Most articles on HDI production list the steps and tell the reader to control them: watch the exposure time, watch the etch time, watch the hole size. That advice is unfalsifiable, because none of it carries a number. A real HDI build is governed by one cumulative quantity — registration error, which grows with every lamination cycle — and by three smaller budgets that interact with it: build-up dielectric thickness, laser drill quality, and via fill flatness. This article gives each of those a window, explains what fails when it is exceeded, and shows why the choice between stacked and staggered microvias is a yield decision rather than an aesthetic one.
What the Original Version Left Out
The published article ran to twelve headings and about 770 words, and contained no tolerance, no window and no failure mechanism. It also made three claims that would mislead a buyer:
| Original statement | Correction |
|---|---|
| Overly small hole diameters may lead to poor signal transmission | Backwards. A smaller via is a smaller discontinuity and a shorter stub; the limit on hole diameter is drill and plating capability, not signal quality |
| Soldering temperature and time are considerations in HDI production | Assembly is the customer’s process. What a fabricator controls is the finish and its thickness window; the reflow profile is set downstream |
| Control exposure time, development time and etch time | True but unactionable without windows. For fine lines the etch factor and the resulting line width tolerance matter more than time, and the number that has to be held is typically ±10 µm on a 75 µm line |
What the original omitted is the one quantity that dominates HDI yield: registration. A conventional board registers once, at drill. An HDI board registers at every lamination cycle, and each cycle adds error that the next one inherits.
The Process, With a Number at Every Step
The sequence for a 1+N+1 or 2+N+2 build is fixed. What varies between fabricators is not the sequence but the windows.
| Step | What is controlled | Typical window | What fails when it is exceeded |
|---|---|---|---|
| Core drilling and inner-layer imaging | Core via position, line width | ±25 µm position; ±10 % line width | Inner-layer short or open after lamination shift |
| Core lamination | Layer-to-layer registration, thickness | Class 3 registration per IPC-6012; thickness ±10 % | Impedance drift, layer misregistration |
| Build-up lamination | Resin flow, dielectric thickness, registration | 60–80 µm finished, ±10 % | Impedance shift, resin starvation at edges, warpage |
| Laser drilling | Hole diameter, taper, bottom copper remaining | 100 µm ±15 µm; 10–15 µm copper left at the target pad | Pierced target pad, open microvia, or a blind hole that will not plate |
| Desmear | Resin smear removal | Full removal without attacking the dielectric; permanganate for epoxy, plasma or sodium naphthalenide for PTFE | Barrel-to-target separation, intermittent connection that passes test and fails in the field |
| Electroless copper | Seed layer coverage in a blind hole | 0.3–0.5 µm, complete coverage to the bottom corner | Void at the target interface |
| Via fill and planarisation | Fill voiding, copper recess | Void <2 % of cross-section; recess ≤25 µm | Solder wicking into the via, or a pad that is not coplanar for a 0.4 mm ball |
| Outer imaging and etch | 75 µm lines at 75 µm space | ±10 µm; etch factor compensation applied | Impedance out of tolerance, shorts at the fine pitch |
| Solder mask and finish | Mask on 75 µm traces, dam width | Dam ≥0.1 mm where required; finish per IPC-4552 and related | Mask lift, finish bridging, solder balling |
Read down the “what fails” column and a pattern appears: almost every failure is either a registration error or an interface defect. Neither is visible at electrical test, and neither is caused by the operator watching a clock.
Registration Is a Running Total, and Stacked Microvias Spend It
Each lamination cycle carries its own registration error, typically ±25 µm at the layer-to-layer level for a controlled build. A 1+N+1 board has two build-up cycles and one core cycle; a 2+N+2 board has four build-up cycles. The errors do not average out — they add, and the microvia has to land inside a capture pad of about 250 µm for a 100 µm hole, which leaves roughly 75 µm of annular ring on each side.
That is the arithmetic behind the stacked-versus-staggered decision. A stacked microvia must align vertically with the one below it, so the errors of both cycles consume the same 75 µm of ring; a staggered microvia is offset by at least 50 µm and lands on its own pad, so each layer only has to meet its own tolerance. The yield difference is real and repeatable: stacked fields typically run 91–94 % against 96–98 % for staggered, before any reliability testing.
| Via configuration | Registration requirement | Typical yield | When it is the right choice |
|---|---|---|---|
| Staggered | Each layer to its own pad; ≥50 µm offset between levels | 96–98 % | Default for 2+N+2 and above unless density forces otherwise |
| Stacked | Vertical alignment of all levels, ±25 µm cumulative | 91–94 % | Ultra-dense escape, or a thermal path that must be continuous |
| Skip via | Lands two build-up layers down; aspect ratio ≤0.8:1 | 89–92 % | Removing one landing pad and one routing layer |
Two practical rules follow. Ask for staggered vias where the density allows and expect to pay for stacked only where the escape or the thermal path requires it. And when a design does use stacked microvias, specify copper-filled and capped construction — an unfilled stack is a column of voids waiting for the first thermal cycle, because each interface is a place where plating can fail to bond.
Dielectric Thickness: The Variable That Moves Impedance
Build-up dielectric is supplied as resin-coated copper or as prepreg, and after lamination its finished thickness depends on resin flow, copper balance and press profile. On a 60–80 µm layer held to ±10 %, a 50 Ω microstrip moves by about ±6 % — which consumes most of a ±10 % impedance window on its own, before etch variation and Dk tolerance are counted.
Two things make this worse in an HDI build specifically. Copper balance matters more because the layers are thin: a region with 80 % copper and a region with 30 % copper laminate to different thicknesses under the same pressure, so impedance varies across the panel rather than from panel to panel. And the resin system itself is lossier than the laminates used for RF: build-up films are typically Dk 3.3–3.6 with Df in the 0.015–0.025 range, against 0.002–0.004 for a low-loss microwave laminate. An RF line built on a build-up layer therefore loses several times what the same line loses on the core, which is the main reason microwave routing is kept on core layers in a hybrid stack.
The control that works is not tighter press pressure but a coupon. Put impedance coupons on the same panel, in a copper-balanced area, and measure them after lamination and again after finish. A fabricator that controls thickness by recipe and verifies by coupon will hold ±10 %; one that controls by recipe alone will hold it on average and miss it on the edges.
Laser Drilling: An Energy Window, Not a Setting
A CO2 laser removes epoxy by ablation and stops on copper, which is why the process needs a target pad and why 10–15 µm of copper must remain on it. The window is narrow in both directions: too little energy leaves resin at the bottom and the hole will not plate; too much pierces the target pad and creates a blind hole into the layer below, which is an open circuit that electrical test may not catch if the pad is still connected elsewhere.
Hole shape matters as much as diameter. A CO2-drilled microvia is conical, with the bottom diameter typically 70–80 % of the top; the plating has to cover that taper, which is why the aspect ratio limit is set on the bottom diameter, not the top. UV lasers produce a straighter wall and are preferred for 75 µm holes and for anything below 60 µm dielectric.
The material interaction is where PTFE departs from FR-4. PTFE does not absorb the CO2 wavelength efficiently, melts rather than ablates, and leaves debris that is difficult to remove; UV or excimer drilling with plasma desmear is the practical route, and the 237 ppm/°C Z-axis expansion of PTFE loads the barrel far more than FR-4 during any subsequent thermal excursion. For a 0.254 mm microwave dielectric the geometry also rules out HDI microvias entirely: a 1:1 aspect ratio would need a 254 µm hole, which is outside the microvia definition.
Desmear and Metallisation: Where Intermittent Opens Are Born
Laser drilling leaves a resin smear at the bottom of the hole and along the barrel. If it is not removed, the electroless copper seeds on smear rather than on the target pad, and the connection is mechanical rather than metallic — it passes continuity test, survives a few thermal cycles, and then opens. This is the most expensive defect class in HDI because it is invisible at the gate and appears in the customer’s reliability test.
The chemistry is material-specific and not interchangeable. Epoxy build-up films are desmeared with permanganate; polyimide needs a stronger permanganate or plasma; PTFE cannot be desmeared by permanganate at all and requires sodium naphthalenide or plasma. A mixed PTFE and FR-4 build therefore needs a sequence that serves both, or the PTFE layers are drilled and treated in a separate operation.
After desmear, electroless copper has to seed the bottom corner of a blind hole where the electrolyte exchanges poorly. The window is 0.3–0.5 µm of seed, complete to the corner, followed by pattern plating to the finished thickness. For a filled and capped via the plating continues until the hole is closed, then the surface is planarised; recess above 25 µm leaves a pad that is not coplanar, and a 0.4 mm ball placed on a recessed pad wicks solder into the via.
Warpage: A Thin Board Is a Flexible Board
Bending stiffness scales with the cube of thickness, so a 0.6 mm HDI board has about 5 % of the stiffness of a 1.6 mm board of the same construction. Warpage after each lamination cycle is therefore a process output that has to be managed rather than a defect that can be inspected out, and the practical window is 0.2–0.3 % of the diagonal — about 0.3 mm on a 130 mm panel.
Three controls do most of the work: a symmetric stack (same number of build-up layers, same copper weight and same dielectric thickness on both sides), balanced copper distribution per layer, and a press profile that lets the resin cure before the panel is released. Asymmetric builds are the most common cause of a panel that measures flat after lamination and bows after reflow.
What to Inspect, and Against Which Document
HDI acceptance is not covered by a single number in IPC-6012; the HDI-specific requirements sit in IPC-6016, with IPC-6012 Class 3 governing the rest of the build. The inspections that actually catch the failure modes above are few:
| Check | Method | Criterion | Failure it catches |
|---|---|---|---|
| Microvia cross-section | Microsection, 3–5 vias per panel | Void <2 % of cross-section; barrel copper per IPC-6016; target interface continuous | Smear-induced separation, plating void |
| Layer registration | X-ray of registration targets | Within the build’s stated tolerance | Stacked via misalignment |
| Impedance | TDR on coupons from the same panel | Per drawing, typically ±10 % | Dielectric thickness or etch drift |
| Via fill flatness | Profilometry or cross-section | Recess ≤25 µm | Solder wicking, non-coplanar pads |
| Thermal cycling | Air-to-air −40 to +125 °C, or IST | Typically 500–1,000 cycles without resistance change beyond 10 % | Barrel fatigue, interface separation |
The cross-section is the only check that finds the smear defect before the customer does, and it samples a handful of vias out of millions. That is why the process windows in this article matter more than the inspection plan: at HDI densities, inspection is a confirmation, not a screen.
Eight Numbers for the Purchase Order
An HDI order that carries these eight items can be built, argued about and rejected. One that carries adjectives cannot.
1. Build notation (1+N+1, 2+N+2) and total layer count. 2. Microvia diameter, dielectric thickness and aspect ratio per build-up layer. 3. Via configuration: stacked, staggered or skip, with the fill and cap requirement stated. 4. Minimum line and space on each layer. 5. Registration tolerance and the inspection method. 6. Impedance targets with coupon geometry and tolerance. 7. Warpage limit after each lamination and after finish. 8. Acceptance document: IPC-6016 with IPC-6012 Class 3, plus the finish specification.
Conclusion
HDI production is not twelve things to watch; it is one cumulative budget and three local ones. Registration accumulates at roughly ±25 µm per lamination cycle and is spent by the microvia landing on its pad, which is why stacked vias cost 4–6 percentage points of yield against staggered ones and should be used only where density or thermal continuity demands them. Dielectric thickness moves impedance by ±6 % on a ±10 % thickness window, and is controlled by copper balance and verified by coupon rather than by recipe. Laser drilling is an energy window bounded by resin left at the bottom on one side and a pierced target pad on the other. And desmear, which has no visual signature, is where the intermittent opens come from — with a chemistry that changes completely when the laminate is PTFE rather than epoxy.
XCE PCB runs HDI and mixed-dielectric builds under those windows: 2 to 72 layers, 1+N+1 and 2+N+2 with staggered or stacked microvias, impedance ±10 %, panels to 600×1200 mm, samples in as fast as 24 hours, inspected to IPC-A-600G, IPC-6015 and IPC-6016. Tell us the build notation and the via configuration and we will return the registration budget with the quote, because that number decides the price more than the layer count does. The electrical side of the same decision is in HDI performance and functions, and lamination control in lamination for high-frequency boards.
Frequently Asked Questions
What are the critical steps in HDI PCB manufacturing?
Why are stacked microvias less reliable than staggered ones?
How does laser drilling affect microvia reliability?
What causes intermittent opens in HDI boards?
How much warp is acceptable on an HDI board?
Which standard covers HDI acceptance?
Need an HDI Build With a Registration Budget?
Send us your Gerber files and requirements for a manufacturability review and a tailored quotation.
sales1@xcepcb.com
+86 180 3341 4657