Coaxial lighting to make solder balls dark on a shiny die

A polished die carrying an array of solder balls is the classic contrast puzzle: the substrate is a mirror and the balls are shiny too, yet a good bump image shows a bright field with every ball black. That picture does not come from exposure, gain or focus. It comes from geometry. Coaxial lighting sits on the camera axis and returns the flat die's reflection straight into the lens, while each curved ball sends the same light off-axis. What follows is the physics, a worked example with the arithmetic shown, and the tilt tolerance that decides whether the effect survives on your bench.

Why the balls go black: the mirror test

The law of reflection says light leaves at the angle it arrived, measured from the surface normal. Light a flat, polished die that sits square to the lens axis and every ray returns along its own path into the aperture. The die floods the image with signal. That is bright field.

A solder ball is also a mirror, but its normal sweeps through every direction as you travel around the sphere. Only the patch near the apex, where the normal still points back inside the lens acceptance cone, sends light into the aperture. The rest of the sphere reflects the illumination off to the sides and out of the image. The ball is shiny; curvature is what makes it black. A matte ball behaves differently: it scatters light back from the whole face and reads mid-grey.

This is the same bright-field versus dark-field law that decides every specular inspection, worked for flat parts in the bright field vs dark field guide.

What coaxial lighting actually is

A coaxial light folds LEDs onto the camera axis with a beamsplitter plate at 45 degrees. Illumination travels down the axis, the die mirrors it straight back, and the return passes through the splitter to the sensor. A plain non-polarizing plate passes roughly half the light downward and half of the return upward, so about a quarter of the LED output reaches the sensor. Budget for that loss with output headroom or strobing.

The coaxial lights range runs from 50 mm units for single dies to 200-400 mm plates for full panels, in standard, narrow-angle and high-precision builds. Narrow angle suits longer working distances, where a wide cone would spill past the field.

Pick a unit with output headroom for the splitter loss:

Sizing the light to the field

Two apertures matter: the active LED face and the splitter's clear aperture. Both must pass the whole field, so size the face to at least the field diagonal and treat the splitter aperture as the hard limit. Extra face area buys uniformity toward the corners, where the return cone crosses the splitter at an angle and vignettes first.

Three sizes cover most die fields, 50 mm to 100 mm: the 80 x 80 mm narrow-angle LS-CXL-0019, the 50 x 50 mm blue LS-CXL-0227 and the 102 x 100 mm LS-CXL-0067:

Worked example: 12 mm die, 0.4 mm balls, 80 mm working distance

The inputs, and where each comes from:

• Die: 12 mm square, a typical single-die bump field. Scale the numbers linearly for your part.
• Balls: 0.4 mm diameter, mid-range for BGA balls, which commonly run 0.3-0.75 mm.
• Camera: 2448 x 2048 px at 3.45 µm, a standard 5 MP 2/3-inch sensor. Sensor width: 2448 x 3.45 µm = 8.45 mm; height 7.07 mm.
• Working distance: 80 mm, chosen to clear the beamsplitter housing and the fixture above the die.
• Iris: f/4, for depth across a field that is never perfectly flat.

The rules in words: magnification from field coverage, focal length from magnification, then the lit share of each ball from the lens acceptance cone.

The arithmetic:

1. Field with 10% margin: 12 mm × 1.1 = 13.2 mm. Magnification: m = 8.45 mm / 13.2 mm = 0.64.
2. Focal length: f = 80 mm × 0.64 / 1.64 = 31.2 mm, so take the 32 mm standard. At 32 mm: m = 32 mm / (80 mm − 32 mm) = 0.667, and FOV = 8.45 mm / 0.667 = 12.7 mm × 10.6 mm, which leaves 0.34 mm of margin per side.
3. Field diagonal: sqrt(12.7² + 10.6²) = 16.5 mm. A 50 mm-class light face covers it three times over; the splitter aperture must still pass it.
4. Object-side pixel: 3.45 µm / 0.667 = 5.2 µm. Each 0.4 mm ball spans 400 µm / 5.2 µm = 77 px.
5. Aperture radius at f/4: 32 mm / 4 / 2 = 4 mm. With the half-diagonal at 8.3 mm, the acceptance half-angle is θ = atan((8.3 + 4) mm / 80 mm) = atan(0.154) = 8.7°.
6. Lit share of each ball: sin²(8.7°) = 0.023, about 2% of the face. The bright apex dot measures 77 px × 0.153 = 12 px across; the other 98% of each ball is black.
7. Tilt budget: a mirror tilted by α swings its return by 2α. The return must still land inside the 4 mm aperture radius, so 2 × tan α × 80 mm ≤ 4 mm gives α ≈ 1.4°.

Decision: with the die located in a fixture, take the 32 mm lens — 77 px per ball, and each ball reads about 98% black. If placement wanders, take 25 mm: m = 25 mm / 55 mm = 0.455, FOV = 18.6 mm with 3.3 mm of margin per side, and each ball still spans 53 px. Opening the iris to f/2 raises the lit share to sin²(11.5°) ≈ 4% and doubles the tilt budget to about 2.9°, at the cost of depth of field.

Alignment: hold the die square to the axis

The 1.4° figure is the whole art. Fixture the die rather than trusting hand placement, and shim the light square to the stage. On the bench, rotate the light by about a degree and watch the monitor: a field that dims uniformly was square, while a field that flips from white to black all at once has crossed the tilt budget. At that point every ball lights up at its rim and the die goes dark — dark field by accident. Under a flat mirror the geometry folds the aperture back onto itself at every field point, so squareness, not field position, is what to police. The same axis sensitivity shows up as slot angle in the coaxial slot defect guide.

Coaxial vs ring vs dome at the same station

A dome is the wrong shape over a flat mirror: the die simply images the dome back, dark spot at the camera hole included, which the dome camera-hole fix covers in detail. For larger solder fields on brass-like substrates, the working pattern combines geometries instead: a dome for the curved surroundings and a coaxial fill for the flat pads, as the brass solder-joint lighting guide lays out.

Color, strobing and the controller

Geometry is color-blind: white, blue, red or infrared all keep the die bright and the balls dark, because the law of reflection ignores wavelength. Wavelength earns its keep elsewhere. Blue shortens diffraction blur on fine bump pitch — the same reasoning behind selecting blue light for dark and shiny targets. Red and infrared suppress ambient light on the shop floor and sit where mono silicon sensors are most sensitive. Take a monochrome camera: color adds nothing to a bright-versus-dark decision.

The splitter costs light twice, so short exposures want strobing. Pulse the light during the exposure only: a short pulse runs the LEDs brighter than their continuous rating within the duty limit and keeps the housing cool. Iris and pulse energy trade against each other, which is why lower f-numbers gather more light at the same focal length. A strobe-capable digital light controller takes the camera's flash output and drives the pulse; the light controller finder filters by strobe feature and channel count:

When you want the opposite: bright balls on a dark die

Flank and rim defects hide against a black ball. Invert the geometry: steep grazing light around the die at 70° makes the flat mirror throw its return past the lens while the curved flanks catch it. That is the same grazing logic that finds dents and pits on metal with low-angle light.

Checks and traps before you commit

• Field coverage: light face and splitter aperture both clear the field diagonal, 16.5 mm in the example.
• Squareness: hold within 1.4° at f/4 and 80 mm working distance; re-shim after fixture maintenance.
• Ball finish: polished balls go black; matte or oxidized balls sit mid-grey, and no exposure setting changes that.
• Exposure trap: exposure and gain scale the die and the balls together on a linear camera, so the contrast ratio stays fixed; only reflection direction moves it.
• Polarizer trap: cross polarization dims the die's bright return along with everything else; keep it for glare-on-texture problems, as the cross-polarizer guide explains.
• Focus trap: focusing changes blur, not reflection direction; no focus position turns a specular return off-axis.
• Camera side: monochrome sensor, auto gain and auto exposure off, short strobed pulse.

Size the field first, pick the splitter second, pulse third: start at the light finder, or open the configurator at the light step and we will check the strobe budget with you. If the geometry fights back on a real part, ask through the contact page.