Using CCD Vision to Inspect Laser Solder Joints Inline

CCD vision inspection of laser solder joints inline means capturing an image of each joint immediately after the laser reflow step — and usually a pre-solder alignment image as well — then running rule-based or learned a…

CCD vision inspection of laser solder joints inline means capturing an image of each joint immediately after the laser reflow step — and usually a pre-solder alignment image as well — then running rule-based or learned algorithms on that image to accept or reject the joint before the board indexes to the next station. On a laser soldering cell the camera is not a separate AOI box downstream: it shares the motion platform, the fiducial map and the trigger with the soldering head, which is what makes highly inspection possible at roughly 0.5–2 s per device without adding a second handling step. The practical result is a closed loop — align, solder, look, decide — with the temperature trace and the image stored against the same serial number.

What can a CCD camera actually see on a laser solder joint?

It sees the surface, the silhouette and the specular behaviour of the fillet. That covers more of the defect spectrum than most people expect, but not all of it, and knowing the boundary is what keeps an inline inspection station credible.

• Presence and position: component offset, rotation, tombstoning, missing part — measured against the fiducial-derived nominal position, typically to ±10 µm at 5 µm/pixel.
• Solder volume proxies: fillet length and height, wetting angle against the pad edge, and pad coverage percentage. A joint with less than roughly 75% pad coverage on a Class 2 assembly is a flag, not a scrap decision — the operator confirms.
• Discrete defects: solder balls above the agreed size limit (commonly 0.13 mm for Class 2, tighter for fine-pitch), bridging between adjacent pads, splatter on keep-out areas, icicles and sharp points.
• Surface quality: dull, grainy or non-wetted surfaces that correlate with a cold joint — especially useful when cross-checked against the recorded temperature curve for that specific joint.

The honest limit: a top-down CCD image cannot see voids, cracks or the intermetallic layer under an opaque joint. If the application is a power device, a QFN thermal pad or a BGA-like geometry, void fraction has to come from X-ray. Treat CCD as the highly surface gate and X-ray as the sampled or first-article subsurface check.

How is the camera integrated into an inline laser soldering cell?

Two imaging positions are normal, and they use different optics because they solve different problems.

• Pre-solder alignment camera — looks down through or beside the soldering head, finds board fiducials and component edges, and corrects the laser spot position before the pulse. Telecentric lens, coaxial or ring illumination, and a narrow bandpass filter so the imaging light is not swamped by ambient or by laser back-reflection.
• Post-solder inspection camera — either a second head on the same gantry or a scan station immediately after. For fillet height and coplanarity, 3D acquisition (laser triangulation or phase-shift fringe) is preferred over pure 2D, because a 2D image of a shiny fillet is dominated by specular highlights.

Three integration details cause most of the field problems we see:

• Wavelength separation. Diode soldering lasers run around 915 nm or 980 nm; Nd:YAG at 1064 nm. Put an OD4+ blocking filter at the laser wavelength in front of the sensor, and illuminate at a separated wavelength (e.g. 470 nm blue or 850 nm) so the filter does not starve the image.
• Trigger and settling. Fire the inspection trigger only after the solder has solidified — a minimum 80–150 ms dwell after laser off for small joints, longer for high-thermal-mass parts. Imaging a joint that is still liquid produces a fillet shape that will never exist again.
• Fume and flux management. Flux volatiles condense on the lens window within a few thousand cycles. A positive-pressure air curtain plus a scheduled window-clean interval (and a reference-image drift check) prevents the slow slide into blurred, over-tolerant inspection.

What resolution, lighting and optics do you need?

Size the optics from the smallest feature you must judge, not from the field of view. A working rule: the smallest defect dimension should span at least 3–5 pixels.

Use HDR or multi-exposure capture for solder. A single exposure either clips the specular fillet or crushes the matte pad into black; combining two or three exposures and fusing them gives usable grey levels on both. Also keep the light source on a calibrated driver — LED output drops measurably over a few thousand hours, and unreferenced brightness drift shows up as a slow change in pass rate that nobody can explain.

How do you set acceptance criteria and prove the inspection is working?

Start from the drawing, not from the software default. Take the applicable IPC-A-610 class limits, convert each visual criterion into a measurable image feature (fillet length in pixels, coverage percentage, wetting angle band, ball diameter), and write the tolerance band into the recipe. Then validate in three steps:

• Golden-board correlation. Teach the recipe on 5–10 known-good boards and 10–20 deliberately marginal ones (partial solder, reflowed with reduced energy to force dull fillets). If the marginal set passes, the thresholds are too loose.
• Attribute agreement study. Have two or three trained operators judge 50–100 joints independently and compare against the machine. Target Kappa above 0.75; below 0.6 the recipe is not trustworthy regardless of what the pass rate says.
• Ongoing drift checks. One golden sample per shift, plus a monthly review of the false-call and escape logs. A recipe that has not been re-correlated in six months is a guess.

Watch the false-call/escape trade-off explicitly. Tightening a threshold to catch one more marginal joint almost always costs false calls elsewhere; record both numbers so the change is a decision rather than an accident.

Troubleshooting: symptom, likely cause, what to check

• Sudden rise in false “insufficient solder” calls. Check the lens window for flux haze first, then the LED output level, then whether a new board lot has a different pad finish (ENIG vs OSP changes contrast markedly). Re-run the golden sample before touching thresholds.
• Good joints passing that the operator rejects. Usually a recipe taught on one lighting condition and never re-taught; or the inspection image is taken before the joint has solidified, so the machine is judging a smoothed, always-acceptable shape. Verify trigger delay against the temperature curve.
• Intermittent position errors on the same panel. Fiducial contrast or a partially covered fiducial; confirm the alignment camera is finding at least two fiducials and that the residual error is logged per board.
• Machine and X-ray disagree on “voiding”. They are measuring different things. CCD is not a void tool — remove that check from the CCD recipe and put it in the X-ray sampling plan.

FAQ

Can CCD inspection replace functional test or X-ray? No. It replaces the visual inspection step and adds per-joint traceability. Electrical test verifies connectivity; X-ray verifies subsurface voiding. The three are complementary.

What cycle time does inline inspection add? Typically 0.5–2 s per device for 3D acquisition and 0.3–0.8 s for 2D, which usually hides inside the laser soldering takt on a multi-head cell. On a single-head cell with a long solder dwell, it is effectively free.

Do I need a separate inspection station after the soldering cell? Only if the soldering gantry cannot carry a second camera or if the board must cool before imaging. Sharing the platform is cheaper and avoids a second alignment.

How often must the system be re-calibrated? Alignment camera against a certified target monthly or after any crash; lighting and inspection thresholds at each product changeover; golden-sample verification every shift.

If you are specifying an inline laser soldering cell and want the vision parameters matched to your joint geometry and takt, send us your board data and we will work through the optics and acceptance criteria with you.