How RF Engineers Select PCB Materials Using VNA Insertion Loss Measurements
Extracting per-unit-length loss: the Delta-L method
Delta-L is the standard algorithm signal-integrity teams use to extract per-unit-length insertion loss from VNA measurements on coupons of different lengths. Subtracting the shorter coupon’s insertion loss from the longer coupon’s insertion loss cancels fixture and launch parasitics, leaving only the loss of the added trace. Divide by the length difference and you get loss per inch as a function of frequency.
One accuracy note: the noisiness of the extracted curve at low frequencies depends on the length difference between your two structures. A 1-inch delta (5-inch vs. 6-inch coupons) produces more noise below a few GHz than a 6-inch delta (2-inch vs. 8-inch) would. If the evaluation covers frequencies below 5 GHz, check whether the coupon set was designed with that floor in mind.
Read the frequency slope, not just the absolute loss number
Once you have loss-per-inch curves for each material and layer, the shape across frequency reveals more than the loss at any single point in your target band.
Insertion loss has two main contributors that scale very differently with frequency:
- Dielectric loss scales roughly linearly with frequency. A loss/inch curve that climbs steeply and nearly linearly above a few GHz points to Df (dissipation factor) as the dominant mechanism.
- Conductor loss, driven by the skin effect and copper surface roughness, scales roughly as the square root of frequency. A curve that rises more gradually, relative to a linear slope, is conductor-limited.
Most boards operating above 5–10 GHz are dielectric-loss-dominated. That is why Df is the first number engineers compare on a laminate datasheet. At lower frequencies, or with very low-Df materials, copper roughness can take over as the limiting factor. In that case, switching copper foil type — from standard electrodeposited copper to very low profile (VLP) or rolled annealed (RA) — can reduce loss as much as changing laminates.
What the extracted curves tell you about your specific candidates
Plot all your loss/inch curves on the same axes and compare at the highest frequency your channel budget must meet. Materials that look similar at low frequency can separate sharply above 10–15 GHz.
A few things worth checking beyond the absolute loss value:
- How much does loss vary between layer 5 and layer 10 for the same material? Wide spread between layers often points to prepreg vs. core Df differences in the stack-up, not a flaw in the core laminate itself.
- Does the slope change abruptly at a specific frequency? Resonances from connector launches or via stubs in the test coupon can mimic a material characteristic. Cross-check against the coupon’s physical dimensions before concluding anything.
- Are the 70-ohm and 90-ohm structures giving consistent loss/inch for the same material? Normalized for geometry, they should track closely. Large discrepancies usually indicate a measurement artifact.
Common laminate candidates and what to expect
Standard FR-4 carries a Df around 0.02 at 10 GHz. For anything above roughly 1–2 GHz where loss matters, engineers move to purpose-built laminates. The commonly compared options in multilayer RF and high-speed digital work:
- Panasonic Megtron 6: Df around 0.002 at 10 GHz, Dk around 3.7. Processes similarly to FR-4 in most PCB shops. The default starting point for multilayer boards mixing RF and high-speed digital content.
- Rogers RO4350B: Df around 0.0037 at 10 GHz, Dk 3.48. Ceramic-filled hydrocarbon resin with very consistent Dk across temperature and frequency, which matters for impedance-critical designs. Higher cost and less fab-friendly in true multilayer constructions.
- Isola I-Tera MT40: Df in the low 0.004 range at 10 GHz. Designed to be more process-compatible than PTFE while reaching the low-Df territory needed for Gen 4 and Gen 5 PCIe and similar interfaces.
- PTFE-based laminates (Rogers RT/duroid, Taconic TLY): Df below 0.001. Designed for microwave and millimeter-wave work where dielectric loss must be minimized. Requires specialized drilling, handling, and bonding — most standard multilayer shops do not offer this as a routine service.
For 5–15 GHz multilayer work, Megtron 6 is where most teams start. Above 20–25 GHz, PTFE-based options pull clearly ahead, and the measured loss/inch curves will make that gap obvious.
Layer position and prepreg effects in a multilayer stack
Separating structures across layers 5 and 10 is a smart coupon approach. The extracted data will directly show how loss changes with depth in the stack. This matters because inner layers are surrounded by prepreg on both sides, and prepreg commonly has higher Df than the core material at the same frequency.
Most laminate suppliers publish separate Dk/Df values for their core and prepreg systems. If layer 10 consistently shows higher loss/inch than layer 5 for the same material family, prepreg Df is the most likely explanation. That is a prepreg selection problem, not a core laminate problem.
A practical workflow for making the call
The sequence most SI and RF teams follow once coupon data is in hand:
- Extract loss/inch vs. frequency for all candidates with Delta-L.
- Identify whether each candidate is dielectric-limited or conductor-limited at the target frequency using slope analysis.
- If dielectric-limited, rank by Df at your operating frequency using the measured curves rather than the datasheet alone. Datasheet Df values are measured at specific conditions per IPC-TM-650 methods (often 2.5.5.5c or 2.5.5.13), and those conditions may not match your actual operating frequency.
- If conductor-limited, evaluate copper foil type alongside the laminate. Two candidates with identical Df but different copper surface roughness can show meaningfully different loss/inch above 10 GHz.
- Assess manufacturability. Ceramic-filled and PTFE-based materials have specific drilling, bonding, and handling requirements that drive up cost and complexity in multilayer builds.
- Factor in cost and lead time. The laminate with the best measured loss is not the right answer if it adds significant cost and weeks to the build cycle.
The target is the least expensive, most manufacturable laminate whose measured loss/inch fits the channel budget at the operating frequency, not the one with the best number on paper.