Complete guide to PCB layer stackup design, impedance control, and material selection
The PCB stackup is the foundation of every design. It determines signal integrity performance, power distribution efficiency, EMC compliance, and manufacturability. A poorly designed stackup cannot be fixed by routing alone - it must be correct from the start.
This tutorial covers the critical checkpoints for stackup verification during a PCB layout review, with practical examples and tool-specific guidance.
The layer count must meet routing density requirements with at least 20% spare via/channel capacity. It should not be over-specified (increasing cost) or under-specified (causing congestion and DRC violations).
| Design Complexity | Typical Layer Count | Indicators |
|---|---|---|
| Simple (Arduino-class) | 2 | <50 components, no BGA, <200 nets, single voltage domain |
| Moderate (IoT, Sensors) | 4 | 50-200 components, QFP/QFN, 200-500 nets, 2-3 voltage domains |
| Complex (Embedded Linux) | 6 | 200-500 components, BGA <400 pins, 500-1500 nets, DDR3 |
| High-Complexity (SoC) | 8-10 | BGA 400-900 pins, DDR4, 1500-3000 nets, multiple high-speed interfaces |
| Very High (Server/FPGA) | 12-20+ | BGA 1000+ pins, DDR4/5 multi-channel, PCIe Gen4/5, >3000 nets |
A design with a 676-pin BGA (1.0mm pitch) uses 8 layers. Signal layers have 65% utilization. Two dedicated ground planes and one power plane provide solid references. Escape routing from BGA uses only 4 signal layers with dog-bone via fan-out.
Same design forced into 4 layers. Signal layers at 95% utilization. Ground plane heavily split by routing. Multiple DRC violations remain unresolved. Traces forced to route far from their reference planes. Designer adds ground stitching vias to "fix" SI issues.
Choosing layer count based on cost alone. A 4-layer board that requires 3 design revisions due to SI/EMC failures costs far more than an 6-layer board done right the first time. Budget an additional 2 layers beyond minimum routing needs for signal integrity.
Every high-speed signal layer must have an unbroken reference plane on an immediately adjacent layer. The arrangement must provide continuous return paths and minimize layer transitions that cross reference plane boundaries.
Layer 1 (TOP): Signal + Components | 1.0 oz Cu
--- Prepreg (4-5 mil) ---
Layer 2 (GND): Ground Plane | 1.0 oz Cu
=== Core (40 mil) ===
Layer 3 (PWR): Power Plane | 1.0 oz Cu
--- Prepreg (4-5 mil) ---
Layer 4 (BOT): Signal + Components | 1.0 oz Cu
Total: ~62 mil (1.57 mm)
Impedance: ~50 ohm single-ended with 7 mil trace on 4 mil prepreg (Dk=4.2)
Layer 1 (TOP): Signal (High-Speed) | 0.5 oz Cu
--- Prepreg (3.5 mil) ---
Layer 2 (GND): Ground Plane | 1.0 oz Cu
=== Core (10 mil) ===
Layer 3 (SIG): Signal (General) | 0.5 oz Cu
--- Prepreg (28 mil) ---
Layer 4 (SIG): Signal (General) | 0.5 oz Cu
=== Core (10 mil) ===
Layer 5 (PWR): Power Plane | 1.0 oz Cu
--- Prepreg (3.5 mil) ---
Layer 6 (BOT): Signal (High-Speed) | 0.5 oz Cu
Total: ~62 mil (1.57 mm)
Note: Layers 3 and 4 reference planes on adjacent layers (L2-GND, L5-PWR)
Layer 1 (TOP): Signal (High-Speed) | 0.5 oz Cu
--- Prepreg (3 mil) ---
Layer 2 (GND): Ground Plane | 1.0 oz Cu
=== Core (5 mil) ===
Layer 3 (SIG): Signal | 0.5 oz Cu
--- Prepreg (5 mil) ---
Layer 4 (PWR): Power Plane | 1.0 oz Cu
=== Core (20 mil) ===
Layer 5 (GND): Ground Plane | 1.0 oz Cu
--- Prepreg (5 mil) ---
Layer 6 (SIG): Signal | 0.5 oz Cu
=== Core (5 mil) ===
Layer 7 (PWR): Power Plane | 1.0 oz Cu
--- Prepreg (3 mil) ---
Layer 8 (BOT): Signal (High-Speed) | 0.5 oz Cu
Total: ~62 mil (1.57 mm)
All signal layers have adjacent reference planes. Layer 4 (Power) and Layer 5 (Ground) are tightly coupled for embedded capacitance. High-speed signals on L1 and L8 have the thinnest dielectric to their reference planes.
Signal layers 1-2 and 7-8 are adjacent with no reference between them. Crosstalk between these layers will be severe. Inner signal layers have thicker dielectric to references, making impedance control difficult.
Stackup must be symmetric about the center axis with maximum asymmetry less than 10% of total thickness. Copper distribution, dielectric types, and thicknesses should mirror from center.
During lamination, the PCB is heated to 180-200C. Different materials expand at different rates. An asymmetric stackup creates unbalanced stress that causes the board to bow or twist when cooled. IPC-6012 Class 2 allows maximum 0.75% bow/twist; Class 3 allows 0.5%.
Even with a symmetric stackup structure, unbalanced copper fill can cause warpage. If Layer 1 has 80% copper coverage (ground fill) but Layer 8 has only 30% copper coverage (sparse routing), the board may warp. Solution: Add copper fill (hatched or solid) to balance coverage, ensuring all mirrored pairs have similar copper area within 20%.
| Parameter | Symmetric (Good) | Asymmetric (Bad) |
|---|---|---|
| L1 copper weight | 0.5 oz | 1.0 oz |
| L8 copper weight | 0.5 oz | 0.5 oz |
| Top prepreg | 1080 + 1080 (4 mil) | 2116 (5 mil) |
| Bottom prepreg | 1080 + 1080 (4 mil) | 1080 (3 mil) |
| Core structure | Core-Prepreg-Core (centered) | All cores on top half |
| Warpage risk | Low (<0.5%) | High (>1.0%) |
Every controlled-impedance trace routes on the designated layer with width matching field-solver calculation within 5%. Dielectric thickness values must be from the fabricator's standard offering.
Single-ended impedance (microstrip, outer layer):
Z0 = (87 / sqrt(Er + 1.41)) * ln(5.98 * H / (0.8 * W + T))
Where:
Z0 = characteristic impedance (ohms)
Er = dielectric constant (Dk)
H = dielectric height to reference plane (mils)
W = trace width (mils)
T = copper thickness (mils)
| Target Z0 | Dielectric (mil) | Trace Width (mil) | Copper (oz) | Type |
|---|---|---|---|---|
| 50 ohm SE | 3.5 | 5.5 | 0.5 | Microstrip |
| 50 ohm SE | 4.0 | 6.5 | 0.5 | Microstrip |
| 50 ohm SE | 5.0 | 8.0 | 0.5 | Microstrip |
| 50 ohm SE | 4.0 | 5.0 | 0.5 | Stripline |
| 50 ohm SE | 5.0 | 6.5 | 0.5 | Stripline |
| 90 ohm Diff | 4.0 | 4.0/5.0 s | 0.5 | Microstrip |
| 100 ohm Diff | 4.0 | 3.5/6.0 s | 0.5 | Microstrip |
| 100 ohm Diff | 5.0 | 4.0/5.5 s | 0.5 | Stripline |
Note: "s" denotes spacing between differential pair traces. All values approximate - always verify with a 2D field solver.
| Parameter | JLCPCB | PCBWay | Advanced Circuits |
|---|---|---|---|
| Min dielectric | 3.0 mil | 2.5 mil | 3.0 mil |
| Impedance tolerance | +/- 10% | +/- 10% | +/- 5% (premium) |
| Standard Dk (FR-4) | 4.2-4.5 @1GHz | 4.2-4.6 @1GHz | 4.2-4.4 @1GHz |
| Dk consistency | +/- 0.2 | +/- 0.15 | +/- 0.1 |
| Available prepregs | 1080, 2116, 7628 | 1080, 2116, 3313, 7628 | All standard types |
| Min core thickness | 4 mil | 3 mil | 2 mil |
Use Design > Layer Stack Manager. Click "Impedance" tab to define impedance profiles. Altium uses the Simberian solver for accurate results. Each impedance profile links to a net class with automatic width assignment. Export stackup as IPC-2581 for fabricator communication.
Use Board Setup > Physical Stackup. KiCad has a built-in impedance calculator under Inspect > Board Statistics or use the standalone "PCB Calculator" tool. For controlled impedance, define net classes with appropriate track widths in Board Setup > Net Classes.
Use Setup > Cross-section to define the stackup. The integrated Allegro PCB SI tool provides field-solver impedance calculation. Define impedance constraints in the Constraint Manager under "Physical > Impedance" for automatic DRC checking.
Copper weight per layer matches current capacity needs. Power layers should be at least 1 oz for high-current paths. Signal layers typically use 0.5-1 oz. Verification against IPC-2152 temperature rise charts is required.
| Weight (oz) | Thickness (mil) | Thickness (um) | Typical Use |
|---|---|---|---|
| 0.5 oz | 0.7 | 17.5 | High-density signal layers, fine-pitch routing |
| 1.0 oz | 1.4 | 35 | Standard signal/power layers, most designs |
| 2.0 oz | 2.8 | 70 | High-current power, automotive, power supplies |
| 3.0 oz | 4.2 | 105 | Very high current, bus bars, power converters |
| Trace Width | 0.5 oz | 1.0 oz | 2.0 oz |
|---|---|---|---|
| 5 mil | 0.35 A | 0.50 A | 0.70 A |
| 10 mil | 0.60 A | 0.85 A | 1.20 A |
| 20 mil | 1.00 A | 1.40 A | 2.00 A |
| 50 mil | 1.80 A | 2.50 A | 3.50 A |
| 100 mil | 2.80 A | 4.00 A | 5.60 A |
Note: Internal layers carry approximately 50% of the external layer capacity due to reduced convection cooling.
When copper weight changes between layers (e.g., 0.5 oz signal, 1.0 oz plane), the trace thickness affects impedance. A 50-ohm trace designed for 0.5 oz copper will be approximately 48 ohms if actually fabricated with 1.0 oz copper (after etching). Always specify the finished copper thickness in impedance calculations, not the base copper weight.
Material type (FR-4, high-Tg, Rogers, Megtron) specified with Dk and Df values at operating frequency. Dk tolerance confirmed by fabricator to be within 5%.
| Style | Thickness (mil) | Resin Content (%) | Dk @1GHz | Notes |
|---|---|---|---|---|
| 1080 | 2.8-3.3 | 62-68 | 3.9-4.2 | Thin, high resin, good for thin dielectrics |
| 2116 | 4.5-5.0 | 48-54 | 4.2-4.5 | Most common general-purpose prepreg |
| 3313 | 4.0-4.5 | 55-60 | 4.0-4.3 | Medium weight, good flow |
| 7628 | 7.0-7.5 | 42-48 | 4.5-4.7 | Thick, low resin, for spacing |
| Application | Material | Dk | Df | Tg (C) |
|---|---|---|---|---|
| General consumer | Standard FR-4 | 4.2-4.5 | 0.018-0.022 | 130-140 |
| Automotive/Industrial | High-Tg FR-4 | 4.2-4.4 | 0.015-0.020 | 170-180 |
| High-speed (>5 Gbps) | Megtron 6 | 3.6-3.8 | 0.004-0.006 | 185 |
| RF/Microwave | Rogers RO4350B | 3.48 | 0.004 | 280 |
| mmWave (>30 GHz) | Rogers RO3003 | 3.00 | 0.001 | 350 |
| Mixed (RF + digital) | Hybrid (Rogers + FR-4) | varies | varies | varies |
Fabrication note states: "Material: Isola 370HR, Dk=4.04 at 1 GHz per IPC-TM-650 2.5.5.5, Tg=180C (DSC), Td=340C. Prepreg: 2x1080 (62% RC) for controlled impedance layers. Core: 0.1mm Isola 370HR."
Fabrication note states: "Material: FR-4" with no Dk value, no specific laminate brand, no Tg requirement. Fabricator uses cheapest available material with Dk anywhere from 4.0-4.8, causing impedance variation up to 15%.
All layers carrying impedance-controlled nets are explicitly identified. Target impedance values, trace widths, and tolerances are documented in both the constraint system and fabrication drawing.
Design Rules > Routing > Matched Net Lengths| Checkpoint | Priority | Key Metric |
|---|---|---|
| Layer count adequate | Major | Signal layer utilization <80%, 20% spare capacity |
| Plane arrangement | Critical | Every signal layer has adjacent reference plane |
| Stackup symmetry | Critical | Asymmetry <10% of total thickness |
| Dielectric for impedance | Critical | Calculated Z0 within 5% of target |
| Copper weight defined | Major | Meets IPC-2152 current/temp requirements |
| Materials specified | Major | Dk, Df, Tg values documented |
| Impedance layers identified | Critical | All controlled-Z nets on correct layers with correct widths |