IPC-A-610 Class 3 changes three things on a QFN: it withdraws the process indicator allowances that Class 2 leaves open on the side termination, it makes the thermal pad joint something you have to prove rather than look at, and it turns the acceptance record into part of the deliverable. What it does not do is set a void percentage. That number comes from IPC-7093 for bottom termination components, or from your drawing, and most arguments at incoming inspection are really about which document the purchase order called out.
Two builds can leave the same line with the same stencil, the same paste lot and the same reflow profile, one accepted under Class 2 and one rejected under Class 3. Nothing about the process changed. Only the criteria applied to identical joints changed.
Where the two classes part company on a QFN
| Criterion | Class 2 | Class 3 |
|---|---|---|
| Side termination, wettable flank | A bottom only joint is accepted where the flank shows no fillet | The flank is expected to be wetted, because that fillet is the only visual evidence the joint formed |
| Thermal pad void | Judged against the drawing or IPC-7093 | Same source, but every void has to sit inside the limit with no path from the outside edge into the joint |
| Offset and overhang | Allowed within the stated share of the termination | Tighter, measured on the bottom termination where the joint actually forms |
| Process indicators such as dewetting or a grainy fillet | Accepted as indicators under several criteria | Not accepted |
| Acceptance records | Internal | Documented, revision controlled, traceable /td> |
Read the table as a direction, not as a replacement for the document. Clause wording moves between revisions of the standard, and the revision named on the drawing is the one that settles an argument.
Why the thermal pad decides the outcome
A 0.5 mm pitch QFN has terminal pads about 0.25 mm wide, and the ground pad in the middle is several millimetres square with a via cluster underneath. Two different paste problems live on the same part. On the perimeter the limiting number is the area ratio: a 0.25 by 0.25 mm aperture in a 0.12 mm stencil gives an area ratio near 0.52, well under the 0.66 that transfers paste reliably, so the perimeter needs a thinner or a stepped stencil. On the ground pad the risk runs the other way. Too much paste and the body floats on a paste bed, tilts, and squeezes solder out to the perimeter; too little and the voids collect at the pad edges, where the thermal path is thinnest.
Voiding is where the two documents separate. A 2D X-ray of a QFN thermal pad shows the sum of every void through the thickness, so a ring of small voids at the die edge and one large void in the middle look alike on the image and behave nothing alike in service. A void in the middle sits under the hottest part of the die. A void ring at the edge breaks the seal between the pad and the outside and becomes a path for flux and moisture. If the drawing says 25 percent total void, a CT slice has to show where that 25 percent sits, not just how much of it there is.
Three things that go wrong with a Class 3 callout
Quoting a void figure as if the standard published it. Buyers often write 25 percent into the purchase order and then describe it as an A-610 Class 3 requirement. The standard governs workmanship and acceptability of the joint; the void limit for a thermal pad comes from IPC-7093 or from the customer drawing, and the two do not always agree. When they disagree, the drawing wins, and it is worth saying so in writing before the first panel is drilled.
Expecting AOI to prove a QFN joint on a part without a wettable flank. A standard QFN with a straight cut flank gives the camera nothing to measure, so the joint is either proven by X-ray or not proven at all. If the program needs optical proof at volume, the part has to be ordered with a wettable flank, and that decision belongs in the BOM, not in a conversation after the first build.
Applying Class 3 to a whole board for one section. A mixed assembly can carry one assigned area at Class 3 while the rest of the board stays at Class 2. Stamping Class 3 across a high volume board to protect one RF section is a cost decision dressed up as a quality decision, and it usually shows up as unnecessary rework on joints that were never at risk.
Rework deserves one line of its own. IPC-7711 covers QFN removal and replacement with preheat and a nozzle matched to the body, and a reworked part ends up with a fillet that differs from the parts around it. Class 3 does not ban rework; it requires the rework to be recorded and the joint to meet the same criteria afterwards. A build that reports zero rework on 40 touch-ups is the one that fails an audit, not the one that shows the record.
How SUPERB supports a Class 3 QFN build
SUPERB Assembly runs the first article to IPC-A-610 Class 3 where the drawing calls for it, with 3D SPI before reflow, 3D AOI and DIP AOI after, X-ray and CT to IPC-7095 for thermal pad voiding, and rework held to IPC-7711 and IPC-7721. Seven SMT lines in a 5,000 square metre ESD controlled area take components from 01005 up to 57 mm at 40,000 CPH, with twelve zone nitrogen reflow below 500 ppm oxygen, so a 0.4 mm pitch QFN sees the same profile on the tenth panel as on the first. Where the ground pad needs it, the PCB side supports via in pad, plugged and filled vias, and HDI stack-ups with 0.30 mm pitch fanout through the PCB capability range.
Records are keyed to the serial and to the panel, not to the shift: first article report, SPI and AOI program revisions, X-ray images, rework log and the test data from ICT, flying probe and functional test on a fixture built for the product. The system is certified to ISO 9001, with ISO 13485 for medical work, and components carry a 730 day warranty. Builds run build-to-print against the quality standards named on your drawing, without an engineering fee.
If a QFN is being rejected at incoming inspection and the argument is about criteria, send the drawing revision, the standard revision it names and the X-ray images. The reply states which rejections are real, which are the wrong document being applied, and what the paste and stencil setup would need to change to hold the limit at volume.