How 3D Structured Light Measures a Solder Lead Lift That Brightness Cannot Show

OPTICS

A lead lifted 30 μm parallel to its pad leaves no trace in brightness and shows up only as a 15.6° phase shift in 3D structured light fringes.

Taking lifted connector leads, I calculated which lifts 2D lighting catches and misses, how strong the fringes must be to read 3 μm by phase shifting, and when shiny solder makes the height disappear.

About 14 min read

The gull-wing leads of QFPs and connectors are plated with tin or nickel and reflect strongly in the specular¹ way, and a solder fillet after reflow becomes a curved mirror. The solder resist around them is a matte surface that scatters light evenly. In a field of view where surfaces this different sit side by side, the task is to find a lifted lead², one lead floating slightly above its pad.

Two leads that went through the same reflow line can come out almost identical in brightness and color in a 2D image. One sits properly on its pad and the other is lifted, yet what the camera records is the brightness of reflected light, so unless the height difference translates into brightness there is no way to tell them apart. A lift like this stays quiet after passing inspection until it surfaces as a poor contact or an open circuit. I split the problem into three questions: which lifts 2D lighting catches and which it misses, what 3D structured light³ reads height from, and when that measurement breaks down in front of shiny solder.

Tilted lifts and parallel lifts

Lifts come in shapes. In one, the toe rises and the foot tilts; in the other, the whole foot floats parallel to the pad. When the tip of a lead with a 0.5 mm foot rises 30 μm, the foot tilts about 3.4°. Assuming the coaxial illumination⁴ is collimated to about the lens half-angle, a reflective face tilted 3.4° bends the reflected light by twice that, 6.9°, while a lens with an object-side numerical aperture⁵ of 0.05 accepts a half-angle of only about 2.9°, so the tilted foot fails to return light to the lens and shows up dark. A lift like this is caught in 2D as well.

A foot raised parallel to the pad is a different story. The angle of the reflective face is unchanged, so its brightness under coaxial light is unchanged too, and a 30 μm height difference does not show up in focus either. Estimating the depth of field⁶ of the same lens as λ/NA² gives 0.55 μm ÷ 0.05² = 220 μm, so a 30 μm difference is buried inside it. Judging indirectly from the shape of shadows or highlights is possible, but the result swings with lighting angle, lead geometry and neighboring parts, which tends to raise false calls and escapes together. In my judgment, the moment a parallel lift is suspected, it is faster to look for a way to measure height directly in place of brightness.

On the left of the figure are two leads that brightness alone cannot separate; on the right, the same two leads split by color in a height map. The right-hand picture is made without comparing brightness even once. All it takes is calculating how far the pattern has shifted at each pixel.

Height read from how far the fringes shift

3D structured light inspection measures the position of a pattern in place of brightness. When a projector casts sinusoidal stripes at an angle, the stripes appear shifted sideways on surfaces of different height. CKD’s US patent US7245387B2 (priority date February 28, 2002; granted July 17, 2007) describes in its abstract projecting three sine-wave patterns with different phases onto a board printed with cream solder and calculating the solder height from the captured images and the relative phase angles between the patterns. In this method, called phase shifting⁷, the quantity measured is how far the pattern has moved.

Geometry sets the relation between height and phase. Assume the camera looks straight down, the projector is tilted 30°, and one stripe period on the board is 400 μm; the equivalent wavelength⁸, the height over which the phase turns one full cycle, is then 400 ÷ tan 30° = 693 μm. A 30 μm lift becomes a phase of 2π × 30 ÷ 693 = 0.27 rad, or 15.6°, and the range readable without unwrapping the phase is ±346 μm. The height of a lifted lead fits easily inside that range.

Inspection that measures height directly has long been used in solder paste inspection. Koh Young Technology’s company page says it launched the world’s first 3D solder paste inspection system the year after its founding in May 2002, and its history lists the KY-3030 launch in February 2003. A 3D AOI system carries the same principle over to leads and fillets after reflow.

How strong the fringes must be to read 3 μm

Phase wobbles with noise. In N-step phase shifting, where the stripes are moved and shot N times, with pixel brightness noise σ and fringe amplitude⁹ B, the phase noise is roughly √(2/N) × σ ÷ B. I check that formula against a Monte Carlo calculation over 200,000 pixels before relying on it. With 4 steps, σ = 2 gray levels and B = 52 levels, the calculation gives 0.0272 rad, matching the formula, and that phase noise equals 693 × 0.0272 ÷ 2π = 3.0 μm of height.

If judging a 30 μm lift requires measurement repeatability within a tenth of it, 3 μm, then with 4 steps B must be 26 times σ, that is, at least 52 gray levels. With the 3 steps in the patent, it takes 30 times, or 60 levels. More steps lower the noise at the same B, but the number of shots and the time grow with them. That is why I measure fringe amplitude B separately on the resist, on the leads and on the fillets early in setup and keep it in a table.

The numbers in that table differ widely from surface to surface. A surface that scatters light, like the resist, spreads the slanted stripes in many directions and sends the camera its share, while a near-mirror surface sends its whole share in a single direction. This is where B can differ by tens of times under the same projector and the same exposure.

Fringes vanish in front of a shiny fillet

A near-mirror fillet sends the stripe light in one direction, at the same angle it came in. If that direction misses the camera, fringe amplitude B drops close to 0; if it points straight at it, the pixels saturate and the tops of the sine curve get clipped. Either way the material for the phase calculation is ruined. In the same 4-step calculation, when B on the sloped face of a fillet falls to 5 levels, the phase noise is 0.30 rad and the height noise about 33 μm, larger than the 30 μm lift itself.

The heights at these pixels hardly qualify as measurements. The inspection algorithm has to mark pixels with fringe amplitude below a threshold as invalid and drop them from the decision, and when that invalid area covers the toe of a lead, the basis for judging the lift disappears with it. Software can fill the gap from surrounding values, but a filled value is an estimate, hard to use as is for a lift decision. How much invalid area appears as solder gloss and fillet curvature vary from lot to lot cannot be guaranteed before a sample test.

This is why equipment makers keep several sets of lighting. Koh Young Technology’s US patent US7400413B2 (priority date November 13, 2006; granted July 15, 2008) claims an apparatus that measures 3D shape by shadow moiré¹⁰, with a control part that switches several illuminating parts on and off selectively according to the shape of the test object. The same company’s US8437533B2 (priority date October 13, 2008; granted May 7, 2013) claims a method that takes reflection images with a lighting device for 3D measurement and one for 2D measurement, realigns the inspection area by comparing feature information from a database, and then inspects that area for distortion. Both patents keep 3D measurement from resting on a single set of lighting.

Projection directions that erase shadows

Light cast at an angle brings shadows with it. Behind a connector body 3 mm tall, a 30° projection leaves a shadow 3 × tan 30° = 1.73 mm long, and no stripes reach the leads inside it. Adding a projector on the opposite side lets the two shadows fill each other, and projecting from four directions gives every lead around a part fringes from at least one direction. More directions also raise the chance of meeting a sloped fillet face that sends light back toward the camera.

The price is time. Four directions at four steps each means 16 shots per field of view, and assuming a camera at 60 frames per second, that alone takes about 0.27 seconds. Images for the 2D decision and data processing time come on top, so I would work backward from the line takt to the number of fields of view and set the number of directions and steps from there. On a board carrying only low parts that cast no shadows, cutting to two directions to save time is an option.

Catching a 30 μm lift, and when 2D is enough

When choosing the equipment, write down three things together: how many degrees of phase the defect height becomes, where and how much fringe amplitude is secured, and how invalid pixels are handled. Taking 10 μm/px on the board with a 4,096 × 3,000 px sensor of 3.45 μm pixels gives a field of view of 41.0 × 30.0 mm at a magnification of 0.345×, and one 400 μm stripe period spans 40 px. With a telecentric lens¹¹, whose magnification does not change with height, and a lens front radius assumed at 28 mm, the WD¹² that keeps the 30° projection path clear of the lens body must be at least (28 + 20.5) ÷ tan 30° ≈ 84 mm.

Item	① Minimum defect size	② Optical setup · lighting · WD	③ Algorithm parameters
Height resolution	30 μm lift = 15.6° of phase (equivalent wavelength 693 μm)	Projector at 30°, stripe period 400 μm on the board = 40 px, 10 μm/px, field of view 41.0 × 30.0 mm, telecentric lens at 0.345×	4-step phase shifting, height repeatability target 3 μm, fringe amplitude at least 52 levels (noise 2 levels assumed)
Reflection · invalid pixels	Height noise 33 μm on a fillet (B = 5 levels)	Measure fringe amplitude separately on resist, leads and fillets; check for saturated pixels	Mark pixels below the amplitude threshold invalid; hold the decision when the toe falls in an invalid area
Shadows · WD	1.73 mm shadow behind a 3 mm tall part	Four projection directions; WD at least 84 mm with a lens front radius of 28 mm assumed	4 directions × 4 steps = 16 shots, about 0.27 s per field at 60 fps. Separate coplanarity limits per lead

Only with all three rows filled does 3D structured light become equipment that actually judges a 30 μm lift. Geometry sets the height resolution, and keeping that resolution on the floor is shared among fringe amplitude, invalid-pixel handling and projection directions.

Plenty of joints, though, need nothing more than 2D. For defects that show up as differences in color or pattern, such as print defects and foreign matter, or for missing parts and reversed polarity, 2D is faster and the equipment simpler. As the calculation above showed, a lead whose toe tilts up can also turn dark under coaxial light and be caught in 2D. The cost of 3D earns its keep when the defect is height at heart, as with parallel lifts, insufficient solder height and component coplanarity, and even then how far the real detection rate goes on a board full of shiny fillets cannot be guaranteed before a sample test.

The calculation that decides whether a lift is a brightness problem or a height problem is the first sum an engineer choosing inspection equipment does, and the same judgment keeps returning in assembly process and quality work. If you plan to bring in 3D structured light, ask for the invalid-pixel ratio on fillets next to the height accuracy on the spec sheet. Height accuracy only means something on pixels that hold a value. This article explains technology and is not investment advice. See our Disclaimer.

Field Note

If this connector line were to catch lifts in 3D, my first measurement would be fringe amplitude at three spots: the resist, the top of a lead and the sloped face of a fillet. If the resist clears 52 levels while the fillet drops to single digits, that gap predicts the size of the invalid area, so 20 of the same leads get reshot with the projection direction changed each time to see whether the invalid area covers the toe. The lift limit stays at 30 μm, and the last check is whether the standard deviation of height over 10 repeated measurements of the same lead stays within 3 μm.

Field Checkpoints

Once the fringe amplitude table and repeat-measurement results are in, the questions below reach the verdict on whether the 3D decision goes into production.

• WD: is the WD long enough to keep the 30° projection path clear of the lens body? (at least 84 mm with a 28 mm lens front radius and 20.5 mm half field assumed)
• Has fringe amplitude been measured separately on the resist, the lead tops and the fillet slopes and compared with the 52-level mark (4 steps, noise 2 levels)?
• Are pixels below the amplitude threshold marked invalid and dropped from the decision?
• Is the share of each lead toe covered by invalid area logged lead by lead?
• Was the number of projection directions set from the shadow length by part height (height × tan 30°)?
• Does the height standard deviation over 10 repeated measurements of the same lead stay within 3 μm?
• If only tilted lifts matter, has it been checked whether coaxial 2D is enough?

Frequently Asked Questions

Are the moiré method and the phase-shift method different technologies?

Both use the way stripes shift with height. Shadow moiré reads the moiré pattern formed where a grating overlaps its own shadow cast on the object, while phase shifting computes the phase at each pixel from several shots taken as the pattern moves by set steps. A shadow moiré apparatus with a grating-moving actuator, like Koh Young’s US7400413B2, uses both ideas at once, so on the floor the bigger difference comes from how many shots are taken and how much fringe amplitude is secured, more than from the name.

Do two machines with the same height accuracy figure give the same results?

Start by checking what kind of sample the spec sheet’s height accuracy was measured on. If it was a flat sample with ample fringe amplitude, then on a board full of shiny fillets the decision can differ even under the same figure, depending on how many invalid pixels appear and how they are dropped or filled. I would ask for the invalid-pixel ratio and the lead repeatability on samples of our own boards and compare those.

How is the lift height limit set?

It starts from the gap solder can fill between pad and lead and from the customer’s coplanarity requirement. The text uses 30 μm as an example, but the real limit has to come from the part datasheet and the customer specification, and repeatability is checked against a tenth of that limit. With a 50 μm limit, the requirement relaxes to 5 μm repeatability and a fringe amplitude of about 31 levels at 4 steps.

Glossary

• [1] Specular reflection specular reflection
  Reflection that leaves a smooth surface in one direction at the same angle the light came in. It is strong on plated leads and solder fillets. ↩
• [2] Lifted lead lifted lead
  An assembly defect where a component lead floats instead of touching its pad. The toe alone may rise, or the whole foot may float parallel. ↩
• [3] Structured light structured light
  Projecting light with a known pattern, such as stripes, and computing 3D shape from how the pattern bends or shifts on the surface. ↩
• [4] Coaxial illumination coaxial illumination
  Lighting that turns light with a half mirror and sends it down along the lens axis. Flat reflective faces look bright and tilted ones look dark. ↩
• [5] Numerical aperture NA
  The sine of the largest half-angle of light a lens accepts. An NA of 0.05 corresponds to a half-angle of about 2.9°. ↩
• [6] Depth of field DOF
  The range of distance over which the image still looks in focus. For a diffraction-limited lens it is roughly the wavelength divided by NA squared. ↩
• [7] Phase shifting phase-shift profilometry
  Shooting several images while moving a sinusoidal pattern by set steps, computing the phase at each pixel from its brightness changes and converting it to height. ↩
• [8] Equivalent wavelength equivalent wavelength
  The height that corresponds to one full turn (2π) of phase. It is the stripe period divided by the tangent of the projection angle. ↩
• [9] Fringe amplitude fringe modulation
  Half the difference between the bright and dark parts of a sinusoidal pattern. The larger it is, the smaller the phase noise. ↩
• [10] Shadow moiré shadow moiré
  Reading height from the moiré pattern formed where a grating overlaps the shadow it casts on the object. ↩
• [11] Telecentric lens telecentric lens
  A lens whose object-side chief rays run parallel to the axis, so magnification barely changes with the height of the object. ↩
• [12] WD working distance
  The distance from the front of the lens to the inspected part. It decides how much room the projectors and lights have. ↩

References

• CKD Corp., US7245387B2 “Three-dimensional measuring instrument”, priority date 2002-02-28, granted 2007-07-17. Google Patents
• Koh Young Technology Inc., US7400413B2 “Three-dimensional shape measuring apparatus using shadow moire”, priority date 2006-11-13, granted 2008-07-15. Google Patents
• Koh Young Technology Inc., US8437533B2 “Method of measuring a three-dimensional shape”, priority date 2008-10-13, granted 2013-05-07. Google Patents
• Koh Young Technology, “How we began” company page (founded May 2002, KY-3030 launched February 2003). kohyoung.com

Related reading — When 3D+AI Inspection Answers the Shape Defects 2D Can’t Catch · The Defect Zone You Miss When You Switch to 3D Inspection · In 2D/3D Fusion Inspection, the Bottleneck Is Lighting Interference