Sunking PCB

Multilayer PCB Via Types: Your Connection Strategy Decides the Stackup

Design Guide2026-10-22Sunking Circuits Technical Team
Multilayer PCB Via Types: Your Connection Strategy Decides the Stackup

Via types are not four freely mixable options

When a multilayer board is discussed, layer count is almost always the first number spoken, and interconnect is treated as a later detail: decide four layers or eight, then "figure out the holes." The real order is the reverse.

Through-holes, blind vias, buried vias and back drilling look like four items on one menu, to be ticked according to design need. On the manufacturing side they are not the same class of action: a through-hole is a single subtractive pass across the whole board; a buried via requires drilling and metallisation completed before lamination; a blind via must enter from an outer layer and stop at a specified layer; back drilling removes material a second time from an already finished through-hole. Each one corresponds to a different position on the process timeline. The moment one via type requires metallisation before lamination, that board cannot be pressed only once — not as a matter of process preference, but because no other ordering of operations exists.

So the stackup is not drawn first and populated with holes afterwards. Connection intent is fixed first, the process route follows from it, and the stackup can only be released within what that route permits. Design sees layer count and routing space; manufacturing sees "how many times this board gets pressed, which layers in each press, and which holes are drilled before which press." Both are discussing the same board, but the constraints come from the second view.

What each via type costs manufacturing

Through-hole: passes from one face to the other and connects every layer it crosses. Its process position is the simplest — drill and metallise after the full board is laminated. The cost sits elsewhere: it consumes a pad and keep-out on every layer it passes, even when only the two outer layers need connecting. The more layers and the denser the routing, the more expensive that whole-stack footprint becomes. But it is expensive in routing area, not in process steps.

Buried via (between inner layers, reaching no outer layer): drilling and metallisation must be finished while those layers are still an independent sub-board, and only then is that sub-board pressed into the full stack. This means the presence of a buried via splits lamination into at least two stages. What it buys is inner-layer connection that consumes no outer-layer space.

Blind via (enters from an outer layer, stops at a specified inner layer): must enter from the outer side and terminate at depth. Its process cost depends on the layer pair it spans. If it falls within the range that can be machined directly after the final press, it adds no lamination stage. If its termination layer sits deeper and is incompatible with the layer pairs of other via types, it will pull in another lamination stage with it.

Back drilling: strictly speaking not a connection but the local cancellation of one. After the through-hole is metallised, the unwanted section of plated barrel is removed from one side. It does not change the number of lamination cycles, but it introduces depth control as a separate process capability requirement, and its acceptance object is what remains after removal, not how deep the tool went. The stub judgement logic and acceptance method are covered in back drill stub control and acceptance.

Placed side by side, the real distinction between these four is not "which is harder" but which one demands action before lamination. That single line splits the cost structure in two: items that only affect routing area (through-holes), and items that affect the number of process stages (buried vias, and blind vias at certain layer pairs).

Which combinations force sequential lamination

Multiple lamination cycles are not an option you tick separately. They are the consequence of certain via-type combinations. The test is not complicated: list the layer pair each via type connects, then ask whether any connection must be completed before lamination and whether those connections can be assigned to the same sub-board stage.

  • Through-holes only: one lamination. Every hole is machined after the full board is pressed.
  • Through-holes plus shallow blind vias: usually still one lamination, with the blind vias machined from the outer side afterwards. This is the common case of "a little more process capability required, but the lamination structure unchanged."
  • Any buried via present: at least two laminations. The sub-board carrying the buried via is drilled and metallised on its own, then pressed with the remaining layers.
  • Several non-nested buried via layer pairs: the number of lamination stages grows with how the layer pairs nest. If they nest — the inner pair contained within the outer pair — stages can be accumulated level by level. If the pairs cross each other and neither contains the other, then no lamination sequence can satisfy them simultaneously. That is not a problem money solves; the design has to change.
  • Blind via and buried via layer pairs in conflict: a blind via requires a layer interface to remain machinable from outside after pressing; a buried via requires it to be finished before pressing. When both point at the same interface, lamination splits further.

The level structure of sequential lamination and its cumulative constraints have their own article, HDI sequential lamination buildup level constraints; registration control within lamination itself is covered in multilayer lamination registration control. What this article adds is the step upstream of both: the number of lamination cycles is a function of the via layer pair table, not an independent decision.

What design freedom a wrong layer pair destroys

The layer pair — the start and end layer a via connects — is the input to the entire process route. Once it is fixed, several downstream freedoms disappear, and the design side usually discovers this late.

First, stackup symmetry becomes constrained. Sub-board division comes from the buried via layer pairs, and once sub-boards are defined, dielectric thickness and copper distribution can no longer be adjusted freely: each sub-board needs a manufacturable structure of its own, and the full board still has to stay balanced. Symmetry requirements and their effect on warpage are covered in multilayer symmetry and buildup requirements.

Second, dielectric thickness is no longer free. A blind via's termination layer determines the dielectric combination it must cross, and that combination also has to satisfy impedance and availability. The set of genuinely available core and prepreg combinations is already finite — see core and prepreg availability constraints. After layer pairs are fixed, the viable set shrinks further, sometimes to empty.

Third, layer function assignment gets locked. Which layer serves as a reference plane and which carries signals normally has room to move. Once buried vias bind several layers into one sub-board, adjusting layer function across sub-boards means rearranging the process route. The manufacturing-side requirements are covered in multilayer layer function assignment.

Fourth, the cost of a late revision stops being local. On a through-hole-only board, moving one trace usually affects that layer alone. On a board with buried vias, changing a connection that crosses sub-boards may require re-dividing the sub-boards, which is a new process route.

These four losses do not scale linearly with layer count. They jump at the single step of "introducing the first buried via." Which makes one question worth asking before release: do these inner-layer connections genuinely need buried vias, or can they be traded for through-holes that cost more area but fewer process stages?

How far the drawing must declare connection intent

A board shop does not need "how many blind vias." It needs connection intent from which the process route can be reconstructed. Omit any of the items below and the shop can only sequence operations according to its own default reading — and default readings differ between shops.

  • The start and end layer for every via type: written as layer numbers, not as the words "blind" or "buried." The layer pair table is the only input to the process route.
  • Functional distinction per hole: electrical connection, thermal, mechanical fixing or tooling. Their acceptance criteria differ. Plugging and pad treatment for thermal vias are covered in thermal via arrays and plugging.
  • Whether manufacturing may adjust the layer pair: if a given pair is only a means of implementation rather than a hard constraint, saying so lets the shop propose an equivalent with a shorter route. Leave it unsaid and the shop simply builds what is drawn.
  • Plugging and fill requirements: which holes must be filled, the surface condition required after filling, and whether capping is needed. The acceptance basis for resin fill is covered in resin filled via acceptance.
  • The target and acceptance object of back drilling: declare which section is to be removed and what is inspected, not just a depth.
  • Hole wall quality and inspection method: inspection methods for wall condition are covered in drilled hole quality inspection, and the plating capability boundary for laser microvias in laser drilled microvia plating and aspect ratio.
  • A priority declaration: when layer pairs, dielectric availability and impedance cannot all be satisfied, state which one yields first. This is almost never written, and it is what the shop needs most.

The full release checklist is covered in multilayer stackup drawing release checklist. Product scope and buildable structures are on multilayer PCB and manufacturing capabilities, with tolerance definitions on tolerances. The layer count trade-off itself is covered in 4 vs 8 layer buildup planning and the multilayer stackup guide.

The sequence to follow before release

Compressing the sections above into something executable:

  1. List every required inter-layer connection as layer pairs only, with no implementation named.
  2. Check whether any layer pairs cross without nesting. If they do, change the design first.
  3. Assign as many connections as possible to through-holes, trading routing area for process stages.
  4. For what genuinely requires buried or blind vias, derive the lamination stage count from the nesting relationship.
  5. Only then choose dielectric combinations and layer function assignment within that lamination structure, not the other way round.
  6. Put the layer pair table, functional distinctions, fill requirements and priority declaration into the drawing.

The value of this order is that "it cannot be built" surfaces before release rather than after the first prototype.

FAQ

Q: Why can't the stackup be fixed first and the holes decided afterwards?

A: Because a buried via requires drilling and metallisation before lamination, which directly sets how many lamination stages are needed. Only once the stage count is fixed do the available dielectric combinations and layer function assignments have boundaries. Drawing the stackup first commonly ends with the stackup being redone.

Q: Is a blind via always more expensive than a through-hole?

A: Not necessarily. They are expensive in different currencies. A through-hole is expensive in routing area, because it occupies a footprint on every layer it crosses. Blind and buried vias are expensive in process stages. On a board with loose routing, through-holes are the better deal; only a space-constrained board is worth trading process steps for area.

Q: Can several buried via layer pairs be done if we pay more?

A: It depends on the relationship between the pairs. Nested pairs can accumulate lamination stages level by level. Pairs that cross each other with neither containing the other cannot be satisfied by any lamination sequence. That is not a cost question; the design has to change.

Q: Does back drilling add lamination cycles?

A: No. Back drilling happens after through-hole metallisation and does not change the lamination structure. What it introduces is depth control as a separate process capability requirement, and its acceptance object is what remains after removal.

Q: Is writing "blind via" on the drawing enough?

A: No. The shop needs a layer pair table given as start and end layer numbers, because the layer pair is the input to the process route. With only the via type named, the shop can only sequence operations by its own default reading.

Q: What if layer pairs, dielectric availability and impedance cannot all be met?

A: Declare the yield priority in advance on the drawing. The shop can see the conflict but cannot decide on your behalf which requirement to sacrifice. Without a priority declaration, the back-and-forth consumes the whole engineering window.

Want this stackup and HDI process checked?

Layer count, blind/buried vias and impedance control drive both yield and cost. Send us the stackup and we will return DFM feedback with optimisation options.