High Frequency RF PCB: Materials, Stack and Design

What Makes a Board High Frequency

A board is high frequency when the behaviour of its conductors stops being a wire problem and becomes a wave problem. The practical threshold is not a specific number but a relationship: when a trace is longer than a fraction of the wavelength of the signal it carries, the reflections, the loss and the coupling all become significant, and the geometry of the trace becomes part of the circuit.

Two things follow. The first is that the substrate becomes an electrical component rather than a mechanical support, because its dielectric constant sets the wavelength and its loss tangent sets the attenuation. The second is that the design cannot be validated by inspection; it has to be simulated and then measured.

Choosing the Laminate

The material is chosen from the loss budget on the worst line, not from the frequency of the design in general. The loss of a line has three parts: the conductor loss, which falls with a wider trace and a smoother copper, the dielectric loss, which rises with the loss tangent and with frequency, and the radiation loss, which rises with frequency and is controlled by the stack rather than the material.

Working through the options:

• Standard FR-4. Dielectric constant around 4.2 to 4.5, loss tangent around 0.02. Adequate below a few gigahertz for short traces, and unusable for a long line at higher frequencies.
• Low loss FR-4 and modified FR-4. Loss tangent reduced by a factor of two or three, with the same processing and much of the same cost. The right first step when standard FR-4 is marginal.
• Mid loss laminates. A hydrocarbon or a modified resin with a loss tangent of around 0.004 and a dielectric constant around 3.5. The workhorse for base stations, radar modules and high speed digital boards.
• PTFE laminates. Dielectric constants from 2.2 to 3.0 and loss tangents of 0.001 to 0.002. Used for millimetre wave work, long transmission lines and wideband structures.
• Ceramic filled materials. Higher dielectric constants, used to make compact resonators and small antennas where the physical size matters more than the loss.

Three properties matter as much as the loss. The dielectric constant has to be stable with temperature, because a filter or a matching network designed against a drifting value loses its alignment. The batch tolerance of the dielectric constant has to be tight, because a printed filter is produced in quantity without individual tuning. And the material has to be available in the thicknesses the stack requires, because changing the thickness changes the impedance. Our notes on PCB manufacturing describe how the material and stack are controlled in production.

The Stack

The stack is designed from the radio frequency section outward. The requirements are simple to state and demanding to satisfy.

A solid ground plane directly beneath every radio frequency signal layer. The plane sets the impedance, provides the return path and shields the layer from what is below it. A radio frequency layer with a plane two layers away has an uncontrolled impedance and a return path that wanders.

A symmetrical construction. The board has to stay flat through lamination and reflow, and an unbalanced stack does not. On a hybrid stack with a speciality material on some layers, the fabricator adjusts the prepreg thicknesses to balance the construction.

Short ground returns. Ground on the outer layers is stitched to the internal planes with a via array, spaced a small fraction of a wavelength so that the planes behave as one conductor rather than as a set of resonant patches.

A defined separation between the radio frequency and the digital sections. Physical distance, a ground plane between them and a single reference point are all used together. Where both sections share a plane, the digital return current has to be kept out of the region beneath the radio frequency traces.

Our notes on PCB design and layout describe the routing that follows from these constraints.

Impedance Control

Every radio frequency trace has a target impedance, usually 50 ohms single ended and 100 ohms differential, and the trace geometry is calculated from the dielectric constant and the distance to the reference plane. Four practical rules govern the result.

Keep the geometry constant along the trace, because a change in width or in reference distance is a discontinuity. Keep the reference plane continuous, because a trace crossing a plane split has no defined return path. Place a return via beside every signal via, because a layer transition without a local return is an inductance. And verify the impedance on a coupon built on the production panel with the same material and the same geometry, rather than assuming that the calculated value was achieved.

The coupon matters more than it sounds. The difference between the designed and the produced impedance comes from the dielectric thickness and the dielectric constant, neither of which the designer controls directly, and the coupon is the only way to know whether the board meets the specification before the product is assembled.

One transition deserves special mention. A via at a high frequency is not a connection but a short section of transmission line with its own impedance, and the stub below the layer where the signal leaves is an open circuited stub that produces a notch in the response. Back drilling removes the stub, a blind via avoids it, and both are used where the frequency justifies the cost.

Shielding and Isolation

A radio frequency board is both a transmitter and a receiver of interference, and the two problems are addressed together.

Via fences. A row of ground vias along the edge of a radio frequency section, spaced a small fraction of a wavelength, forms a wall that stops the field from spreading along the surface.

Shield cans. A metal can soldered over the transceiver or the filter confines the field and protects the circuit from outside sources. The can is designed with the board: its attachment pads, its clearance and the ground connections all come from the layout.

Isolation between stages. The amplifier and the mixer, or the transmitter and the receiver, are placed so that the field from one does not reach the other. Physical separation, a ground plane between them and a shield wall are used in combination.

Supply filtering. Each radio frequency stage is fed through a filter that presents a low impedance at the operating frequency. A single capacitor is not enough at high frequency; the decoupling is a network with the smallest component closest to the device.

Testing

Three measurements confirm the board. The impedance coupon on the production panel checks the geometry. The insertion loss and the return loss of the transmission lines are measured with a network analyser, either on a dedicated test structure or on the assembled product. And the functional performance, the output power, the sensitivity and the spurious emissions, is measured on the complete assembly.

The assembly should be measured in the condition it will be used, which means in the enclosure and with the shield fitted. A board that measures well open and degrades in the housing had its enclosure effects designed rather than measured. Our notes on PCBA testing describe the test methods, and our notes on quality management describe the process control behind production.

Cost

The cost is driven by the laminate, the panel utilisation, the tolerance the design requires and the test programme. A mid loss laminate on a six layer board is a moderate premium over FR-4; a PTFE material on the radio frequency layers of a hybrid stack is a large one; and a board with tight impedance tolerances, back drilled vias and fine lines is more expensive again because the yield falls.

Two decisions save the most money. Use the speciality material only where the loss budget requires it, and use the cheapest material that meets the budget rather than the best material available. A design that specifies a PTFE laminate for a short trace at five gigahertz has paid for performance it will not use. Our notes on telecommunications PCBA describe the equipment these boards serve.

FAQ

At what frequency does a board become high frequency? It is a relationship rather than a threshold: when a trace is a significant fraction of a wavelength, its geometry becomes part of the circuit. That can happen at a few hundred megahertz on a long trace or at tens of gigahertz on a short one.

Can FR-4 be used for RF? Below a few gigahertz and on short traces, yes. Above that, or on a long line, the loss and the variation in dielectric constant become the limitation.

Why is a solid ground plane required under every RF layer? Because the plane sets the impedance, provides the return current path and shields the layer from the circuits below.

How is the design verified? By simulation before the layout is released, an impedance coupon on the production panel, and network analyser measurements of the assembled board in its enclosure.

Conclusion

A high frequency board is designed around its substrate and its ground. Choose the laminate from the loss budget on the worst line, design the stack from the radio frequency section outward with a solid plane under every signal layer, hold the trace geometry constant and verify it on a coupon, give every via a return path, and measure the finished assembly in the enclosure. The material is the largest single cost and the smallest single risk; the layout is where the design is won.