BGA Voiding in Power Stages — Causes, X-ray Patterns, and Profile Fixes

The first thing to understand is that not all BGA voids matter equally. IPC-7095D allows up to 30% total voiding under a signal ball and is silent on thermal pads — which is exactly where the problem usually lives. A 0.1mm² void under one of 600 BGA-256 signal balls is a curiosity. A 25% void under the central thermal pad of a 7mm × 7mm power QFN is a heat-removal failure waiting for its first thermal cycle.

Voids in the thermal pad concentrate because three forces converge there:

• Outgassing has nowhere to go. The pad is large, the paste deposit is large, and the escape paths to the perimeter are long. Flux volatiles that would happily vent under a 0.5mm ball end up trapped under a 5mm thermal slug.
• Wetting starts at the perimeter. Once the solder collapses and the perimeter seals, anything still gaseous in the centre is locked in.
• The pad is solder-mask-defined (or vias break the surface). Mask dams, exposed via barrels, and tented-via dimples each pin a small pocket of flux. Multiply by thirty vias on a typical thermal pad and the void footprint adds up fast.

On our Madurai line we audit voiding rates per programme, not per panel. Signal-ball voiding stays under 8% on Class 3 builds without intervention. Thermal-pad voiding, left alone, can drift past 35% — and we've seen DC-DC converters fail thermal shock at month four because of exactly that.

"A thermal pad with 30% voiding is a heatsink with 30% delamination. Nobody would ship that as a heatsink. Why are we shipping it as a thermal joint?" — Pioneer Horizon SMT process lead

The rest of this article walks the diagnostic and corrective steps in the order we run them on the line: X-ray signature reading, paste/aperture changes, profile changes, design feedback. Get them in that sequence and you'll spend hours instead of weeks.

A 2D X-ray image gives you a void map. A trained eye gives you a root cause. We classify every voided joint into one of four signatures, and each points to a different fix.

1. Macro-void, single bubble, centred

One large void (often > 25% pad area) sitting near the geometric centre. This is a paste-volume problem: too much paste, too little outgassing time at soak. The flux liberates volatiles faster than they can escape and a single dome forms.

2. Constellation of small voids, edge-biased

Dozens of small voids (each 1–3% pad area) clustered toward the perimeter. This is a reflow profile problem — specifically a ramp rate above 2.5°C/s through 150–200°C. Flux is being driven off too fast for the paste to absorb it without trapping.

3. Voids tracking visible via locations

Void positions correlate one-to-one with via-in-pad locations under the pad. This is a design + paste-mask issue: open via barrels are pulling flux down and trapping it. Tented vias on the secondary side, or filled-and-capped vias, eliminate this.

4. Pinhole field, uniform distribution

A scattered field of sub-50µm pinholes. Usually moisture pickup in the paste — either old paste, paste that wasn't conditioned properly, or boards baked inconsistently before placement. Less common but easy to miss because each individual void is small.

For 2.5D oblique imaging we tilt to roughly 45° and look for the dome shadow — a real void casts a hemispherical shadow, while solder mask artefacts cast a flat one. The difference matters when you're trying to decide whether a 12% reading is genuine voiding or a software false-positive.

The classification step takes about a minute per joint once you've trained your X-ray operator. We keep a reference library of annotated signature images on the line PC so a new operator can pattern-match within a few weeks.

Before touching the reflow profile, exhaust the paste and stencil interventions. They're faster to validate, cheaper to revert, and they typically deliver the biggest single-step reduction.

Aperture window-paning

The most reliable fix for centred macro-voids is to break the thermal pad aperture into a grid pattern — typically a 3×3 or 4×4 split with 0.2mm dams between sub-apertures. The dams force venting paths from the centre to the perimeter during reflow. We've seen voiding drop from 28% to 9% on a 5mm thermal pad with no other change.

Aperture area reduction

For thermal pads, reduce paste deposit to 50–70% coverage of the pad area. Yes, this gives up some thermal mass, but it leaves room for the part to seat fully and for volatiles to escape. The thermal-resistance penalty of 50% coverage versus 100% is typically under 0.5°C/W on a real-world stackup — a price worth paying.

Paste choice

• Type 4 (20–38µm) over Type 3 (25–45µm) for any pad with vias breaking through. Finer powder packs better around obstructions and leaves less trapped flux volume.
• Low-voiding flux chemistries exist as a specific SKU from most major paste vendors (Indium 8.9HF, Senju M40, AIM M8). They cost 15–25% more per kg, but on thermal-pad-heavy builds the yield arithmetic favours them.
• Halide-free chemistries typically vent slower than their halide-containing counterparts. If you're locked into halide-free for compliance reasons, plan on tighter profile control to compensate.

Stencil thickness

A 100µm stencil deposits roughly 70% of the volume of a 150µm stencil at the same aperture. For mixed-pitch boards (0201 alongside 1.0mm BGA), step stencils let you thin the deposit over thermal pads while keeping fine-pitch deposits intact. The tooling cost is recovered within one or two production runs on most programmes.

We log every stencil change against the resulting void rate on a per-programme basis. That history is what lets us walk a new customer through the right starting point in our stencil aperture reduction guide rather than restarting the experiment from scratch.

If paste and stencil changes haven't pulled voiding into spec, the profile is the next lever. We're cautious here because profile changes affect every joint on the board, not just the BGA thermal pads — so the experiments need to be done with profiling thermocouples in place and yield data captured for the whole board.

Soak duration and temperature

A longer soak (60–90 seconds in the 150–200°C band) gives flux volatiles time to vent before the solder collapses. The trade-off is that excessive soak time degrades the flux, leaving you with poor wetting on perimeter signal balls. Our default starting point for a thermal-pad-heavy build is 75 seconds at 165–195°C — adjust from there based on void measurement.

Ramp-to-peak rate

Faster ramps trap more gas. Slower ramps degrade flux. We aim for 1.5–2.0°C/s from soak exit to peak — anything outside that window has consistently produced either constellation-pattern voids (too fast) or wetting defects (too slow).

Time above liquidus (TAL)

Extending TAL from 60 seconds to 90 seconds typically reduces voiding by 3–5 absolute percentage points. Past 90 seconds you start incurring intermetallic growth that costs you on long-term reliability — so the win plateaus quickly.

Peak temperature

SAC305 paste fully reflows at 217°C. We aim for a peak of 235–240°C on thermal-pad-heavy boards — high enough for surface tension to pull voids out, low enough to stay under the IPC J-STD-020 MSL rating of the BGA. Above 245°C you're paying for marginal void reduction with component-stress dollars.

Vacuum reflow when economics permit

A vacuum reflow oven drops voiding to under 2% on thermal pads, full stop. The capex isn't trivial — typically 2.5–3× a standard convection oven — but on power-electronics-heavy programmes it pays back in field-reliability terms within two to three years. We added vacuum capability on Line 4 specifically for our automotive customers.

Every profile change gets a five-board confirmation run before going to production. Profile drift caught at validation is profile drift that didn't ship.

Process levers have limits. After exhausting stencil, paste, and profile, the remaining voiding is structural — it lives in the design and can only be fixed by the layout engineer. This is the conversation we have at the DFM review, and the one that decides whether the next revision lands at 8% voiding or 22%.

Via-in-pad strategy

1. Tented secondary side for power QFNs where you can afford to lose some thermal conduction. Good enough for most consumer power stages.
2. Filled and capped (epoxy-filled vias plated over) for any thermal pad where you need full pad area for paste. Adds 8–12% to bare-board cost but unlocks 1–2 absolute percentage points of voiding reduction reliably.
3. Microvia stacked stacks where layer count and HDI budgets allow. The footprint is smaller, the via volume is smaller, and the trapped-flux pocket is smaller.

Thermal pad size and shape

A thermal pad larger than 5mm × 5mm should always be window-paned in the stencil. A thermal pad larger than 7mm × 7mm should also be subdivided in the copper itself with 0.15mm solder-mask dams. The dams give the perimeter wetting a stopping point and let the centre vent independently.

Component placement near thermal pads

Small components (0402 caps, 0603 resistors) within 1.5mm of a thermal-pad BGA can tombstone or be drawn off-pad by the asymmetric wetting forces of a high-mass solder joint. Push them out to 2mm minimum, or accept that you'll spend rework time on the line.

Documentation back to the design team

We send a void-rate report with every Class 3 lot — joint-by-joint percentage, signature classification, deviation from target. The design team uses this for the next revision's stackup discussion. That feedback loop is what turned one EV customer's BMS programme from 22% thermal-pad voiding on rev A down to 6% on rev C, with no change to the reflow line.

"Voiding is a manufacturing problem until it isn't. Past a certain point it's a layout problem, and the line lead is the wrong person to ask for the fix." — Pioneer Horizon DFM engineer

If you're tracking voiding numbers you don't like, send us your X-ray images and stack-up — we'll classify the signature, identify which lever is most likely to move it, and quote a corrective sample run within a week.