Where loss comes from
Insertion loss in a digital channel is the sum of four things: dielectric loss, conductor loss, reflections from impedance discontinuities, and radiation from stubs. HDI touches the last two.
A conventional through-hole via is drilled through the entire board. The part of the barrel below the layer where the signal enters is a stub: an open-ended transmission line that resonates at a frequency set by its length. On a 3 mm board a stub of 1.5 mm already resonates inside the band of a 10 Gb/s signal, and the usual fix is back-drilling — an extra process step with its own yield cost.
What a microvia removes
A laser microvia is 0.1 mm wide and connects adjacent layers only. There is no barrel running past the signal layer, so there is no stub to back-drill and no resonant notch. The path is also shorter and the return current stays closer to the signal, because the reference plane is one thin dielectric away rather than several.
Two secondary effects follow. Sequential lamination allows ground planes to sit between dense routing layers, which improves shielding between aggressor and victim nets. And the thinner dielectric raises the coupling to the reference plane, which reduces crosstalk for a given spacing.
What it does not fix
- Dielectric loss. Polyimide, FR-4 and high-frequency laminates differ by an order of magnitude in dissipation factor. No via strategy compensates for the wrong laminate.
- Copper roughness. At high frequency the current crowds into the surface of the foil; rough copper adds loss whatever the stack-up.
- Connectors and transitions. A pair of connectors and a cable can contribute more loss than the whole board.
- Impedance discontinuities. A microvia with a pad larger than the trace is still a discontinuity, and a 0.075 mm trace entering a 0.1 mm via needs its antipad and pad sized together.
Where the geometry has to be agreed
HDI dielectrics are thin — often thinner than the trace width needed for a 50 Ω line. A 0.075 mm trace on a very thin core can arrive well below the target impedance, so the stack-up is designed around the impedance rather than checked afterwards. That is why the impedance tolerance is stated as ±10 % and the stack-up is reviewed in DFM.
If the design also needs a fold or a connector tail, a rigid-flex build can keep the high-speed section on rigid material and use flex only where the geometry has to move. The process limits for both are on the capability page, and the stack-up options on the HDI product page.
