Silk Screen on High-Frequency PCB: The Legend Is a Keep-Out Rule, Not an Ink Choice

Legend printing is the last step that touches the board and the only one whose electrical effect is normally solved by geometry rather than by material. That distinction gets lost in most descriptions, which recommend low-Dk, low-loss legend ink for high-frequency work — advice that cannot be followed, because the ink that actually gets used is white, pigment-loaded with titanium dioxide, and has a dielectric constant several times that of the laminate underneath. The rule that works is simpler: the legend does not go on RF traces, and once that is decided the ink only has to survive assembly.

The standard recommendation for high-frequency legend is an ink with low dielectric constant and low loss tangent. It reads well and it is not implementable. Legend ink has to be opaque, and opacity in a white ink comes from titanium dioxide, whose dielectric constant is in the tens. A 20 µm film of that over a microstrip is a high-Dk layer sitting on the conductor, and no commercially available white ink avoids it.

Run the same first-order estimate used for solder mask, on a 0.254 mm 5880 build with a 20 µm legend film at Dk 6.5: the effective dielectric constant of the line rises about ten per cent, and a 50 Ω trace reads roughly 47.5 Ω. A 2.5 Ω shift from a printed label is not something to manage by ink selection.

The practical answer is geometric: keep the legend off RF traces, off impedance-controlled lines, off pads and off fiducials, and the ink’s dielectric properties stop mattering. That is a design rule in the data package, and it costs nothing to apply — unlike a material specification that cannot be met.

This is the last step of the twelve-step sequence, applied after the mask described in the solder mask guide.

These numbers interact. A 0.15 mm stroke through a coarse mesh prints thick and ragged; a 0.10 mm stroke needs a fine mesh and a thin film, which reduces opacity. That is why 0.15 mm is the working floor rather than a target, and why reducing the character height to fit more text usually produces a board that cannot be read after conformal coating.

The board arrives at this step masked, cured and often finished. Preparation is a clean, dry, fully cured surface: no mask residue, no finish contamination, no moisture. Legend printed over an under-cured mask will lift with the mask rather than with the ink, which makes the failure look like an ink problem when it is not.

On PTFE the same surface-energy problem described for mask applies again. The legend sits on mask rather than on laminate in most areas, so the adhesion question moves to the mask-to-laminate interface — but wherever legend lands directly on fluoropolymer, plasma treatment or a promoter is required, exactly as for the mask itself.

The pattern is transferred through a screen: a woven mesh, usually polyester or stainless steel, coated with a photosensitive emulsion that is exposed and developed so that the image areas are open. Ink is forced through the open mesh by a squeegee.

This is worth stating precisely because the equipment is often called a stencil, which in PCB work means the laser-cut metal foil used for solder paste printing. The two are different tools with different failure modes: a stencil wears by clogging and by foil deformation; a screen wears by mesh stretching, emulsion breakdown and tension loss, and a stretched screen prints out of registration without any visible defect.

Screen tension and emulsion thickness are the two parameters to hold. Tension is checked per screen and re-tensioned or replaced on a schedule; emulsion thickness sets the ink film, which sets opacity and edge definition together.

For high-frequency boards the selection criteria are practical rather than electrical: adhesion to the mask chemistry in use, resistance to the assembly process (reflow, wave, cleaning), and colour contrast for automated optical inspection and for human rework.

Mixing and viscosity control matter more than the datasheet suggests: an ink that has thickened prints with ragged edges and skips, and thinning it to recover viscosity reduces opacity and can attack the mask underneath. The pot life on a two-part ink is a hard limit, not a guideline.

Squeegee pressure, angle and speed set the ink film and the edge definition together. Too much pressure and the ink spreads under the emulsion, closing fine strokes and producing haloes; too little and the mesh does not empty, leaving a starved print that fails opacity on the second pass.

The design-side rules are what protect the RF performance, and they belong in the fabrication drawing rather than in a shop-floor conversation:

• No legend on RF traces or impedance-controlled lines. The ink has a dielectric constant several times the substrate; 20 µm over a 0.254 mm PTFE line is a 2.5 Ω shift.
• No legend on pads, test points or fiducials. Ink on a pad blocks wetting; ink on a fiducial blinds the placement machine.
• Keep 0.15–0.2 mm clear of pad edges. Registration is ±0.15 mm, so anything closer will print onto a pad on some panels.
• Do not use legend as a substitute for mask openings. A printed boundary does not insulate; the mask does that.

Legend ink is not cured in one pass. The standard sequence is a low-temperature pre-cure to release solvent, followed by a higher-temperature final cure to complete the reaction.

Skipping the pre-cure and going straight to 150 °C skins the surface while solvent is still inside; the film blisters or, more often, looks fine and fails a tape test after the first reflow. Conformal coating over an under-cured legend is the common way this reaches a customer.

Legend is the least electrical step on the board and the one most often given electrical advice. The advice that works is geometric: keep the ink off the conductors that carry RF, away from pads and clear of fiducials, and the high dielectric constant of a white pigment stops being a design problem.

The claims worth discarding are that a low-Dk, low-loss legend ink solves the RF interaction (white ink is titanium-dioxide loaded and no commercial grade is low-Dk), that the pattern comes through a stencil (it comes through a mesh screen; the stencil is the solder-paste tool), and that a single bake cures the ink (it takes a pre-cure and a final cure, or a UV dose plus a bake for shadowed areas).

XCE PCB prints legend with 0.15 mm minimum strokes, documented keep-outs from RF traces and pads, tension-controlled screens and two-stage cure, on Rogers, Taconic and F4B builds: 2–72 layers, impedance ±10%, panels to 600 × 1200 mm, 5,000 m² of production space and 80+ engineers, with samples in as fast as 24 hours.