Automotive radar operates in the 76 to 81 GHz band, and at those frequencies the printed circuit board stops being a carrier for components and becomes part of the antenna. Substrate choices that are inconsequential at 6 GHz produce measurable range and angular resolution differences at 77 GHz. Worse, the automotive temperature range turns a property most RF designers rarely examine into the dominant selection criterion.
Dielectric constant stability, not dielectric loss
The instinctive priority at millimetre wave is low loss, and loss does matter. But an automotive radar module must hold its performance from roughly −40°C to +125°C, and every laminate’s dielectric constant drifts with temperature. On a patch antenna array, a shift in Dk shifts the resonant frequency and the beam pattern. A design tuned at room temperature that detunes on a cold morning is a functional safety problem, not a performance nuance.
This is the reason RO3003 dominates automotive radar despite RT/duroid 5880 having a lower dissipation factor. RO3003 is a ceramic-filled PTFE with an exceptionally low thermal coefficient of dielectric constant, holding Dk near 3.00 ± 0.04 across the automotive range. The purpose-developed RO3003G2 variant further reduces loss at millimetre-wave frequencies while retaining that thermal stability. Selecting on Df alone leads you to the wrong material.
The magnitude is easy to underestimate. A patch antenna’s resonant frequency scales inversely with the square root of the effective dielectric constant, so a Dk drift of even one percent moves resonance by roughly half a percent — several hundred megahertz at 77 GHz. Against an antenna bandwidth that may itself be only a few percent, that is a meaningful fraction of the design margin consumed by temperature alone.
Copper roughness becomes a first-order effect
Skin depth in copper is approximately 0.65 micrometres at 10 GHz. At 77 GHz it falls to roughly 0.24 micrometres. Once skin depth approaches the scale of the copper surface profile, current is forced to follow a longer path along that roughness, and conductor loss rises accordingly.
The practical consequence is that foil selection is not a detail. Standard electrodeposited foil with a typical tooth profile can add loss comparable to the dielectric contribution at these frequencies. Very-low-profile and rolled-annealed foils reduce it substantially. Any fabricator quoting a 77 GHz board without asking which foil profile you want has not built one before, and the difference is large enough that it should appear explicitly in your fabrication notes rather than being left to the house default.
There is a trade-off to acknowledge. Smoother foil bonds less strongly, because the tooth profile that costs you loss is also what mechanically keys the copper to the laminate. Peel strength drops as profile drops, which matters for fine features and for any part of the board that sees mechanical stress. This is a genuine engineering decision rather than a case of smoother always being better.
Where the loss budget actually goes
It helps to separate the loss contributions rather than treating insertion loss as a single number from the datasheet. Three mechanisms dominate at 77 GHz, and they respond to entirely different design decisions.
Dielectric loss scales with the dissipation factor and with frequency, and it is the term the material datasheet describes. Conductor loss scales with the square root of frequency in smooth copper, but rises faster once surface roughness becomes comparable to skin depth — which, as above, it does at these frequencies. Radiation loss is the term most often forgotten: an open microstrip structure radiates increasingly as the substrate thickness approaches a meaningful fraction of a wavelength, and at 77 GHz that threshold arrives at thicknesses that seem entirely ordinary on a low-frequency board.
The practical implication is that thinner substrate is generally better at millimetre wave, up to the point where trace widths become too narrow to etch repeatably. That trade-off between radiation loss and etch tolerance is worth working through explicitly with your fabricator, because the answer depends on their actual line-width capability rather than on a general rule.
Surface finish: avoid the nickel
The same argument applies with more force here than anywhere else. Electroless nickel immersion gold places a ferromagnetic, comparatively lossy nickel layer directly in the path of surface currents. At 77 GHz this is not a marginal penalty. Immersion silver or a nickel-free alternative is the standard approach for radar antenna surfaces, and the finish should be specified deliberately rather than left to the fabricator’s default.
Transitions and via design
At millimetre wave, the discontinuities between structures often cost more than the structures themselves. A via transition that behaves as a simple connection at 6 GHz becomes a resonant stub at 77 GHz, and the launch from a package pad onto a transmission line can dominate the loss budget of a short trace.
Grounded coplanar waveguide deserves particular mention here. At millimetre-wave frequencies, conventional microstrip becomes prone to radiation and surface-wave modes, and its dimensions shrink to the point where etch tolerance dominates impedance. A grounded coplanar structure with a well-designed via fence confines the field more effectively and is more tolerant of the etching variation any real process produces. The via fence pitch is a real design parameter rather than a drawing convention, and it needs to be tight relative to the wavelength in the substrate.
These structures depend on registration and drill position accuracy holding to tighter limits than general production work. It is worth confirming the fabricator’s actual positional tolerance rather than the figure on their capability sheet, which usually describes a best case rather than a routine one.
What the automotive supply chain adds
Beyond the RF engineering, automotive programmes bring requirements that eliminate a large share of otherwise capable suppliers.
• IATF 16949 quality management, which is a substantially different audit from ISO 9001 alone.
• Production part approval process documentation, including dimensional results and process capability studies.
• Material lot traceability maintained for the life of the programme, not merely at delivery.
• Formal change control, so that a laminate or process substitution cannot occur without notification and re-qualification.
• Capacity and continuity commitments across a multi-year production life.
These are procurement questions as much as technical ones, and they should be settled during supplier selection rather than after design freeze. When shortlisting a Rogers PCB manufacturer for radar work, ask for the certification scope document rather than a logo on a website, and confirm that the certified scope covers the specific facility that would build your boards.
Stackup and construction notes
| Decision | Common approach | Why |
| Antenna layer laminate | RO3003 / RO3003G2 | Dk stability across −40 to +125°C |
| Digital section | FR-4 or RO4350B | Cost; hybrid construction |
| Copper foil | Very low profile or rolled | Skin depth ~0.24 µm at 77 GHz |
| Surface finish | Immersion silver | Avoids lossy nickel layer |
| Impedance structure | Grounded coplanar waveguide | Better mode control at mmWave |
Qualification before volume
Build a qualification lot from the actual design, then measure across temperature rather than at ambient only. Compare the measured resonance shift against your simulation of the material’s thermal Dk coefficient. Then repeat the build from a different material lot several weeks later and compare. A Rogers PCB supplier running mid-volume production with in-house testing can usually support this sequence without a separate tooling charge, which is worth confirming during quotation rather than discovering afterwards.
The failure mode this catches is subtle and expensive: a design that works perfectly on the first lot and marginally on the third, because material variation invisible at room temperature becomes visible at the edge of the operating range. Finding that during qualification costs a few thousand units of schedule. Finding it after a production release costs considerably more, and in an automotive programme it may cost a field action.
Moving into volume
Radar boards are usually small, which means panel utilisation and array yield drive unit cost far more than the laminate price per square metre. A design that fits forty per panel instead of thirty-two changes the economics of a multi-year programme considerably, and the board outline is normally still adjustable when this is discovered early. It rarely is later.
Yield behaviour at millimetre wave is also different from conventional work. Boards do not usually fail outright; they fail marginally, with an antenna pattern slightly outside specification or an insertion loss a few tenths of a decibel high. That makes end-of-line test strategy a design consideration rather than a manufacturing afterthought, and it is worth agreeing with the fabricator during qualification exactly which measurements they will make on every panel and which they will make on a sample basis.
Closing
For 77 GHz radar, rank your material criteria in this order: dielectric constant stability across temperature, then conductor loss including copper profile, then dielectric loss, then cost. That ordering is different from almost every other RF application, and it is the reason radar programmes converge on ceramic-filled PTFE rather than the lowest-Df laminate on the datasheet.
