Channel Insertion Loss Budget and Maximum Reach Calculator
Enter the data rate, your loss budget and the laminate. The calculator splits dielectric from conductor loss, subtracts the fixed cost of connectors, vias and packages, and tells you how many inches of trace are actually left — and which lever will buy you more.
| Where the budget goes | dB | Share | Note |
|---|---|---|---|
| Dielectric loss | scales with frequency and Df — the material choice | ||
| Conductor loss | scales with √f, width and roughness | ||
| Connectors and vias | fixed cost, spent before any trace | ||
| Package and die | not yours to change |
Runs entirely in your browser. Nothing is uploaded, stored, or sent anywhere.
Reach is a budget, not a distance
There is no such thing as “how far can I route this”. There is only a loss budget — set by the standard or the SerDes — and a list of things spending it. Connectors and vias spend theirs before a single inch of trace is routed, the package and die spend theirs at both ends, and whatever remains divided by the loss per inch is your actual reach.
That framing changes decisions. If four connectors and vias plus package loss have already consumed 5 of a 20 dB budget, upgrading the laminate is competing for the remaining 15, and removing one connector may buy more reach than changing material.
Two loss mechanisms that scale differently
αdielectric ≈ 2.3 · fGHz · √Dk · Df dB/inch — linear in frequencyαconductor ∝ √f · roughness / width — skin effect, square root of frequency
This difference is the single most useful thing on this page. Dielectric loss is proportional to frequency; conductor loss only to its square root. Double the data rate and dielectric loss doubles while conductor loss grows by about 1.4×. So the balance shifts with speed: at 1 Gbps copper geometry dominates and cheap FR-4 is fine, while somewhere between 10 and 25 Gbps the laminate becomes the deciding factor and no amount of trace widening rescues a lossy material.
The tool shows which one currently dominates and says which lever to pull, because pulling the wrong one is expensive: low-Df laminates carry real cost and lead-time penalties, and they are wasted money on a channel that is conductor-limited.
Roughness is the specification nobody reads
Copper foil is deliberately roughened so it adheres to the laminate, and at high frequency current flows in the skin, following that roughness rather than a straight path. The effective path lengthens and loss rises — by up to about 2× for standard foil compared with very low profile copper at the same geometry.
It is specified as a foil type rather than a number, and it is easy to leave to the fabricator’s default. On a marginal channel it is often the cheapest single improvement available, because it changes no geometry and no stackup — only the material line on the fab drawing.
Nyquist, and why the third harmonic matters anyway
Loss budgets are quoted at Nyquist, half the NRZ data rate, because that is the fundamental of the fastest alternating pattern. It is the right place to compare channels. But a real edge carries energy well above it, and the loss keeps rising, so a channel that just meets its budget at Nyquist still has visibly degraded edges. This is why equalisation exists and why “meets the spec mask” and “has margin” are different statements.
What is not in the number
- Impedance discontinuities, which cause reflections rather than loss. A channel can meet its insertion-loss budget and still fail on return loss, and via stubs are the usual culprit — the via stub calculator covers that.
- Crosstalk from neighbouring aggressors, which behaves as noise, not attenuation.
- Fibre weave effect, where a trace running along a glass bundle sees different effective Dk than one running over resin — a real source of intra-pair skew on loosely woven laminates.
- Equalisation. Transmit de-emphasis and receive CTLE/DFE recover a substantial amount of loss, which is why published budgets are often far larger than an unequalised eye would survive.
More free tools
Each of these runs entirely in your browser. Nothing is uploaded, stored or sent anywhere, and none of them needs an email address.
See all 28 engineering tools →
New tools and notes, once a month at most
We add tools here fairly often and write up the things worth writing up — a stackup that behaved oddly, a standard that turned out to be obsolete, a calculator that was quietly wrong. Join and you get told when something new lands.
One email a month at the very most, and usually less. No drip sequence, no sales cadence, no sharing your address with anyone. Unsubscribe whenever you like.