PCB Antenna and Feed Resources

mmWave Antenna in Package | Substrate & Assembly

This guide covers mmWave antenna in package (AiP) design from substrate material selection and stackup architecture through flip-chip assembly, thermal management, and production-readiness testing.

mmWave antenna in package module with patch array and BGA balls

What Is Antenna in Package (AiP)?

An antenna in package (AiP) places one or more radiating elements directly inside the IC package substrate rather than on the main PCB. The die, feed network, and antenna share a single laminate or ceramic module—typically smaller than 15 × 15 mm—eliminating long lossy board-level transitions. AiP is the dominant architecture for 24–77 GHz automotive radar, 28/39 GHz 5G front-end modules, and 60 GHz WiGig radios because it keeps the mmWave transition loss budget under control while enabling volume production in standard SMT lines.

Compared with antenna-on-board (AoB) approaches, AiP trades some aperture area for dramatically shorter interconnects and tighter impedance control. For a deeper look at when each topology wins, see the comparison of AiP versus AoB architectures.

Why AiP Matters Above 24 GHz

Cross-section of AiP substrate with microvias and flip-chip die

Free-space wavelength at 60 GHz is about 5 mm, and at 77 GHz it shrinks to roughly 3.9 mm. A half-wavelength patch at 60 GHz is under 2.5 mm on a side—small enough to fit inside a BGA package footprint with room for a 4 × 4 array. At these frequencies every millimeter of trace between the RFIC and the antenna adds measurable insertion loss (often 0.15–0.3 dB per mm on organic substrates at 60 GHz). Integrating the antenna inside the package keeps that path to a few hundred micrometers, which can save 2–4 dB of system loss compared with routing across a motherboard.

The short wavelength also means that package-level tolerances of ±25 µm (about ±1 mil; 1 mil = 0.0254 mm) are a small fraction of a wavelength, keeping phase errors across an array well within the ±10° budget needed for beam-steering accuracy.

Substrate Material Selection

Organic Laminates

Most commercial AiP modules use high-frequency organic laminates. Popular choices include Megtron 7 (Dk ≈ 3.4, Df ≈ 0.002 at 20 GHz) and Rogers RO4350B (Dk 3.48, Df 0.0037 at 10 GHz). For 77 GHz radar, lower-loss fluoropolymer-based materials such as RO3003 (Dk 3.00, Df 0.001) are preferred despite higher cost, because every 0.001 increment in Df adds roughly 0.3 dB of loss per wavelength of trace at that frequency.

LTCC (Low-Temperature Co-fired Ceramic)

LTCC offers hermetic packaging, stable Dk (typically 5.6–7.8 depending on tape system), and the ability to embed cavities and vertical transitions in a monolithic fired body. It dominates military and aerospace AiP where environmental robustness matters more than unit cost. Line/space capability is around 75–100 µm, adequate for 77 GHz patch elements.

Glass and Silicon Interposers

Emerging AiP substrates use thin glass (Dk ≈ 5.3, very low Df ≈ 0.0005) or high-resistivity silicon (>1 kΩ·cm). Glass interposers allow through-glass vias (TGVs) at 100 µm pitch and extremely flat surfaces for fine-pitch flip-chip bonding. These are still in early volume production but appear in research prototypes up to 140 GHz.

Substrate Dk (typical) Df (typical) Min Feature Cost Tier Best Fit
Organic (Megtron 7) 3.4 0.002 30 µm L/S Low–Med 5G FEM, 60 GHz
Organic (RO3003) 3.0 0.001 75 µm L/S Med 77 GHz radar
LTCC 5.6–7.8 0.001–0.003 75 µm L/S High Mil/aero, harsh env.
Glass interposer ≈ 5.3 0.0005 10 µm L/S Very High Research, D-band

Stackup Architecture

A typical organic AiP stackup has 6–10 metal layers. From bottom to top the functional zones are:

  1. BGA land layer – solder-ball pads that connect the module to the motherboard. Pitch is usually 0.4–0.8 mm.
  2. Power and ground planes – solid copper references that also serve as the antenna ground plane. Keeping this plane continuous under the array is critical; any slot or split shifts the antenna resonance.
  3. Feed network layer(s) – stripline or substrate-integrated-waveguide (SIW slot radiator) routing from the die to each element. Stripline is preferred over microstrip here because it is shielded between planes, reducing coupling to the mold compound above.
  4. Antenna layer – the top copper, where patch elements or grid-array slots are etched. This layer faces the package mold or an air cavity.

Microvias stitch signals between layers. Stacked microvias (as few as 75 µm drill diameter) are common in HDI mmWave stackups because staggered vias consume too much routing area at half-wavelength element spacing.

Antenna Element Design Inside the Package

Patch antennas dominate AiP because they are low-profile, easy to array, and naturally broadside-radiating. At 60 GHz on a Dk 3.4 substrate, a rectangular patch is roughly 1.4 × 1.4 mm. Bandwidth is typically 5–8 % for a single patch; stacked patches or aperture-coupled feeds can widen this to 15–20 %, enough to cover the full 57–71 GHz WiGig band.

Grid-array and slot antennas are alternatives when wider bandwidth or dual polarization is needed. For high-gain single-module solutions, mmWave antenna array design techniques—corporate feed networks, sub-array tiling, and amplitude tapering—apply directly inside the package.

After fabricating the array, accurate antenna pattern measurement in an anechoic chamber or spherical near-field scanner validates gain, beamwidth, and sidelobe levels before the module ships.

Assembly Process: Die Attach to Overmold

AiP assembly follows a sequence that blends semiconductor back-end and SMT practices. The steps below assume an organic substrate with a flip-chip RFIC—the most common commercial configuration.

Step-by-Step Assembly Flow

  1. Wafer bumping – Gold stud bumps or copper-pillar bumps are formed on the RFIC wafer. Bump pitch for mmWave dies is typically 100–150 µm.
  2. Flux and placement – The die is flipped and placed onto the substrate bond pads using a high-accuracy die bonder (±5 µm). Tacky flux holds the die during reflow.
  3. Thermocompression or mass reflow – Solder bumps (SAC305 or SnAg) are reflowed, usually peaking at 245–260 °C. Thermocompression bonding at 300 °C with copper pillars gives a more robust joint for automotive-grade parts.
  4. Underfill dispense and cure – Capillary underfill flows under the die to relieve CTE mismatch stress between silicon (CTE ≈ 2.6 ppm/°C) and organic substrate (CTE ≈ 12–17 ppm/°C). Cure is typically 150 °C for 30–60 minutes.
  5. Overmold or lid attach – A low-Dk mold compound (Dk ≈ 3.0–3.8) encapsulates the die side. Because the mold sits above the antenna layer, its Dk directly detunes the patch; the antenna must be designed with the mold’s Dk already accounted for. Alternatively, a metal or plastic lid with an air cavity avoids this detuning but increases cost and height.
  6. BGA ball attach – Solder balls are placed on the bottom land pads and reflowed.
  7. Singulation – The panel is diced into individual modules.
  8. Final test – Each module is tested over-the-air (OTA) on an automated handler for EIRP, beam patterns, and S-parameters.

Not sure which assembly steps you can handle in-house versus outsource? The quiz below walks you through your current capabilities and recommends a process split.

Loading Assembly Process Quiz…

If the quiz flags flip-chip placement or thermocompression bonding as gaps, that is normal—most board-level assembly houses lack the ±5 µm placement accuracy and controlled-atmosphere reflow profiles these steps demand. Those steps are typically done at a semiconductor OSAT (outsourced assembly and test) facility, while BGA-level SMT onto the motherboard stays in a standard line.

Thermal Management

Flip-chip die bonder placing RFIC onto AiP substrate panel

A 77 GHz radar MMIC can dissipate 2–4 W in a package footprint under 12 × 12 mm. Without a clear thermal path, junction temperature rises fast. Key thermal strategies include:

  • Thermal vias under the die – A dense array of plated-through or filled microvias directly beneath the RFIC conducts heat to the BGA side. Via density of 20–30 vias per mm² is typical.
  • Exposed copper pad – A large copper pad on the BGA side, soldered to a thermal landing on the motherboard, can drop thermal resistance by 30–50 % compared with vias alone.
  • Thermally conductive underfill – Filled underfills with thermal conductivity of 1–2 W/m·K (versus 0.2 W/m·K for standard underfill) help, though they raise Dk slightly and must be characterized at mmWave frequencies.

Buy vs. Build: Module Sourcing Decision

Some teams buy a complete AiP module from a semiconductor vendor (TI AWR series, Infineon XENSIV, NXP TEF82xx), while others build a custom substrate around a bare RFIC die. The trade-offs are significant:

  • Buy – Faster time to market, pre-validated antenna performance, vendor-supported reference designs. Limited customization of beam pattern, polarization, or array size.
  • Build – Full control over element count, scan range, and form factor. Requires substrate design expertise, flip-chip assembly capability or OSAT partnership, and OTA test infrastructure. Development cost is typically six figures before first silicon.

Use the tool below to compare the total cost and timeline of buying an off-the-shelf module versus building a custom AiP substrate around a bare die for your volume and performance requirements.

Loading Buy vs Build Assembly Tool…

If the tool shows that your annual volume is under roughly 50 k units and your beam requirements match a commercial module’s spec, buying almost always wins on both cost and schedule. Above 200 k units, or when you need a non-standard array geometry, a custom build starts to pay back the NRE within one to two production years.

Mold Compound and Radome Effects

The material directly above the antenna layer—whether mold compound, a conformal coating, or a radome lid—shifts the patch resonant frequency downward and narrows bandwidth. A mold with Dk 3.5 sitting 0.2 mm above a 60 GHz patch can shift resonance by 3–5 %. Designers must co-simulate the mold in a full-wave solver (Ansys HFSS or CST) from the start, not add it as an afterthought. For deeper guidance on dielectric cover effects, see the discussion on antenna radome Dk and thickness budgeting.

Test and Validation

AiP modules are tested at three levels:

  1. Substrate-level RF probe – Before die attach, probe pads on the substrate verify feed-network loss and impedance. A calibrated probe station with GSG probes rated to 110 GHz is standard.
  2. Module-level OTA – After overmold, each module is placed in a compact antenna test range or near-field scanner. EIRP, EIS (equivalent isotropic sensitivity), and beam-steering accuracy are measured. Refer to IEEE Std 145 antenna definitions for terminology.
  3. System-level on-board – The module is soldered onto the application board and re-tested to catch motherboard coupling, ground-plane truncation, and connector mismatch effects.

Common Pitfalls and How to Avoid Them

  • Ignoring mold Dk tolerance – Mold compound Dk can vary ±5 % lot to lot. Design the patch bandwidth wide enough to absorb this shift, or specify a tighter-tolerance mold grade.
  • CTE mismatch cracking – Without adequate underfill, thermal cycling (–40 to +125 °C automotive profile) cracks solder bumps within 500 cycles. Underfill is not optional.
  • Ground-plane discontinuity – Routing a signal trace across the antenna ground plane creates a slot that radiates and detunes elements. Keep the ground plane under the array unbroken; route all signals on inner layers or outside the array footprint.
  • Insufficient via fencing – Without via fences between elements, surface waves couple adjacent patches and raise sidelobe levels by 3–5 dB. Place via rows at λ/10 spacing around each element or sub-array.

Frequently Asked Questions

What frequency range is antenna in package used for?

AiP is most common from 24 GHz to 77 GHz (automotive radar, 5G NR FR2, WiGig). Research prototypes extend to 140 GHz and beyond using glass or silicon interposers.

Can AiP achieve more than 20 dBi gain?

Yes, but it requires a large array (e.g., 16 × 16 elements) that may exceed typical package sizes. A 4 × 4 patch array at 60 GHz on an organic substrate achieves roughly 16–18 dBi. For higher gain in a compact form, consider a package-level transmitarray lens placed above the AiP.

Is LTCC better than organic for AiP?

LTCC offers superior hermeticity and stable Dk over temperature, making it preferred for military and space applications. Organic laminates win on cost and scalability for commercial volumes above 10 k units. Neither is universally better; the choice depends on environment, volume, and loss budget.

How does AiP handle heat from high-power MMICs?

Thermal via arrays, exposed copper pads on the BGA side, and thermally conductive underfill are the primary techniques. Junction-to-board thermal resistance below 5 °C/W is achievable with careful via planning on organic substrates.

What EDA tools support AiP co-design?

Full-wave solvers (HFSS, CST, FEKO) model the antenna and package electromagnetics. These are coupled with package layout tools (Cadence Allegro Package Designer, Siemens Xpedition) for physical design and DRC. Co-simulation between the two is essential because standalone antenna simulation without the package geometry can miss resonance shifts of several percent.

Ready for a quotation?

Send your requirements for a material, stackup and manufacturing review. Price and schedule are confirmed for your project.

Upload Gerber