Define the measurement boundaries
Draw where the network starts and ends. A bare transmission-line coupon, a connector-to-connector channel, and a package-inclusive simulation represent different systems. If fixtures or launches are removed mathematically, record the method and the structures used to support it. A graph without these details can look precise while remaining unsuitable for a design decision. For differential links, confirm that the reported quantity is the appropriate differential transmission response rather than an unrelated single-ended result.
Separate the contributors before optimizing
AMD's material guidance compares loss with specified material and conductor assumptions. Use that approach to establish a baseline line model, then add transitions and connectors. A smooth loss trend and a localized transmission notch suggest different investigations; neither should automatically be corrected by choosing a lower Df. Check whether the problematic frequency region aligns with a via structure, a launch, or the distributed line. The remedy should address the contributor that consumes margin.
Compare equal electrical structures
When evaluating materials, maintain the target impedance and document the width changes required by each candidate. Equal physical widths do not necessarily create an equivalent comparison. Include copper profile and actual dielectric thickness in the model. If a shorter route is feasible, compare its benefit with the proposed material change. An architectural adjustment can improve the channel while also reducing sensitivity to uncertain material inputs, but confirm that it does not introduce new transitions or routing congestion.
Use coupon data to improve the next decision
Correlate the electrical result with the coupon's measured dimensions and material record. Keep launch and fixture uncertainty separate from the line model. If the result fails the agreed requirement, first establish whether the coupon represents the production structure accurately. Then investigate the specific discrepancy. Preserve raw data and model revisions so the next board does not repeat the same correlation work. A single passing trace does not establish compliance for every route length, layer, and transition on the product.
A continuous reference keeps the return path close
Dielectric · geometry + material
Core or build-up dielectric
Illustrative only. Copper thickness, dielectric construction and via spans require a reviewed stackup.
Common questions
Is Df the same thing as insertion loss?
No. Df is a dielectric material property under stated conditions. Insertion loss belongs to a defined network and includes additional contributors.
Can I scale every measured loss curve directly by route length?
Only the distributed portion may support that approximation under matching conditions. Launches, connectors, and discrete discontinuities do not generally scale with trace length.
References & further reading
Method & assumptions · General design guidance. Validate the proposed construction and acceptance criteria for your project.