X-ray Void Measurement: Thresholds, Angles and Repeatability
Two laboratories can measure the same BGA and report void fractions that differ by a factor of two without either of them making a mistake, because the number depends on the threshold, the magnification and the projection angle. Controlling those three variables is the whole discipline of void measurement.
What Void Fraction Means and What the Specification Actually Limits
Void fraction is the projected area of the gas pockets inside a solder joint divided by the projected area of the joint itself. It is an area ratio measured through the full thickness of the joint, not a volumetric ratio, and that distinction matters: a single large void near the top surface and a scatter of small voids through the bulk can give the same number while behaving completely differently in service.
Specifications usually limit two things at once, the total void fraction and the size of the largest single void. A thermal pad might be limited to 25 percent total area with no single void over 5 percent of the joint area, because the largest void is what creates the hot spot. Read the limit in the drawing or the component specification before configuring the measurement, since a limit written for total area cannot be verified with a tool that reports only the largest void, and the reverse. Where the customer supplies a limit expressed as a diameter, convert it to an area before setting up the software, and state the conversion in the report. A limit of 0.3 mm diameter on a 3 mm joint is about 1 percent of the joint area, which is a very different requirement from 10 percent total area and will change the pass rate dramatically.
How an X-ray System Turns a Shadow into an Area
In X-ray inspection a transmission image is a map of attenuation, not a picture of the joint. The detector records how many photons passed through each pixel, and the software converts that intensity into a grayscale level that is displayed as an image. A void attenuates less than solder, so it appears brighter, and the measurement is a matter of deciding which pixels are bright enough to count.
Three system properties limit the accuracy of that decision. The focal spot size sets the geometric unsharpness, the detector pixel pitch sets the sampling limit, and the beam energy sets the contrast between solder and void. A system with a 5 micron spot and a 50 micron pixel can measure a 100 micron void reliably; the same void on a 200 micron pixel detector is a single grayscale level above the surrounding solder and will be lost in noise. That is why the smallest void a system can measure should be stated before the measurement method is written. If the specification calls for detecting 50 micron voids, the imaging chain has to support it, and no amount of threshold adjustment will recover information the detector never captured.
Thresholding: The Single Biggest Source of Disagreement
Thresholding converts the grayscale level image into a binary map of void and solder, and the threshold value decides the answer. Lower the threshold and small dark features are counted as voids; raise it and they disappear. Because the transition between solder and void is not a step but a gradient spread over several pixels, the measured area changes by a few percent for every grayscale level the threshold moves, and the effect is largest for small voids.
There are two defensible ways to set the threshold. The first is a fixed value referenced to a calibrated standard, which gives repeatability across time and across operators but may over or under report on a system whose contrast has drifted. The second is an adaptive method such as Otsu’s algorithm applied per joint, which tracks the local contrast but can produce different answers for the same joint imaged at different brightness. Whichever is chosen, write it down and use it for every measurement of that product. Where a supplier and a customer disagree, compare the threshold first, then the magnification. In most disputes the joint is identical and only the threshold differs, and the difference is enough to move a borderline joint from pass to fail. Reference the measured areas against the solder joint acceptance criteria in the same document, and state the threshold in the report so that the number can be reproduced by anyone reading it.
Magnification, Resolution and the Smallest Measurable Void
Magnification is a trade off, not a setting to maximise. Higher magnification increases the number of pixels across the joint, which reduces the error introduced by the edge gradient, but it also reduces the field of view and forces more images to be stitched. A joint measured at 4x and the same joint measured at 10x will not give the same void fraction unless the threshold is adjusted, because the edge gradient occupies a different fraction of the joint area in each image.
Choose the magnification so that the joint occupies at least 200 pixels across its shorter dimension, and state it in the method. That gives an edge gradient of a few pixels and keeps the area error below a few percent. Where the smallest void of interest is near the pixel limit, report the measurement as an upper bound or as below the detection limit rather than as a number, because a void represented by three or four pixels cannot be measured accurately at all. Store the magnification, the field of view and the detector binning mode with each image. Two images of the same joint taken with different binning are not comparable, even if the displayed magnification is the same.
Projection Angle, Board Tilt and Superimposed Features
A transmission image is a projection through the whole thickness of the board, so any feature above or below the joint contributes to the image. A void measured on a bottom side BGA can be the shadow of a via, a cut out or a component on the top side. That is the most common cause of a void that appears on one system and not on another, and it is resolved by oblique views rather than by image processing.
Oblique viewing uses the fact that a genuine void moves within the joint as the angle changes, while a superimposed feature moves independently. Take a second image at 30 to 45 degrees and compare the position of the bright region relative to the joint outline. Where the board is not flat, the tilt of the joint itself changes the apparent void area, so a warped board measured flat will report a different number from the same board measured with the joint plane normal to the beam. For ball grid arrays and other hidden joints, use computed tomography where the specification is tight. A reconstructed slice removes the superposition problem entirely and gives a genuine cross section rather than a projection, at the cost of imaging time. Where the limit is a total void fraction only, a well controlled 2D measurement with an oblique confirmation is usually sufficient.
Calibration: Phantoms, Step Wedges and Reference Balls
Calibration turns grayscale levels into a statement about geometry. The practical calibration set for void measurement is a resolution target for pixel size, a step wedge for contrast linearity, and a set of reference balls with known voids for the measurement itself. The reference ball is the most important, because it is measured under the same conditions and with the same threshold as the production joints.
Measure the reference ball at the start of each inspection session and record the result. A reading that has moved by more than a small percentage means the tube has aged, the detector has drifted or the threshold has changed, and any production measurement taken that session carries the same shift. Keep the reference balls in a sealed container and check them for oxidation, because a corroded reference ball is worse than no reference at all. Where a system has automatic calibration routines, they still need verification with an artefact. A calibration that relies on the system’s own internal reference cannot detect a fault in the internal reference.
Repeatability and Reproducibility Testing
Repeatability is the same operator measuring the same joint on the same system ten times; reproducibility is different operators on different systems measuring the same joint. Both should be measured before a void specification is enforced, because a limit that is smaller than the measurement variation cannot be applied to production.
A workable approach is to select ten joints spanning the range of interest, have three operators measure each joint twice, and calculate the spread. If the reproducibility is plus or minus 5 percentage points, then a limit of 25 percent void fraction can only be applied as a band from 20 to 30 percent, and a joint measured at 26 percent should not be scrapped on that number alone. Publish the reproducibility figure with the specification so that both sides of the agreement know what the number means. Repeat the study after any change to the system, the software version, the threshold or the magnification. An upgrade that improves the image quality also changes the measurement, and a specification written against the old response will start failing joints that were previously accepted.
Reporting Results So Two Suppliers Agree
A void report that can be reproduced contains the system model, the tube voltage and current, the magnification, the detector binning, the threshold method and its value, the number of joints measured, the definition of the joint area used as the denominator, and the reference artefact reading for the session. Without those fields, two competent laboratories will not agree, and the disagreement is not evidence that either is wrong.
State the denominator explicitly. Whether the joint area is taken from the pad outline, the solder fillet boundary or the ball diameter changes the void fraction for the same void by a large margin, and in a dispute this is often the real difference. Where a customer specifies a percentage, ask for the denominator in writing before the first batch is measured, not after the first rejection. Include at least one image of a rejected joint and one of an accepted joint in the report. Images settle arguments that tables of numbers cannot, because they show the threshold in action and make an unstated assumption visible.
When X-ray Is the Wrong Tool
X-ray measures projected area, so it cannot report a void that is smaller than the detector resolution, and it cannot distinguish a void from a low density region such as a poorly wetted interface or an unfused solder powder agglomerate. Those features look similar and behave differently, and a void specification is not the right instrument to control them.
Where the requirement is the mechanical integrity of the joint rather than the size of the gas pockets, use a microsection or a mechanical test instead. Cross sectioning gives a true view of the interface and the intermetallic layer, and it settles questions about whether the void is a process artefact or a symptom of a wetting problem. Use X-ray for screening and for process control, and use sectioning to qualify the process and to resolve disputes.
FAQ
What void fraction is acceptable in a solder joint? Read the limit from the component or assembly specification rather than from a general rule. Thermal pads are commonly limited to about 25 percent total area with no single void over 5 percent, while some high reliability products allow far less. Confirm whether the limit applies to total area or to the largest single void.
Why do two X-ray systems give different void percentages? Almost always because of the threshold setting and the magnification, not because of the measurement itself. A lower threshold counts more of the edge gradient as void, and a different magnification changes how much of the joint the gradient occupies. Compare those two settings before questioning the joint.
Can X-ray see a void that is 50 microns across? Only if the imaging chain supports it. The detector pixel pitch and the focal spot size set the smallest feature that can be resolved, and a 200 micron pitch detector cannot measure a 50 micron void no matter how the threshold is set. State the detection limit before writing the specification.
