Every panel leaves the press carrying three accounts that were opened before it went in: a position, a thickness and a stress. The position decides whether a 100 µm microvia lands inside its pad; the thickness decides whether 50 Ω is 50 or 53; the stress decides whether the board is flat at assembly or warps in the reflow oven. Everything the factory does after lamination — drilling, plating, imaging, finish — is spent paying those three accounts down, and nothing downstream can close one that is already overdrawn. This is a comprehensive analysis of the barriers specific to multilayer lamination on high-frequency material: what each account contains, what the numbers look like, and which ones a design can settle before the first panel is cut.
Three Accounts, Not One Problem
Lamination is usually discussed as a bonding step, and on FR-4 that framing is adequate: heat, pressure, resin flow, done. On high-frequency material it is not, because the material is expensive enough that a laminated scrap is a serious loss, and because three independent error families are fixed at the same instant.
The first is position: every layer has to arrive at the drill station within a tolerance of every other layer, and the error is cumulative across press cycles. The second is thickness: the dielectric height that the designer assumed has to be the height the press actually produces, including the bondply that was pressed into it. The third is stress: the panel cools from 200 °C or 300 °C to room temperature with materials that contract at different rates, and whatever cannot relax stays in the board as bow, twist and stored strain around every hole.
The reason to separate them is that they are owned by different people and fixed by different decisions. Position is a tooling and pinning question, thickness is a stack-calculation question, stress is a stack-symmetry and cooling question. A factory that treats lamination as one problem optimises the press recipe and leaves two of the three accounts untouched.
How This Differs From the Lamination Overview
The step-by-step view of lamination — how the dielectric height is set, how voids form, how the three bonding systems differ — is set out in lamination in high-frequency PCB manufacturing. That article answers “what does the press do to the impedance”. This one answers a narrower question: what is different when the board has four, eight or twelve layers instead of two, and what does the difference cost.
| Concern | Two-layer board | Multilayer, one press cycle | Multilayer, two cycles (PTFE + thermoset) |
|---|---|---|---|
| Registration | Not applicable | One accumulation of error | Two accumulations, roughly doubled |
| Thickness | Core only | Core + pressed bondply, one interface | Two interfaces, second press re-flows the first |
| Stress | Low, symmetric by nature | Depends entirely on stack symmetry | Highest — two thermal histories, two contraction events |
| Scrap value | Low | High | Highest, and the loss happens at the last step |
Account One: Position, and Why It Grows With the Layer Count
Registration error accumulates because each press cycle and each imaging step contributes its own offset, and the offsets add rather than cancel. A practical planning figure is ±25 µm per lamination for the material movement, plus the imaging and drilling errors on top. On a single-cycle four-layer board that is one term; on a three-cycle twelve-layer hybrid it is three.
| Source of error | Magnitude | Depends on |
|---|---|---|
| Prepreg cure shrinkage | 0.02–0.05 % → 0.12–0.30 mm on a 600 mm panel | Resin content, warp vs weft orientation |
| Thermal expansion during the cycle | 0.31 mm on 600 mm for TLY-5 over 20 K, 0.37 mm for 5880 | Material X/Y CTE, panel size |
| Tooling and pin fit | ±25 µm typical for a four-slot pin set | Pin condition, slot wear |
| Press platen parallelism | Small but systematic, tilts one side | Maintenance history |
The budget is easiest to see on a microvia. A 100 µm via on a 250 µm pad has 75 µm of ring on each side when everything is nominal. Two press cycles consume two thirds of that before drill wander and image distortion are counted, which is why stacked microvias yield worse than staggered ones and why a design that can be staggered should be. The same arithmetic explains a rule that is often stated without its reason: on PTFE, keep the aspect ratio at 6:1 or better, because deep small holes amplify every positional error at the bottom of the barrel. The hole-design side of that rule is in PCB through hole.
Account Two: Thickness Is Not the Number on the Datasheet
The dielectric height in an impedance model is the distance between two copper layers after pressing. The laminate datasheet gives a core thickness; the bondply datasheet gives a prepreg thickness; the press delivers something else, because the bondply flows and consolidates. On a single interface the difference is small. On a multilayer it appears at every interface, and the errors accumulate in the same direction.
| Stack element | Nominal | What the model must use | Consequence if ignored |
|---|---|---|---|
| Core | 0.254 mm | 0.254 mm ±10 % | ±6 % impedance |
| Bondply, one layer | 0.10 mm | Thickness after flow, not before | 0.10 mm missed = 50.0 → 61.1 Ω |
| Resin flow | 15–25 % of the bondply | Redistribution, not loss — it fills the copper gaps | Local h variation over dense copper |
| PTFE cold flow | Up to 4 % thinning | A permanent reduction, not elastic | 50.0 → 48.8 Ω |
The resin-flow row is the one that catches experienced people. Prepreg does not disappear when it flows; it moves into the spaces between traces. Over a dense copper area the copper consumes most of the resin and the dielectric ends up thinner than over a bare laminate area, which means the same stack can produce two different impedances on the same layer depending on what is under it. The countermeasure is to balance copper distribution across each layer at design time — not to ask the press to fix it.
Two further thickness terms are worth stating because they are frequently omitted from the model. A buried stripline is more sensitive than a microstrip to the same percentage error, because the field is bounded on both sides: a ±10 % variation on a 0.5 mm buried dielectric moves the impedance further than the same percentage on a 0.254 mm microstrip. And on PTFE the height moves with temperature in service: 237 ppm/°C on 5880 means a 0.254 mm dielectric grows about 3 % between 20 and 150 °C, which is roughly +1.9 % on impedance — a real effect, and one that a coupon measured at room temperature cannot show.
Account Three: Stress Is What Remains After Cooling
The panel is bonded at temperature and then cooled. Every material in it contracts at its own rate: copper at 17 ppm/°C, RO4350B at 32 in the Z direction, RO4003C at 46, PTFE at 217–237. Because the layers are bonded, they cannot contract independently, so the difference is stored as elastic strain and as curvature.
Three consequences matter. Bow and twist appear when the stack is not symmetric about its centre line: an unbalanced stack with more copper on one side will curl, typically by 0.2–0.3 % of the diagonal, which is enough to fail a flatness requirement on a large panel. Stored strain around holes is what makes a barrel crack later: the copper is already stretched by the Z expansion and has no reserve when thermal cycling adds more. Dimensional change is what breaks registration on the next layer if the board is imaged after lamination rather than before.
The controls are structural rather than procedural. Make the stack symmetric in copper weight and in dielectric thickness about the centre. Keep the high-CTE material in the middle rather than at the surface, so that its expansion is constrained by both neighbours. Cool under pressure until the panel is below the temperature at which the bond is set — opening the press early leaves the stack free to warp while it is still soft. And do not rely on Tg as a proxy for any of this: PTFE has no useful Tg, and its dimensional behaviour is governed by its melt point near 327 °C and by its CTE, not by a glass transition.
One Cycle or Two: What the Hybrid Actually Costs
The single largest barrier in multilayer high-frequency work is not a process detail, it is the number of press cycles the stack requires. That number is set by chemistry, and it is decided on the stack drawing.
| System | Bonding temperature | Can it share a cycle with the others? | Where it is used |
|---|---|---|---|
| FR-4 / epoxy | 170–190 °C | With RO4000, no; RO4000 needs 190–200 °C | Digital, control, power layers |
| RO4000 + 4450B | 190–200 °C | With itself, yes | Antenna, radar, base-station RF |
| PTFE + FEP film | 260–300 °C | No — destroys cured epoxy if pressed after it | Millimetre wave, low-loss cores |
| PTFE + low-flow prepreg | 180–190 °C | Sometimes, at the cost of lower peel strength | PTFE stacks that must bond to FR-4 |
The order matters as much as the number. When PTFE and FR-4 appear in one board, the PTFE sub-assembly is pressed first at high temperature and the FR-4 is bonded to it second at low temperature — the reverse order destroys the cured epoxy. That means the second cycle re-heats a finished PTFE assembly, which re-opens the thickness account (a second flow event) and adds a second registration term. Designs that can keep the PTFE to the outer layers and the thermoset in the centre pay one penalty instead of two, which is the main reason that construction is so common on hybrid antenna boards.
The Stack Calculation That Prevents Most of It
Most lamination scrap on high-frequency work is arithmetic that was never done. The calculation is short, and it has to be done before the laminate is ordered, not after the first coupon comes back out of tolerance.
Step 1. Fix the target impedance per layer and the frequency at which it will be measured. Step 2. Choose the dielectric height the model needs, not the height that is convenient to stock. Step 3. Subtract the pressed bondply thickness from the dielectric height to obtain the core thickness, and check that the core is a stocked item; if it is not, change the geometry rather than the order. Step 4. Compute the line width from the chosen Dk — process Dk for building, design Dk for what the customer’s simulator uses, and they are not the same number. Step 5. Check the resulting width against the minimum the etcher can hold at that copper weight, because a 0.075 mm line at 2 oz is not a tolerance question, it is an impossibility. Step 6. Check the aspect ratio of every hole through the finished stack. Step 7. Check that the stack is symmetric.
Failing to do step 3 is the single most common error in the data packages that reach a fabricator: the customer calculates impedance on the core thickness alone. The error is systematic, always in the same direction and often larger than every manufacturing tolerance combined. The data that must accompany the stack is described in mixed-pressure PCB design files, and the design-side sequence in stack-up design steps.
Voids: The Error That Hides Inside the Average
A void in the bondply is not a cosmetic defect. It is a series air layer of Dk 1.0 in the middle of the dielectric, and it changes the impedance of the traces above it without changing anything that an average measurement can see.
| Void thickness | Effective Dk | Impedance from 50.0 Ω | Visible to |
|---|---|---|---|
| 5 µm | 3.48 | 51.2 Ω | Per-trace TDR only |
| 10 µm | 3.31 | 52.3 Ω | Per-trace TDR only |
| 20 µm | 3.03 | 54.4 Ω | TDR, sometimes cross-section |
| 50 µm | 2.40 | 60.1 Ω | Cross-section, and the customer |
Voids come from three sources: air trapped under vacuum failure, moisture flashing out of the prepreg, and particles holding the bondply off the surface. All three are controlled the same way — a vacuum press that actually reaches vacuum, a bake profile for the prepreg, and cleanliness in the lay-up room — and none of them are controlled by pressure alone. Increasing pressure on a stack with a trapped void does not close it; it flattens the panel around it. How much that void costs depends on where it is found, which is the argument in common scrap causes.
Pressure and Flow: Two Levers That Fight Each Other
Pressure has to do two jobs at once: consolidate the stack to the target thickness, and drive resin into the gaps between traces without starving the bond line. On thermoset stacks the working range is roughly 200–400 psi with a resin flow of 15–25 %. On PTFE the upper end of that range is dangerous, because PTFE does not resist load — and the thinning is plastic and permanent.
The practical resolution is to use the lowest pressure that achieves consolidation and to compensate with a thicker starting laminate, rather than to press a thin laminate down to the target. This is why a PTFE stack is often specified with a core slightly thicker than the electrical model wants, and why the model should be built on the pressed height rather than the ordered height. High pressure is not a quality measure; on PTFE it is a source of impedance error that no coupon will attribute correctly, because the coupon is pressed in the same cycle and thins by the same amount.
Cooling and Opening: The Step With No Instrument on It
The press cycle is monitored for temperature and pressure and almost never for the one variable that decides flatness: the temperature at which the load is released. A thermoset panel opened above its cure temperature, or a PTFE panel opened while still soft, is free to distort without any constraint, and the distortion is frozen in as it cools.
The rule is simple to state and easy to violate under schedule pressure: cool under pressure, open below the temperature at which the bond is set, and let the panel equalise before the next operation. For thermoset stacks that means below the cure temperature with a controlled ramp rather than a quench; for PTFE it means well below the point where the material still deforms under its own weight. A panel that is 0.3 % out of flat will pass every electrical test and fail at the assembly fixture, which is the worst possible place to discover a lamination decision.
Verifying the Three Accounts
Verification has to match the account being verified, and the most common mistake is to verify one and assume the others.
| Account | Measurement | Frequency | What it does not tell you |
|---|---|---|---|
| Position | X-ray of layer-to-layer targets | First board of every lamination batch | Whether the drill will stay inside the ring |
| Thickness | Coupon TDR, or microsection | Per lot per impedance layer | Whether one trace differs from the average |
| Stress | Bow and twist, thermal stress test | Per lot, plus after finish | The stored strain that has not yet cracked |
| Voids | C-SAM or microsection | First article, and on any process change | Voids smaller than the resolution |
One correction to a claim that appears in a great deal of supplier literature: impedance is not measured on 100 % of controlled-impedance lines. Continuity and isolation are tested on 100 % of boards; impedance is measured on coupons that are designed to represent the layers, per lot, with a stated method and frequency. A supplier who claims full TDR coverage on every net is either describing something else or describing a sampling plan with unusual language. What each test stage can prove is set out in high-frequency PCB testing.
Ten Points to Release Before the Press Cycle
1. Stack symmetric in copper and dielectric about the centre line. 2. Core thickness = required dielectric height minus pressed bondply, and the core is a stocked item. 3. Process Dk and design Dk both stated, and the model uses the right one. 4. One bonding system per press cycle, and the number of cycles written on the drawing. 5. PTFE confined to the layers that carry the field wherever the application allows. 6. Every hole checked against a 6:1 aspect ratio in the finished stack. 7. Copper distribution balanced on each layer, so that resin flow does not create local thickness variation. 8. Prepreg bake profile specified per family, and the store controlled at 23±2 °C and 40±5 % RH. 9. Coupon design agreed, with the measurement frequency and the number of points per layer. 10. Cooling and opening temperature specified, with the flatness limit stated as a percentage of diagonal rather than as “flat”.
All ten are answerable in a single review meeting. Nine of them cost nothing to change at that point, and eight of them are expensive or impossible to change once the first panel has been pressed.
Conclusion
The barriers specific to multilayer lamination on high-frequency material are three accounts opened simultaneously: position, thickness and stress. Position accumulates at roughly ±25 µm per press cycle and is the reason stacked microvias yield worse than staggered ones. Thickness is the core plus the pressed bondply, not the core alone, and neglecting a 0.10 mm bondply costs 22 % of impedance. Stress is what cooling leaves behind when the stack is asymmetric, and it decides flatness and barrel life rather than impedance.
The largest single lever is the number of press cycles, which is set by chemistry and not by process skill: 170–190 °C, 190–200 °C and 260–300 °C cannot be opened at once. Confining PTFE to the outer layers and using one bonding system is worth more than any press-recipe optimisation, because it removes an entire accumulation of error rather than reducing one term in it.
XCE PCB laminates 2 to 72 layers on panels up to 600×1200 mm, in PTFE, hydrocarbon and ceramic systems, with impedance held to ±10 % on coupons, samples in as fast as 24 hours and inspection to IPC-A-600G and IPC-6015. Send the stack drawing before the laminate is ordered — steps 3 and 4 in the calculation above are where most of the avoidable cost sits. The step-by-step press behaviour is in lamination in high-frequency PCB manufacturing, and the wider route in high-frequency PCB manufacturing process.
Frequently Asked Questions
Why does a PTFE multilayer need two press cycles?
How much registration error does each lamination add?
Is the dielectric height the laminate thickness on the datasheet?
How much does a void in the bondply change impedance?
Can CTE be matched between the laminate and copper to avoid barrel cracks?
Is impedance measured on every board?
Have Your Stack Checked Before the Press Cycle
Send us your Gerber files and requirements for a manufacturability review and a tailored quotation.
sales1@xcepcb.com
+86 180 3341 4657