Four issues appear once the antenna moves inside the package: the board becomes part of the radiator through the ground it presents, any material above the package detunes it, the joints have to hold one reference plane flat across the array, and the mold compound brings a moisture sensitivity a bare board never had. None of them looks like a placement error, and all four pass a visual inspection at the end of the line.
The board is now part of the antenna
An antenna in package still radiates against the ground plane it is soldered to. At 60 GHz a patch array on a 2.5 mm half wavelength grid sits a few hundred micrometres above a reference plane that the board has to continue without interruption. A break in that plane under or beside the package becomes a slot radiator, and a slot beside an array does more damage than the same break elsewhere on the board.
The usual guidance for a via fence is a pitch of lambda over twenty, which at 60 GHz is 0.25 mm and cannot be stitched inside a 0.4 mm ball grid field. The fence therefore has to live outside the land area, on a ring the package outline allows, with the plane under the land continuous rather than stitched. Where a signal crosses that boundary it crosses on a layer below the reference, never by cutting the reference.
Anything above the radiator detunes it
Coating, underfill, potting, label stock and a foam gasket all sit in the near field. At 60 GHz the free space wavelength is 5 mm and at 77 GHz it is 3.9 mm, so a 50 micrometre layer is one percent of a wavelength thick, but its dielectric constant decides more than its thickness. A layer with a dielectric constant near 3 raises the effective permittivity of the radiating structure by a few percent and moves the resonancof megahertz to a couple of gigahertz.
The failure is quiet. A coated unit still transmits and receives, just off the frequency the filter was set for, with a few tenths of a dB of extra loss, and a functional test does not flag it. That is why the coating drawing has to carry a mask over the antenna window, and why underfill wicking towards the antenna side is a defect rather than a cosmetic issue.
Coplanarity and the joint that holds the reference
A thin package with a large body is the worst case for coplanarity. Package and board expand differently through the profile, the package bows, and the joints that carry the load are the perimeter lands, which on an antenna in package are usually ground. A corner ground land that opens does not fail a DC test: it removes part of the reference plane, the pattern tilts, and the loss appears in a radiated measurement weeks later.
What controls it is a land pattern that does not put the only mechanical support at the corners, a stencil aperture that delivers enough paste on the ground lands to close a small gap without leaving a void large enough to matter, and a reflow profile that brings the whole assembly to uniformity instead of chasing peak temperature. Void acceptance to IPC-7095 under 5 percent on ground and thermal pads, checked by X-ray and by CT where the void location matters, separates a joint that conducts from a joint that holds a reference plane.
Moisture inside the mold compound
An antenna in package arrives as a molded part with an MSL rating. A part rated MSL 3 has 168 hours of floor life at 30 degrees and 60 percent relative humidity, and after that it is baked to J-STD-033 at 125 degrees before reflow. Moisture left in the mold compound turns to steam, and the damage shows as delamination between the mold and the laminate, which is the interface the antenna stack depends on.
A popped package still passes a continuity test and fails as a shift in resonance and a drop in gain, usually after the build is finished. Floor life tracking at the line, dry storage and a bake record keyed to the panel date code are the only controls that work, because the defect cannot be seen from outside.
Three mistakes that cost a build
Coating to a note instead of a drawing. One coating instruction across the panel puts material over the antenna window, and the batch then measures low on gain with no visible cause. The mask belongs on the drawing and in the coating program, with the window marked physically at the station.
Reworking a package with hot air. A hand nozzle heats the mold compound above its glass transition while the joint is still solid, and the antenna layer delaminates from the laminate underneath. Rework belongs on a profile controlled station with preheat from below, and the unit belongs in a radiated check afterwards, not only a continuity check.
Qualifying on a bare board and adding the housing later. The first article is measured without the radome or the housing, the unit passes, and the shipped configuration is measured nowhere. Run the radiated check on the assembly as it ships, once per build, and keep the result with the first article report.
How SUPERB supports an antenna in package build
On the board side, SUPERB PCB handles RF and microwave work from 500 MHz to 77 GHz with 0.05 mm lines and 0.035 mm fanout on HDI stack-ups, via in pad plugged, filled and capped, impedance held to plus or minus 5 percent and plus or minus 2 percent on critical nets, a TDR coupon on every panel, and layer to layer registration of 0.025 to 0.06 mm. High layer count work runs to 68 layers with back-drilled vias held to a 0.25 mm residual stub.
On the assembly side, seven SMT lines in a 5,000 square metre ESD controlled area place from 01005 up to 57 mm at 40,000 CPH, with 3D SPI before reflow, 3D AOI after, X-ray and CT to IPC-7095, twelve zone nitrogen reflow below 500 ppm oxygen, MSL control and baking to J-STD-033 keyed to the panel, and selective coating with masking held to the drawing. Test is ICT, flying probe and functional test on a product specific fixture, and components carry a 730 day warranty.
Builds run build-to-print, with the Gerber, the BOM and the assembly drawing supplied by you and no engineering fee in the quotation. mmWave modules are assembled through the 60 GHz mmWave radar PCBA line and the wider turnkey assembly flow, with fine pitch and via in pad work covered by the HDI PCB capability and void control explained in the X-ray acceptance note.
If a build is losing gain after coating or after a rework, send the coating drawing, the keep-out, the MSL record and the X-ray images. The reply states which of the four mechanisms the evidence points at.