problem is worst for the defects that matter. A crack that runs part way around a barrel, a void that sits at the interface between the plating and an inner layer pad, and a solder joint that has separated only at its edge all project as a small change in optical density. An operator can see that the joint is not identical to its neighbours without being able to say which layer has failed or how much of the joint is affected.
How A Slice Is Produced
A slice is built the same way a medical scan is built, but at a much smaller scale. The board sits on a rotation stage between a microfocus source and a flat panel detector. The source is a spot a few micrometres across, and its small size is what allows geometric magnification: move the board closer to the source and its shadow grows, so a three micron detail can cover many detector pixels. The stage rotates through one hundred and eighty or three hundred and sixty degrees in small increments, and a projection is captured at each step.
A single projection is still a shadow. The reconstruction is what produces the slice: the projections are combined by filtered back projection or an iterative algorithm into a volume where each voxel carries an absorption value. Software then cuts that volume at the plane of interest, and the same data set can be re-cut at any other plane without rescanning the board. That is the main practical advantage over a physical cross section, which destroys the sample and shows only the plane that was ground and polished.
What Computed Tomography Adds
The same rotational data set can be presented as a series of slices or as a rendered volume, and the two views answer different questions. A stack of slices through a ball grid array shows every row of joints at its own equator, the plane where the true joint diameter is greatest and where a void area is at its maximum. A rendered volume treats the whole interconnection as one object and is better for showing a crack that wanders between planes or a barrel that is thinner on one side than the other.
Resolution and field of view trade against each other, and the trade is set before the scan starts. A large board viewed as a whole gives a voxel size of perhaps ten to twenty micrometres, enough to see that a joint is missing or grossly voided but not enough to measure a thin plating. Zooming to a single package or a single region of interest on the same stage brings the voxel size down to one or two micrometres, at the cost of scanning only that region. In practice the whole-board scan finds candidates and the zoomed scan characterises them.
Reading Voiding In A Solder Joint
A void is a pocket of gas trapped in the joint during reflow, and on a slice normal to the joint axis it appears as a well defined dark region inside the bright solder. The measurement that matters for an acceptance criteria decision is the total void area as a percentage of the joint cross section, taken on the plane where the joint is widest. Several small voids scattered through a joint behave very differently from one void that spans the whole interface, so most specifications also limit the largest single void and the voiding where the joint meets the pad.
The judgement must be made from a slice rather than from a top-down image, because the densest path through the joint dominates the shadow. A void that sits low in the joint is hidden behind the solder above it in a two-dimensional view and can be completely invisible. This is why voiding measured from above is usually lower than voiding measured on a slice, and why the two numbers should never be compared. Reference limits for the alloy itself are set out in the guidance on lead free versus leaded solder.
Sample Mounting And Throughput
A slice needs no sample preparation beyond mounting the board, which is the reason it is used for production screening and for failure analysis on boards that must survive the examination. Cycle time, however, is dominated by the rotational scan: a full set through a small region can take several minutes, and a large board at coarse resolution can take far longer than a laboratory schedule allows. Throughput improves when only the planes and regions flagged by a top-down pass are scanned.
Fixturing matters more than most operators expect. Any movement of the board during the rotation blurs the reconstruction, and the blur looks like a smeared joint rather than a motion artefact. Heavy assemblies are supported from below, connectors are braced so they cannot flex, and boards are scanned flat, the orientation that puts the least mass in the beam path.
Acceptance Criteria And Reporting
The acceptance criteria for a slice come from the same source as any other inspection limit: the workmanship standard the customer has invoked, the internal specification derived from it, and the class of the product. For plated holes the limits are barrel copper thickness, crack length and separation at the inner layer connection. For solder joints the limits are void area, void position and the presence of any crack between solder and pad. Barrel defects are described in more detail under copper plating defects prevention.
A report should state the voxel size, the plane of the slice, the magnification and the measurement rule used. Without the voxel size the measurement cannot be judged, because the same defect measured at two micrometres and at fifteen micrometres gives different numbers. The report should also state what was not inspected. A slice samples one plane through one feature, and a clean slice does not certify the rest of the board.
Limits Of The Technique
Density contrast is the fundamental limit. Two materials with similar absorption, such as a lead-free solder and a copper pad, or a void filled with flux residue rather than gas, can be indistinguishable in the volume, and a thin crack closed by compression may not appear at all. Very thick or very dense assemblies, such as those with a metal stiffener or a heavy heat sink, absorb so much that the region behind them is noisy.
Interpretation is the other limit. A slice is easy to produce and hard to read, and telling a genuine crack from an artefact of beam hardening or cone angle takes experience. Laboratories usually scan a known good reference board with the same settings, so the operator compares a suspect slice against a real example rather than against a mental image of what a good joint ought to look like. The same discipline of comparing against a defined reference runs through every quality characteristic of a board design.
FAQ
Can a slice replace a physical cross section? For most production and failure analysis questions it can, and it leaves the board usable afterwards. A physical section is still used to confirm plating thickness and grain structure, and to check features that carry too little density contrast for X-ray.
How much voiding is acceptable? It depends on the specification the product is built to and on the thermal and mechanical duty of the joint. Many commercial requirements allow a total void area of around twenty five percent with no single void above ten percent, while high reliability assemblies tighten the limit at the interface.
Why does the same joint measure differently on two machines? Voxel size, reconstruction algorithm, threshold setting and the plane chosen for the measurement all change the number. Comparing results requires identical settings, and those settings belong in the report.
