4.3 AXI Techniques

Automated X-ray inspection uncovers the hidden world beneath BGAs, QFNs, and plated-through holes, where optical inspection and in-circuit test can’t reach. By visualizing collapse, voiding, and hidden defects, AXI provides an inside look at solder joint health that directly links to long-term reliability. The choice of method—fast 2D, angled 2.5D, or detailed laminography—balances speed with clarity, ensuring that every board gets the right level of scrutiny without choking throughput. When paired with calibrated setups, clear acceptance limits, and MES-linked decision rules, AXI transforms invisible risks into quantifiable, traceable data that engineers can act on.

4.3.1 AXI in one minute (why we bother)

Automated X-ray inspection sees hidden joints—BGAs/CSPs/WLCSPs, QFN thermal pads, stacked packages, even plated-through holes. AXI answers three questions fast:

Is something there? (missing balls, misaligned packages)
Did it wet? (collapse/shape for BGAs, edge fillets on QFNs/LGAs)
How gassy is it? (voiding percentage and pattern)

Use AXI where AOI is blind and ICT can’t reach.

4.3.2 The three flavors (speed vs detail)

Mode	How it works	What it’s great at	Trade-offs
2D (single projection)	One top (and/or bottom) shot	Fast inline screening: missing balls, bridges under BGAs, big voids in QFN thermals	Overlap/“stacking” hides detail; copper shadows can trick it
Oblique / 2.5D	Slight angle sweep, a few projections	Better at spotting lifted balls, skewed rows, large void pools	Slower; still some overlap artifacts
Laminography (planar CT)	Many angled shots → reconstruct a slice at one height	Separates layers: reads QFN thermal pads cleanly; inspects one BGA plane without neighbors	Slowest inline; often used as sampling or offline root-cause

Rule of thumb: inline 2D/2.5D for every board, laminography for the tricky few (NPI, audits, or when a defect spikes).

4.3.3 BGA health: collapse, voids, and HIP clues

Ball collapse: good joints look like “hourglasses” (thinner middle, bonded at the pad). Flat/round “marbles” suggest no wetting or cold joints.
Void limits: set per ball and per package. Typical starting points (tune to customer spec):
- Per-ball void area: ≤25% (flag >25–30%).
- Cluster shape matters: one big void near the intermetallic interface is worse than many tiny ones.
HIP (head-in-pillow) signs: halo or “cap and ball” with a dark seam between paste and sphere, often non-collapsed center balls. If HIP rises, revisit TAL/soak and consider N₂ (see 9.3/9.5).

4.3.4 QFN / LFPAK thermal pads: measure what matters

Use laminography or at least oblique views to isolate the pad plane—2D often lies when copper stacks up.
Judge both total void % and pattern: lots of tiny, spread-out voids are kinder than one giant bubble under the die center.
Typical starting guardrails: total void ≤25–35% (customer-specific). If you’re higher: check windowed apertures (7.4) and soak profile (9.5).

4.3.5 Thick boards, heavy copper: making physics your friend

Thick FR-4, backplanes, or big heat-spreaders absorb X-rays and create overlap.

What to change

Energy & current: bump kV/µA to punch through—but avoid blasting noise into the image. Pair with longer integration or frame averaging for clarity.
Geometry: use magnification wisely; too high magnification on thick boards narrows the field and increases blur.
Angles: add oblique views to separate stacked features; flip the board when practical to examine the other side’s pads.
Regions of interest: don’t image the world—scan only the BGA fields, QFN thermals, and suspect zones to keep cycle time sane.

4.3.6 PTH and thick barrels (if you inspect THT)

AXI can estimate barrel fill by grayscale through the hole:

Judge % fill, void gaps, and solder cones.
On high-aspect holes in thick boards, expect longer exposures or dual-angle shots. Pair findings with wave/selective process tuning rather than tightening AOI.

4.3.7 Build a stable AXI program (library + recipe)

Import CAD/centroid so the tool knows pad centers, ball pitch, and row/column counts.
Teach per package: expected ball count, window masks for QFN thermals, and slice heights for laminography.
Golden images: store “good” joints for each BGA/QFN so reviewers have a reference.
Throughput plan: inline = 2D/2.5D with targeted ROIs; offline (NPI/FA/escapes) = laminography/CT on a handful of parts.

4.3.8 Limits & decisions (so reviews are fast)

Set three bins per feature:

Pass (green): below void/shape limits, ball counts OK.
Review (yellow): borderline void %, odd shapes, localized clusters.
Fail (red): missing ball / severe HIP image / gross voids or bridges.

MES link should block WIP on red, queue yellow for human review, and store images with the SN for board-to-box traceability.

4.3.9 Common artifacts → quick fixes

What the image shows	Likely artifact	Fix
Grainy “snow,” low contrast	Low exposure or motion blur	Raise integration/avg frames; check conveyor stop
Dark streaks under copper pours	Shadowing/overlap	Use oblique angle or laminography; shrink ROI
Balls look oval in one direction	Geometric blur at high magnification	Back off magnification or refocus at pad plane
“Voids” that move with angle	Part features, not voids	Lock slice height; confirm on second angle
Random “missing” balls near tray edges	Teach/align error	Re-teach package; verify CAD import origin/θ

4.3.10 Calibration & GR&R (trust the picture)

Flat/dark field correction on schedule (removes detector drift).
Geometric check with a reference grid so distances and counts stay true.
Golden board recheck: same packages, same ROIs, same results.
GR&R sampling (3 ops × 3 repeats) on void % and ball count so numbers survive audits.

4.3.11 What to do when AXI finds trouble (smallest fix first)

BGA HIP/poor collapse → extend and smooth TAL, add N₂ if marginal; verify VIPPO is filled/cap-plated; improve board support.
QFN voiding → adjust windowing and soak; N₂ can help; avoid more peak unless needed.
Random missing balls → feeder/pick shock or board warp; check placement logs and supports.
Uniform high voids across a lot → paste/lot issue or profile drift; A/B with a fresh jar and re-profile.

(Ties back to Chapter 9’s defect playbook and Chapter 7’s printing.)

4.3.12 Pocket checklists

Program & setup

CAD/ROIs loaded; packages taught (ball counts, slice heights)
Energy/integration set for board thickness; oblique views defined for hidden fields
Golden images attached for quick review

Limits & flow

Void % limits per ball and QFN thermal set (customer-aligned)
Pass/Review/Fail routing to MES with image save
Inline = 2D/2.5D; laminography reserved for NPI/audits/escapes

Health

Flat/dark & geometric calibration current
GR&R spot check scheduled; sample runs plotted
Image artifacts list posted (so reviewers fix causes, not widen limits)

When AXI programs are tuned with the right imaging technique, stable calibration, and well-defined review thresholds, they shift from reactive troubleshooting to proactive quality control. The benefit is consistent visibility into hidden joints, fewer latent escapes, and confidence that what can’t be seen by eye is still under control.