Engineering tool
Channel loss budget estimator
Estimate the stripline insertion loss of a PCB channel from 10G NRZ to 224G PAM4, split into dielectric and conductor loss, and compare it with your channel budget. It is a first-order model for comparing laminates and copper foils, and it opens with a 112G PAM4, 20 cm Megtron 6 example checked against the IEEE 802.3ck 28 dB channel budget.
Inputs
Budget defaults follow IEEE backplane (KR1) die-to-die limits: 112G PAM4 28 dB (802.3ck), 224G PAM4 40 dB (802.3dj); enter your own budget for other rates or for a PCB-only allocation. Editing Dk or Df switches the laminate to "Custom".
Result
First-order estimate, ±20%, for material comparison only.
Laminate comparison at the same length
The same channel on one laminate from each of four loss grades, holding length, rate, trace width and copper roughness constant.
| Laminate | Dk | Df | Dielectric dB | Conductor dB | Total dB | Budget used |
|---|
Formula and assumptions
Nyquist frequency f is half the baud rate at the IEEE 802.3 electrical lane rates: 10.3125 Gb/s NRZ → 5.16 GHz, 25.78 Gb/s NRZ → 12.89 GHz, 53.125 Gb/s PAM4 → 13.28 GHz, 106.25 Gb/s PAM4 → 26.56 GHz, 212.5 Gb/s PAM4 → 53.13 GHz. PAM4 carries 2 bits per symbol. (OIF CEI-112G quotes 112 Gb/s, i.e. 28 GHz; the difference is under 6%.)
α_d (dB/in) ≈ 2.3 · f(GHz) · Df · √Dk α_c (dB/in) ≈ (0.54 · √f(GHz) / w(mil)) · K_rough K_rough = 1 + (2/π) · atan(1.4 · (Rq/δ)²), δ(µm) = 2.09 / √f(GHz) total (dB) = (α_d + α_c) · length(in), 1 in = 2.54 cmThe dielectric term is the standard first-order approximation for a homogeneous dielectric. The smooth-copper conductor term has the √f / w form of Bogatin's rule of thumb α_c ≈ 36·√f / (w·Z0), with the constant calibrated to a published field-solver result (0.404 dB/in of conductor loss for a 5 mil, 100 Ω stripline pair at 14 GHz, Signal Integrity Journal); the plain rule of thumb at Z0 = 50 Ω gives 0.72, about 33% higher. Copper roughness is applied with the Hammerstad–Jensen correction, which rises with frequency as the skin depth δ shrinks and saturates at 2× for foil much rougher than δ. As an independent check, for a 5 mil trace on FR-4 at 4 GHz with standard foil the model gives about 0.80 dB/in, against about 0.78 dB/in reported by Bogatin for a PCIe Gen3 channel of that construction.
RMS roughness per foil class is a planning assumption: conventional ED foil measures about 1.2–4 µm Rq on its treated side, very-low-profile foils about 0.47–0.7 µm, and HVLP3-class foil is specified at Rz ≤ 1.1 µm; the Rq values in the selector sit inside those ranges. Hammerstad under-predicts loss for the roughest foils above about 20 GHz, where a Huray-model field solver is the better tool. Laminate Dk and Df are the supplier's datasheet values at 10–14 GHz and are held constant across frequency; real laminates drift slightly upward in Df with frequency. Via, connector, package and crosstalk losses are not included.
Budget defaults: IEEE 802.3ck 100GBASE-KR1 allows 28 dB at 26.56 GHz and IEEE 802.3dj 200GBASE-KR1 allows 40 dB at 53.125 GHz, both die-to-die including packages and connectors. The PCB trace gets only part of that; if you have a trace allocation, enter it as the budget.
Budget status: green up to 85% of the budget, yellow above 85%, red above 100%. When a 112G or 224G channel exceeds the budget, the estimator reports the length at which this construction would use the full budget, as the point beyond which a retimer or a lower-loss laminate is needed.
Result accuracy is about ±20%: use it to compare laminates and foils, not to sign off a channel. Channel sign-off needs a field-solver or measured S-parameter model of the actual stackup.
Request a stackup review
A stackup review returns named CCL and glass constructions and the fabricator's impedance calculation for your longest channel, usually within five business days.
References
- E. Bogatin, Signal and Power Integrity – Simplified; and E. Bogatin, "Loss in a channel: Rule of Thumb #9", EDN (edn.com/loss-in-a-channel-rule-of-thumb-9/).
- Signal Integrity Journal, "How to Reduce Attenuation in a Differential Channel" (signalintegrityjournal.com/articles/1734).
- Supplier datasheets: Panasonic MEGTRON 6 R-5775(N) (Apr 2022), MEGTRON 7 R-5785 (2022), MEGTRON 8 R-5795; Isola Tachyon 100G and Astra MT77; EMC EM-890K and EM-892K (Jul 2024); Rogers RO3003; Isola 370HR Dk/Df tables.
- IEEE 802.3ck (100GBASE-KR1, 28 dB at 26.56 GHz) and IEEE 802.3dj (200GBASE-KR1, 40 dB at 53.125 GHz).
- E. O. Hammerstad and Ø. Jensen, "Accurate models for microstrip computer-aided design", IEEE MTT-S 1980 (roughness correction); IPC technical paper "Effect of Conductor Surface Roughness upon Measured Loss and Extracted Values" (foil Rq values).