SMT Assembly Inspection: SPI, AOI and X-Ray Checks Explained
Inspection in a modern SMT assembly line is not a final gate that sorts good boards from bad ones. It is a set of measurements placed at each stage of the process, so that a deviation is caught where it is created rather than after it has been buried under a component. A solder paste deposit that is slightly small is invisible after reflow, but it is plainly visible immediately after printing, and that is why the first inspection station sits before the placement heads rather than after the oven.
Why Inspection Sits Inside the Process
Every defect has a point of origin. Insufficient paste comes from the printing step. A shifted component comes from placement. A void under a thermal pad comes from the reflow profile. Voiding inside a ball comes from the paste chemistry and the profile together. If the only inspection is at the end of the line, the feedback loop is hours long and every board produced between the cause and the detection is suspect.
The practical answer is a short loop. Measure the paste, correct the printer. Measure the placement, correct the feeder. Measure the joint, correct the profile. Each station reports into the same traceability record, so a board that fails at the end can be traced back to the machine state that made it. That structure is what a mature PCBA development process looks like in production.
<img src="https://www.gopcba.com/wp-content/uploads/2026/05/呼吸机PCBA-2.png" alt="SMT assembly line with solder paste inspection station” />
Solder Paste Inspection
Solder paste inspection is the earliest and cheapest measurement in the line. It is a laser or structured-light scan of the printed deposits, and it reports volume, area, height and position for every aperture on the stencil. Volume is the number that matters most, because a joint that fails for lack of solder almost always began as a deposit that was short of volume.
A typical control window holds printed paste between roughly 80 and 120 percent of nominal volume, with height close to the stencil thickness, commonly 0.12 to 0.15 millimetres for fine-pitch work. The two failure directions have different causes. Low volume points to a clogged or partially blocked aperture, a worn stencil, insufficient squeegee pressure, or paste that has dried on the stencil. High volume points to a gasket that no longer seals, so paste is forced under the stencil and smeared around the aperture.
Placement Accuracy and Component Size
Placement accuracy is quoted in two ways that are easy to confuse. The repeatability figure describes how tightly a machine returns to the same point, and the accuracy figure describes how close it gets to the position it was told to reach. For a chip component of 0402 size or larger, either number being small relative to the pad is sufficient. For 01005 and smaller parts, where the body is around 0.4 by 0.2 millimetres, the tolerance is a large fraction of the component and the machine specification becomes a real constraint rather than a datasheet formality.
The reason placement drift shows up as a defect is worth stating plainly, because it is usually a solder problem rather than a component problem. A part placed slightly off centre is pulled into position by the surface tension of molten solder, provided the deposits are correct and the pad geometry is symmetric. When parts tombstone, the cause is normally an imbalance in paste volume or thermal mass between the two ends, not the placement accuracy itself. The physical mechanisms are set out in SMT component shift causes.
Automated Optical Inspection
Automated optical inspection is the workhorse station after reflow. A camera system images the board, usually with structured light or several viewing angles so that height can be inferred, and compares what it sees against a program built from the CAD data. It detects missing components, wrong components, polarity errors, shifted or rotated parts, insufficient and excess solder, bridging, and lifted leads.
The limit of optical inspection is that it only sees what light can reach. Solder joints under a BGA, under a QFN thermal pad, or inside a land grid array are hidden, and no amount of camera resolution changes that. This is not a small gap, because those packages carry the highest pin counts and are the most expensive to rework. The other limit is program quality: a threshold set too tight produces false calls that operators learn to dismiss, and a threshold set too loose misses real defects. Both failure modes reduce the station to a formality.
<img src="https://www.gopcba.com/wp-content/uploads/2026/08/high-speed-circuit-board.webp" alt="X-ray inspection of BGA solder joints on an assembled board” />
X-ray Inspection
X-ray inspection exists to cover the region that optical inspection cannot see. A transmission image shows the solder under a package as a density map, so a missing ball, a bridged pair, or a ball that has not collapsed is visible. Two-dimensional imaging is fast and sufficient for gross defects. Three-dimensional computed tomography, reconstructed from a set of angled views, separates the layers of a package and is what makes void measurement meaningful, because a void percentage read from a single projection is the sum of every void along the path.
Voiding deserves a careful reading. Some voiding under a thermal pad is normal and acceptable; the question is how much, and where. Voiding concentrated in one corner of a large pad creates a hot spot, because the heat path is thinnest exactly where the remaining solder is smallest. A specification written as one percentage for the whole pad can pass a joint that will run hot. Where the joint is a thermal path rather than an electrical one, the rule should be written in terms of the largest contiguous void rather than the total.
Reading Yield and DPPM Data
Defects per million opportunities is the standard measure of line quality, and it is only useful when the opportunity count is defined consistently. A board with two thousand joints and one defect has a different rate from a board with two hundred joints and one defect, and comparing the two figures without normalising makes a simple product look worse than a complex one. When a supplier reports a rate, the useful question is what counts as an opportunity and whether the figure covers every inspection station or only the final one.
Trend matters more than the absolute number. A stable rate with defects spread evenly across types is a healthier line than one that averages the same figure while alternating between a clean week and a week dominated by bridging. The second pattern means the process is drifting in and out of control, and the average conceals it.
Reflow, Traceability and First Article
The reflow profile is measured rather than assumed. A thermocouple attached to a representative board, with the assembly instrumented at the thermal mass that heats slowest, gives the real profile: the ramp, the soak, the time above liquidus and the peak. The profile that matters is the one at the coldest joint on the board, not the one at the sensor.
Traceability ties the measurements together. If the paste inspection record, the placement program revision, the reflow profile and the inspection results are keyed to the same board serial number, a field failure can be traced to the machine state that produced it. Without that, a defect report produces a guess. Layer onto that the practice of inspecting the first board of a run in full rather than by sampling, and the loop closes: the placement order and pad positioning of a new product are validated before the run starts, and the measurements that follow keep it in control.
FAQ
What does solder paste inspection measure? The volume, area, height and position of each printed deposit. Volume is the most useful figure, because a joint short of solder usually began as a deposit short of volume.
Why is X-ray inspection needed when there is optical inspection? Because optical inspection cannot see under a BGA, a QFN or a land grid array. X-ray transmission shows the solder hidden beneath the package, and computed tomography separates its layers.
Is all voiding under a thermal pad a defect? No. Some voiding is normal. What matters is the largest contiguous void, since a single large void in one corner of a pad concentrates heat where the solder path is thinnest.
What is a good DPPM figure? The absolute number matters less than its definition and its stability. A consistent rate with defects spread evenly across types indicates a controlled process; a fluctuating rate does not.
