中文English

White-Light Structured-Light Thickness and Deformation Measurement | Non-Contact Inspection of Carrier Warpage, PCB Warpage and Solder Paste Thickness

On the SMT and EMS floor, the flatness of carriers (Carrier / Fixture) is a variable that is easily overlooked and hard to measure. When a carrier is not flat, solder paste thickness is distributed unevenly and the Z axis of component placement is off, and the result is open joints, tombstoning and product warpage. This article explains the principle of white-light structured-light measurement, the system components, measured validation data, and its actual relationship with international warpage standards.

1. Problems with traditional measurement methods

  • Visual inspection: cannot be quantified, judgment varies from person to person, and it leaves no traceable record.

  • Contact measurement (such as a micrometer): can measure just points or lines, with coverage of just 5% to 10%. Carrier warpage is usually a continuous curved surface, and sampling a few points does not capture the true magnitude of a central bulge or corner warpage.

  • Both make it hard to set up a unified, traceable inspection standard — this is the real pain point: it is not that you cannot measure, it is that even when you measure, you cannot compare across batches and across operators.

2. How structured light measures height

Structured light is an active optical 3D measurement technique. It uses a controllable light source (for example, a DLP projector) to project a known mathematical pattern — optical fringes — onto the surface of the object being measured. When the surface has height variations, these fringes deform correspondingly as seen from the camera’s viewpoint; by analyzing the degree of fringe deformation through geometric triangulation, the 3D depth information of the surface can be derived.

Two fringe algorithms used together

MethodHow it worksWhat it solves
Gray code fringe method (Gray Code)Projects binary Gray code fringesClearly identifies the absolute index of each fringe, avoiding phase-jump errors caused by reflections or occlusion
Phase-shifting fringe method (Phase-shifting)Projects multiple (usually 3 to 9) fringe images with small phase offsets and solves the phase differenceObtains high-resolution continuous depth information

In addition, the system uses a “multi-frequency fringe fusion” strategy, combining high-frequency and low-frequency fringes to solve the phase wrapping (2π ambiguity) problem, so that objects with large height differences over a wide area can also have their 3D shape reconstructed accurately.

These two algorithms are complementary, not a choice of one or the other: Gray code handles “which fringe this is” (absolute positioning, but coarse resolution), and phase-shifting handles “where within this fringe” (fine resolution, but used alone it cannot tell which fringe it is). Using just one of them, you either lack precision or get step jumps at reflective spots.

3. Why white light

Light sourceAdvantagesLimits
White lightLow cost, broad spectrum, high compatibility with general industrial camerasMore sensitive to highly reflective surfaces
Blue lightGood monochromaticity, strong resistance to ambient stray lightHigher cost
Red lightSharp projection, suited to high-precision point scanningLaser safety protection must be considered

Four reasons for choosing white light: for surfaces common on the SMT floor, such as solder paste and PCB green solder mask, white light provides uniform reflection characteristics; its spectrum matches the sensitivity range of standard CMOS cameras, so no special sensor is needed; its cost is significantly lower than blue-light DLP modules; and the accuracy this system needs is about ±25 μm to ±50 μm, which white-light technology already fully meets.

The last point is the key to the selection logic: the accuracy requirement determines the technology choice; higher accuracy is not always better. If the requirement is ±25 μm, choosing a solution that can reach ±5 μm but costs several times more means the extra money buys no corresponding quality improvement. First define clearly “how fine you need to measure”, then choose the technology.

4. System components and stability

ModuleFunctionCore hardware
White-light projection moduleProjects structured-light fringesDLP2010 DMD + white LED
Dual industrial camera moduleCaptures deformed fringe images1 MP CMOS × 2, GigE interface
Control computerProjection synchronization, image computation and reconstructioni7 CPU + GPU acceleration
Precision opto-mechanical mountFixing and calibration positioningAnodized aluminum alloy frame + XYZ fine adjustment
Software systemControl, reconstruction, display, reportingIn-house developed GUI and analysis algorithms

The mount uses a composite design of “aluminum alloy main frame + industrial aluminum extrusion substructure”, with silicone anti-vibration pads added to suppress high-frequency vibration. In a 4-hour stability test, thermal drift was kept within ±5 μm.

The rigidity and thermal stability of the opto-mechanical mount often determine actual accuracy more than the specifications of the camera or projector. The system accuracy is ±25 μm, and the mount’s 4-hour thermal drift is ±5 μm — thermal drift takes up one fifth of the error budget. When accepting this kind of equipment, always ask for thermal drift data, and ask for the measurement duration. A thermal drift figure measured over just 10 minutes is meaningless; the production line runs continuously.

5. Algorithm flow and color interpretation

From image capture to visualized results, the whole process completes within 1 second, in seven steps:

  • Projection and image capture: the projector projects fringes at high speed, and the two cameras capture images synchronously by hardware trigger.

  • Phase decoding: a phase map is solved by combining the Gray code and phase-shifting algorithms.

  • 3D reconstruction: triangulation converts the phase map into a per-pixel height map (Z-map).

  • Plane fitting and de-tilting: a least-squares (minimized sum of squared errors) method automatically finds the reference plane and corrects it to a zero plane, eliminating errors from tilted placement.

  • Color threshold classification: height data are mapped to colors according to preset or custom tolerance ranges.

  • Noise filtering: regional weighted filtering smooths edges and suppresses noise points.

  • Result display and report output: 2D / 3D color maps are displayed in real time, and a statistical report is generated with one click.

Step four, “de-tilting”, is easily underestimated. Without automatically finding the reference plane, the measured height differences would include the tilt caused by “the object not being placed flat” — a completely flat carrier, if just placed askew, would measure as if warped on one side. Whether plane fitting is done determines whether measurement results can be compared from one measurement to the next.

Color grading (example)

LevelHeight rangeColorMeaning
Normal range−50 μm to +50 μmGreenFlat or within acceptance tolerance
Slight deviation+50 μm to +100 μmYellowMild deformation, needs attention
Severe deviationBeyond ±100 μmRedExceeds tolerance; scrapping or repair recommended

The tolerances mapped to the color scale can be adjusted automatically or set manually according to the product type (SMT carriers, PCBs, etc.).

6. Three measured validations

ExperimentConditionsResults
Carrier warpage measurementA carrier with a known deformation of up to +0.21 mmMeasured peak +0.208 mm, error within 0.01 mm; over 10 consecutive measurements the repeatability standard deviation σ = 11 μm
PCB warpage inspectionA 100 mm × 160 mm PCBAccurately showed the 0.13 mm warpage distribution at the center, error range ±0.03 mm; inspection speed more than 8 times faster than manual measurement
Solder paste thickness inspectionSolder paste after printing with a 0.12 mm thick stencilMeasured average thickness 0.109 mm; height difference between left and right PADs less than 10 μm

The figures in the third item are worth a second look: the stencil thickness is 0.12 mm, and the average height of the printed solder paste is 0.109 mm, about 91% of the stencil thickness. This is normal — solder paste slumps slightly after release, and the release rate of an aperture is not 100% in the first place. If the measured ratio is clearly low, what should be reviewed is the aperture area ratio and release, not the measuring instrument. What really matters is “the height difference between left and right PADs is less than 10 μm”, which reflects the uniformity of squeegee pressure and stencil contact.

7. Comparison with traditional sampling inspection

Evaluation itemTraditional sampling methodThis systemDifference
Flatness coverage5 to 10% (point or line sampling)100% full coverageMore than 10 times
Inspection time per piece3 to 5 minutesAbout 1 secondSee explanation below
Inspection error±0.10 mm±0.025 mm4 times the accuracy
Operator dependenceHigh (relies on experience)Low (automated judgment)Can be standardized

For the inspection time row, the initial source data state the improvement as “reduced by 95%”, but the actual calculation is far more than that: 3 minutes is 180 seconds and 5 minutes is 300 seconds; dropping to about 1 second is a reduction of more than 99%. The 95% in the source text clearly underestimates its own figures. When citing, use the before and after times directly (180–300 seconds → about 1 second), which is clearer than a percentage and will not be questioned.

Coverage is where the real value of this method lies. Contact measurement can just take points, while carrier warpage is a continuous curved surface — sampling measurement may happen to miss the areas of greater deformation and produce a “pass” result. What full coverage changes is not accuracy but whether anything gets missed.

8. Corresponding international standards (and an important boundary)

StandardStandard titleHow it corresponds
IPC-7095EDesign and Assembly Process Guidance for Ball Grid Arrays (BGAs)—

There is a boundary here that must be made clear, or you will be caught out in a technical audit: JEITA ED-7306 and JEDEC JESD22-B112 are both “elevated-temperature” package warpage measurement standards — at their core is measuring how warpage changes with temperature during a heating process that simulates reflow. This system currently measures at room temperature, and the subject of both standards is the IC package body, not the carrier; this system’s data can be used just as a reference for comparison and cannot be claimed to conform to these two standards. If a customer requires elevated-temperature warpage characterization per JESD22-B112, a room-temperature system cannot substitute for it.

Incidentally, a correction to one name: JEDEC JESD22-B112 covers warpage measurement at the package level, not the board level. Use the correct name when citing, and you will stand on firmer ground in an audit.

The future development directions in the initial technical document also list “an elevated-temperature version (measuring in a 150 °C environment, for analyzing dynamic warpage changes during reflow)” — which corresponds exactly to the boundary above. Other plans include using a deep learning model to automatically identify deformation types (central bulge, corner warpage) and judge whether a part is repairable or must be scrapped, a fully automatic inline version combined with a conveyor, and fusing a ToF sensor to compensate for highly reflective areas to improve measurement accuracy on mirror surfaces and polished metal.

FAQ

How is structured-light measurement different from AOI?

AOI mainly judges “whether it is there, whether it is aligned, whether there are visual defects”, mostly with 2D or limited 3D information; structured-light measurement outputs a height value for each pixel (Z-map) and answers “how large the height difference is”. The two have different purposes: AOI catches missing parts, offsets and polarity errors, while structured light measures warpage, flatness and thickness distribution.

Why use both the Gray code and phase-shifting algorithms at the same time?

The two are complementary. Gray code determines “which fringe this is”, providing absolute positioning but coarser resolution; phase-shifting determines “where within this fringe”, with fine resolution but, used alone, it cannot distinguish the fringe index. Using just one of them, you either lack precision or get phase jumps at reflective spots. The system additionally uses multi-frequency fringe fusion to solve the phase wrapping (2π ambiguity) problem.

Which should I choose: white light, blue light or red light?

It depends on the accuracy requirement and the surface of the object being measured. White light is low cost, has a broad spectrum and high compatibility with general industrial cameras, and has uniform reflection characteristics on solder paste and PCB green solder mask; its drawback is greater sensitivity to highly reflective surfaces. Blue light has good monochromaticity and strong resistance to ambient stray light but higher cost; red light gives sharp projection and suits high-precision point scanning, but laser safety protection must be considered. The key is to first define how fine you need to measure — this system’s requirement is ±25 to ±50 μm, which white light can meet. Higher accuracy is not always better; accuracy beyond the requirement buys no corresponding quality improvement.

Why do plane fitting and de-tilting before measurement?

Because an object placed askew will be misjudged as warped. The system uses a least-squares (minimized sum of squared errors) method to automatically find the reference plane and correct it to a zero plane, removing the variable of “tilted placement”. Without this step, a completely flat carrier placed at a slant would measure as warped on one side, and since the placement angle differs each time, the data could not be compared from one measurement to the next.

The solder paste measured 0.109 mm, thinner than the 0.12 mm stencil — is there a printing problem?

No. Solder paste slumps slightly after release, and the release rate of an aperture is not 100% in the first place, so measuring around ninety percent of the stencil thickness is normal. What really matters is the height consistency among PADs on the same board — in this case the height difference between left and right PADs is less than 10 μm, reflecting good squeegee pressure and stencil contact. If the overall ratio is clearly low, what should be reviewed is the aperture area ratio and release conditions.

Can this system be used for JEDEC JESD22-B112 elevated-temperature warpage testing?

Not at present. JESD22-B112 and JEITA ED-7306 are both elevated-temperature package warpage measurement standards, whose core is measuring how warpage changes with temperature. A room-temperature system cannot run the full test of the standards, and the subject of these two standards is the IC package, not the carrier, so the data can be used just as a reference for comparison. If a customer requires elevated-temperature warpage characterization, a model with temperature control capability is needed.

What does a repeatability standard deviation of 11 μm mean?

It represents the spread of results when the same object is measured 10 times in a row. This figure is closer to everyday use than “accuracy ±25 μm” — accuracy describes how close you are to the true value, while repeatability describes whether you get the same answer each time you measure. In process control, an instrument with insufficient repeatability will make variation that does not exist appear on the control chart.

What should I ask when accepting this kind of equipment?

Ask at least four things: first, thermal drift data and the measurement duration (a figure measured over just 10 minutes is meaningless; the production line runs continuously); second, the repeatability standard deviation — do not look just at the nominal accuracy value; third, whether the measurement includes automatic plane fitting and de-tilting; fourth, if elevated-temperature warpage standards need to be met, confirm whether the model has temperature control capability.

Standards referenced

For the specifications cited in this article, please refer to the current version from each issuing body.

Further reading

For SMT printing, reflow and other process measurement topics, you can read further:

Related products

Jiahon supplies the items mentioned in this article.

From: Jiahon Engineering Department

Back to top