5G Small-Cell Boards: RF Shielding and Thermal Trade-offs at 28 GHz

Small-cell radios for 5G n257/n258/n261 (24.25–29.5 GHz) operate in a band where every millimetre of geometry matters. A 3 dB margin on the link budget can be the difference between a clean handover and a dropped session, and a meaningful slice of that margin is spent — or saved — inside the can shield around the PA and front-end. We routinely measure 4–7 dB of variation in adjacent-channel emissions across nominally identical builds, traced to nothing more than gasket compression and seam continuity.

The mistake we see most often on first-pass designs is treating the shield as a Faraday cage borrowed from sub-6 GHz. At 28 GHz, the wavelength in free space is roughly 10.7 mm, which means seam-leakage slots become resonant antennas as small as 5.3 mm. The four mechanisms that drive performance on real boards:

• Seam leakage — any unstitched seam longer than λ/4 (~2.7 mm) radiates measurably. We've seen 12 dB of shielding-effectiveness loss from a single 4 mm un-stitched corner.
• Aperture coupling — vent holes, screw clearances, even pad cut-outs for solder access. A 1.5 mm aperture is essentially transparent at 28 GHz unless honeycomb-loaded.
• Gasket compression — the conductive elastomer between can and PCB is rated at a contact-resistance value that only holds at the specified compression range (typically 15–30%).
• Cavity resonance — the enclosed volume itself is a resonant box at frequencies set by its dimensions. A 12 mm × 8 mm × 3 mm cavity has a TE101 resonance near 22 GHz, which then couples to a TE201 at 32 GHz — both in-band.

"The first time we measured a customer's small-cell prototype, the PA was hitting its compliance target on the bench and failing it by 6 dB in the production housing. The difference wasn't the silicon — it was a 0.4 mm shadow gap where the can wasn't fully seating on a slightly bowed substrate." — Pioneer Horizon RF team lead

Three families of can shields dominate the small-cell market. Each has a different cost, height-profile, and shielding-effectiveness curve. Pick the wrong family and you will spend the rest of the programme compensating with gaskets and tape.

Stamped two-piece (frame + lid)

The default for high-volume mmWave. Tin-plated nickel-silver, typically 0.15–0.2 mm wall. The frame solders to the PCB with castellated tabs on 2.0–3.0 mm pitch (we hold to 2.0 mm pitch for any wall facing an active PA — at 3.0 mm pitch the inter-tab slots become λ/4 antennas in n258). The lid press-fits or clips on. Shielding effectiveness up to 60 dB at 30 GHz with proper tab pitch.

Drawn one-piece

Deep-drawn from a single sheet. No seams, which removes the lid-to-frame leakage path entirely. The trade-off is height — drawn cans cap out around 5 mm internal height before the corner thinning becomes a problem. Use these when total stack-height permits and the PA is the dominant emitter.

Cast aluminium

Reserved for outdoor radios where thermal mass and IP-rated sealing matter. SE in excess of 80 dB is achievable but you pay 3× the unit cost of stamped and you fight assembly fixturing on the SMT line. We use cast cans on outdoor small-cell radios with sustained PA dissipation above 8 W where the can also serves as a heat sink.

Pad-stack rules under any can

• Continuous ground-ring pad at least 0.8 mm wide all around the can footprint. Break the ring and you break the shield.
• Via stitching every 1.2 mm along the ring (λ/8 at 30 GHz in FR-4) connecting to the nearest reference plane.
• Solder-mask opening matched 1:1 to the metal frame — over-sized masks invite seam leakage along the meniscus.
• No traces routed under the can-wall footprint. Inner-layer signals must dive at least one plane below before crossing the wall, and surface-layer traces stop at the ring.

The most common single failure on small-cell builds we audit is gasket under-compression. A conductive elastomer rated for 0.4 Ω-cm at 20% compression delivers nothing close to that at 5% compression — and PCB warp plus enclosure-lid tolerance routinely conspires to drop average compression below spec across the gasket length.

Compression math we hold to

Target nominal compression of 22% ±5%, which forces you to budget for:

• PCB flatness — typical FR-4 mid-panel bow is 0.3% of diagonal, or 0.18 mm across a 60 mm board edge. Specify 0.15% max for any board with a peripheral gasket.
• Enclosure lid tolerance — most stamped lids hold 0.1 mm. Combined with PCB bow, that's already 0.28 mm of variation. Choose a gasket whose 22% compression is at least 0.4 mm thick.
• Long-term compression set — fluorosilicone elastomers settle 5–10% over 1,000 hours at 85°C. Spec the initial compression high enough that end-of-life still clears the minimum.

Thermal path under a 12 W PA

The PA dissipates roughly 12 W into a footprint of 5 × 5 mm with a 3 mm-deep enclosure lid above it. Conduction is the only realistic path — convection inside a sealed can does nothing. We use this stack:

1. PA exposed die pad solders directly to a copper coin (2 mm thick, press-fit into the PCB) — this drops θ_jc by roughly 40% versus a thermal-via array.
2. Copper coin contacts the can floor through a 0.5 mm graphite-loaded TIM with bulk conductivity ≥ 6 W/mK.
3. Can floor is mechanically clamped (not just soldered) to the enclosure heat-spreader with M2 screws on 8 mm pitch.
4. Total stack θ_ca measured at 4.2 K/W in our thermal chamber — gives a 50°C rise at 12 W, which lands T_j around 100°C in a 35°C ambient enclosure.

Skip any layer in that stack and the PA throttles. We've seen designs where the can floor was 0.7 mm above the heat-spreader with only air in between — junction temperature ran 30°C hot and the PA's internal protection back-pressured the baseband for hours of every day.

At sub-6 GHz, you can usually trust a 3D EM tool's first answer. At 28 GHz, you need to verify the model against measurement before you trust any sweep. We close the loop with a deliberate three-step process on every small-cell design.

Step 1 — Pre-layout EM sweep

We model the can geometry (frame + lid + ground-plane stitching) in HFSS or CST Studio with a port at the PA reference plane and a virtual probe in the cavity. The sweep covers 20–35 GHz to capture both the in-band response and the first two cavity resonances. The deliverable is a recommended can size, tab pitch, and any required absorber-pad placement before layout starts.

Step 2 — Bench measurement on the first build

We measure three things on the first prototype:

• Reverberation-chamber SE sweep from 20–40 GHz with the can populated and the PA powered down. This isolates passive shielding from active emissions.
• Conducted emissions at the antenna port with the PA in transmit, captured against the 3GPP TS 38.104 spectral mask.
• Cavity probe through a 0.5 mm aperture (later filled with conductive epoxy) to validate the simulated cavity-resonance frequencies. If sim and measurement diverge by more than 1.5 GHz on the first resonance, the model is wrong and the sweep results are not trustworthy.

Step 3 — Correlation and the design-rule update

If measurement and sim agree within 1 dB on SE and within 800 MHz on cavity resonance, the simulation is trusted for the rest of the programme. If they don't, we tune the model — usually by adding measured gasket impedance and PCB material loss at the actual 28 GHz frequency rather than the datasheet 10 GHz value — and re-sweep. This calibration step costs roughly two engineering days and prevents the second silicon spin that would otherwise be triggered by a late compliance failure.

Pioneer Horizon maintains a calibrated EM model library for the 14 most common can-shield SKUs we use in production. New designs start from that library, which collapses the per-programme correlation work to one bench run instead of three.

A perfectly designed can shield can still fail in production for reasons that have nothing to do with RF. The four production-line failure modes we've debugged most often on small-cell builds:

1. Solder voiding under the can frame

Stamped frames with long unbroken castellated edges trap volatile flux during reflow. We see 30–40% voiding on first runs unless the stencil aperture is broken up. The fix is a stencil with 0.4 mm aperture gaps every 6 mm along long walls — paste volume drops 8% but voiding falls under 10% and SE recovers fully.

2. Lid lift after second reflow

Two-piece cans where the lid is clipped on can lift 0.1–0.2 mm during second reflow as thermal expansion mismatches the frame. Spec lids with at least four clip points and a press-fit interference of 0.05 mm; verify lift with a post-reflow flatness check on the first ten boards of every run.

3. Gasket dislodgement at lid install

Adhesive-backed conductive elastomers sometimes detach during lid press-fit if the operator slides rather than presses. We've moved to peel-and-place gaskets with a 30 N/cm peel strength on the lid side and a dispensed conductive elastomer on the PCB side for any radio shipping more than 1,000 units/month.

4. The ones you can't rework

Once a one-piece drawn can is reflow-soldered, it is essentially permanent. Rework requires μBGA-level localised heating that often de-solders nearby fine-pitch parts. Plan for this in your test strategy — anything inspected only after can installation must be inspectable post-can, or you have to accept a fixed scrap rate. For our highest-volume small-cell line we run boundary-scan plus a 40 GHz network analyser test through the antenna port before can placement, then a final spectral mask test after. Defects caught at the pre-can stage cost us roughly ₹200; defects caught post-can run ₹2,800 with a 30% scrap probability.

If you're starting a 5G small-cell programme, see our companion piece on RF compliance pre-screening and our approach to thermal-coupled layouts for the underlying patterns. Or share your RF block diagram and we'll come back with a can-shield + thermal proposal within three working days.