Keep the depth at 0.8 times the drill diameter or less, and do not treat the standard limit of 1:1 as a production target. In the accelerated testing that reliability papers keep quoting, microvias built at 1:1 failed after a few thermal stress cycles while structures at 0.7:1 or below finished the full test, on boards that carried 3,595 microvias each. The number to put on the drawing is 0.75:1. The 1:1 figure defines what a microvia is, it does not define what survives ten years in the field.
Where the ratio actually comes from
Aspect ratio is depth divided by drill diameter, and depth runs from the capture land foil to the target land, so the outer copper foil is part of the number. IPC-T-50M and IPC-6012 both cap a microvia at 1:1 with a total depth of 0.25 mm, and both count the foil in that depth.
That detail decides stack-ups. A 0.1 mm dielectric on 0.035 mm foil is 0.135 mm deep, so a 0.1 mm drill gives 1.35:1 and the structure is outside the definition of a microvia altogether. To land at 0.75:1 the drill has to open to 0.18 mm, or the dielectric has to come down to about 0.04 mm. Production runs sit at 0.075 to 0.1 mm drill through 0.06 to 0.08 mm of dielectric, which lands between 0.6:1 and 0.9:1 depending on which foil is used. The foil is the cheapest lever in the stack and the one that nobody checks.
Depth control is the other half of the same number. Prepreg resin flows during lamination, so the dielectric after press is not the catalogue figure, and a stack quoted at 0.08 mm often measures 0.07 mm in the cross-section. That is comfortable if the drill was sized from the pressed thickness, and a quiet 1.05:1 if it was sized from the datasheet.
Why the bottom of the via is where it fails
Copper does not plate evenly into a hole. Current density is highest at the surface and falls with depth, so the wall thins as it approaches the target pad. At 0.5:1 the difference is small. At 1:1 the copper landing on the target pad is a fraction of the surface thickness, and thin copper at the target is the defect the reliability studies keep pointing at, even on structures below 1:1.
The failure has a name in the trade: eyebrowing. The barrel stays attached on one side, the other side separates from the capture pad, and the arc of copper left behind looks like an eyebrow. It reads as an intermittent open, because the joint can still conduct when the board is cold and open up when it warms.
Thermal cycling is what drives it. Copper expands at 17 ppm per degree and resin above its Tg reaches 200 ppm per degree, so the barrel is anchored between two materials that move at very different rates. Every reflow pass and every field cycle pulls on the same two interfaces. A via at 0.7:1 has enough copper at both ends to absorb that movement, one at 1:1 does not. Stacked builds make it worse, which is why IPC issued a warning on high aspect ratio microvias in stacked arrangements: stacking three or four vias puts the weakest interface directly under the next via landing pad.
The pads matter as much as the hole
The target pad should be at least 80 percent of the drill diameter. A smaller target concentrates stress exactly at the plating interface the barrel is trying to hold, and it also leaves the laser less copper to strike, which changes the shape of the hole from a bowl into a cylinder.
On the capture side, keep at least 0.05 mm of annular ring per side at 0.1 mm drill. Laser position tolerance on a production line runs around 0.025 mm and the ring is what absorbs it. Below that the via breaks out of the pad on one side and the plating starts from a partial pad, which no amount of process control recovers.
The fill has its own ratio. A filled and capped via to IPC-4761 Type VII is what a via-in-pad needs at 0.4 mm pitch, and the cap copper meets the barrel at a second interface. Dimple is held to 10 micrometres or less on our line, and the cross-section has to show both the fill and the target pad in the same cut or neither number means much.
Three ways it goes wrong
Treating 1:1 as the target because it is the standard. The standard is a definition, not a process window. A board built at 1:1 will pass first article, pass functional test, and come back from the field a year later with an intermittent on one BGA corner. The process window for laser drilling and plating sits at 0.75:1 to 0.8:1, and the extra 0.02 mm of drill costs nothing.
Changing the dielectric and leaving the drill alone. Dielectric gets squeezed to fix impedance or to fit another layer, and the drill size on the drawing stays where it was. The ratio moves from 0.8:1 to 1.2:1 and nothing on the drawing shows it, because the drawing carries the two numbers separately. Ratio is the only field that has to be recalculated every time the stack changes.
Stacking on a target pad at the 80 percent limit. Each stage of a stack lands on the pad of the stage below, so the target of one via is the capture of the next. Take the smallest allowed pad at every level and the middle interface ends up with the least copper and the highest stress. Staggered stacks are more forgiving than stacked ones for exactly this reason, and when a stack has to be stacked, the pad grows before the drill does.
What SUPERB PCB supports
Laser drilling from 0.075 to 0.15 mm with a 0.05 mm floor, HDI from 1 to 3 stages, ELIC up to 30 layers, seven lamination cycles, layer to layer registration of 0.025 to 0.06 mm, plating of 20 micrometres or more, and via fill to IPC-4761 Type VII with a dimple of 10 micrometres or less. Cross-sections report the measured capture pad, target pad and pressed dielectric, so the ratio can be checked on hardware instead of on the drawing.
The boundary is worth stating plainly. The stack comes from the customer, and when the ratio lands above 0.8:1 it goes back as a DFM note for written approval. We do not alter a stack on our own, and we do not sign off a ratio we cannot hold in cross-section.
What to send with the stack
Pressed dielectric thickness and the foil used, drill diameter and whether it is the tool size or the finished size, target and capture pad diameters, number of stages and whether they are stacked or staggered, Tg and CTE, and the thermal cycling the product has to pass. With those six items the ratio can be settled before the first panel, which is the only point where changing it is cheap.