Tenting puts solder mask over the via pad and leaves the hole hollow. Plugging fills the hole with mask ink before the surface coat goes on. Filling puts epoxy in the hole and caps it with copper, and it is the only one of the three that gives a surface flat enough for a solder joint to sit on. For a 0.3 mm signal via on a standard board, tenting is enough and it costs nothing. The answer changes the moment the via sits inside a thermal pad, next to a BGA ball, or on a test point.
The practical rule: tent everything that is only a layer change, plug anything that could collect flux or cleaning chemistry, and fill and cap anything that has to be soldered over. Most arguments about vias come from a drawing that says tented while the shop reads plugged, or the other way round, so the first job is to name the type.
The states you can actually buy
IPC-4761 gives them numbers, and the numbers are what belong on the fabrication drawing. Type I is a tented via, mask over the pad, hole still open. Type II is tented and covered, with a second coat of mask on top. Type III is plugged, the hole filled with mask ink. Type IV is plugged and covered. Type V is filled with a non conductive epoxy, Type VI is filled and covered, and Type VII is filled, capped with copper and planarized. Only Type VII gives a flat pad, which is why every via in pad specification ends there.
Single sided protection is the case the standard lists as not recommended, and the reason is mechanical. One side covered with the other open turns the via into a cup. Flux, cleaning chemistry and moisture collect inside, and at reflow the trapped material expands and leaves through the only opening it has, which may be the solder joint sitting on top of it. Assemblies have lost QFN parts exactly this way, and the failure is read as a paste or placement problem for a week before anyone looks at the via.
Why tenting is not a seal
Liquid photo imageable mask is a liquid. Printed over a raised via pad with an open hole underneath, it flows into the hole during cure and thins at the pad edge, which is the pale ring inspectors call via yellowing. It is a property of the process, not a defect, and it does not mean the via is unprotected. What it does mean is that the tent is a skin measured in tens of microns, not a plug.
Tenting holds to about 0.4 mm. At 0.5 mm the skin over the hole becomes unreliable and starts to break, and above that the mask simply falls in. Two other limits matter: a tented via cannot be probed without opening the mask first, and a tented via larger than the mask dam beside it will take the dam with it when it breaks.
When plugging earns its extra step
Plugging fills the hole with mask ink, from one side or from both, to something like 98 percent of the barrel. It removes solder wicking through the via, it removes the yellow ring, and it keeps flux out of a hole that would otherwise act as a reservoir. It does not give a flat surface, so it is not a substitute for Type VII under a BGA ball, and if the ink is not fully cured the plug trades wicking for outgassing and ink bleed onto the pad next door.
Plugging is also the right answer for a via in a ground pad that is not soldered, and for any via within one mask dam width of a fine pitch pad, where a broken tent would leave an exposed ring beside a 0.4 mm pitch row.
Thermal vias are a separate decision
A tented thermal via array still conducts, because the mask is thin. Mask is 20 to 25 um thick with a thermal conductivity near 0.2 W per metre kelvin, which puts a square centimetre of mask at about 1 C per watt. For a 2 W part that is 2 C across the mask, and that is not what fails. What fails is the outgassing path: an array that is open on one side and sealed on the other traps flux and air under the thermal pad, and the void that forms is the one that shows up on the X-ray as a 30 percent void under a QFN.
The fix is to give the array a vent or to fill and cap it. Leaving the far side open is free and works if the board is not conformally coated afterwards. Filling and capping is what a high reliability build does, and it also removes the solder wicking that steals paste from the thermal pad.
Three ways the requirement gets lost
Writing tented and meaning plugged. Each shop has its own default for the word, so the same note produces a hollow tent at one supplier and an ink plug at another. Name the IPC-4761 type and add a section or a photo of the accept condition.
Tenting a 0.6 mm via because the drawing says all vias tented. The mask breaks over the hole during reflow or handling, the ring is exposed, and the via that was meant to be protected is now an open cup with a broken lid.
Probing a tented test point. The probe goes through the mask, the ink lands on the pad beside it, and the test passes while the assembly carries a loose flake of cured ink. Either open the mask on test points deliberately or move the test point to a pad that was never tented.
What SUPERB confirms before the panel is released
Via in pad fill is held to 10 um recess or less, hole wall copper at 20 um and above, mask bridge at 3.5 mil or more on green and 4 to 5 mil on other colours, laser vias from 0.075 mm, HDI at one to three stages, and BGA pads at 8 mil on ENIG. The via treatment is read off the drawing and confirmed back in writing, because Type III and Type VII differ by a plating and a planarization step rather than by a note, and the price difference between them is the reason the note gets quietly simplified.
The honest boundary: we do not decide the type for the customer. What we can do is say which types our process holds on a given hole size, stack and finish, and flag the vias on the panel where the specified type will not hold. That conversation is cheap before the first panel and expensive after. Capability is on HDI PCB capability, the general window on PCB capability, and the layer count side on multilayer PCB.