Loading…
Free tool · Channel budget
PCB insertion loss calculator: dB per inch, budget and maximum length
Turn a stackup, a laminate and a data rate into insertion loss over the run, the share each loss mechanism takes, and the longest length the budget allows.
The short answer
Insertion loss is the attenuation a signal suffers along a trace, measured in decibels at a stated frequency. It has two causes that scale differently: dielectric loss rises in proportion to frequency and follows the laminate's loss tangent, while conductor loss rises with the square root of frequency through skin depth and is multiplied by copper roughness. At 8 GHz on 0.2 mm FR-4 a typical 0.35 mm microstrip loses roughly 0.15 dB per inch, and a low-loss laminate removes much of the dielectric part.
Calculator
Insertion loss and channel budget calculator
Enter the cross-section, the laminate and the frequency the channel is judged at — normally the Nyquist rate, half the data rate. The tool returns attenuation per inch, the total over the run, the split between dielectric, strip and return path, and the longest length the budget buys. The models are the ones behind the impedance calculator, so the loss here is the loss that calculator's panel reports.
A laminate fills Dk, Df and foil roughness together.
Line type: microstrip or stripline. Open the button for its units, why it matters and a typical value.
Cross-section and material
Laminate dielectric constant (Dk). Open the button for its units, why it matters and a typical value.
Loss tangent (Df) at the reference frequency. Open the button for its units, why it matters and a typical value.
Dielectric height under the trace. Open the button for its units, why it matters and a typical value.
Trace width. Open the button for its units, why it matters and a typical value.
Copper thickness. Open the button for its units, why it matters and a typical value.
Copper surface roughness. Open the button for its units, why it matters and a typical value.µm
Channel
Frequency the loss is evaluated at. Open the button for its units, why it matters and a typical value.GHz
Routed length of the channel. Open the button for its units, why it matters and a typical value.
Insertion-loss budget at the evaluation frequency. Open the button for its units, why it matters and a typical value.dB
Judge the channel at its Nyquist rate: half the data rate. 16 GT/s is 8 GHz; 32 GT/s is 16 GHz.
Insertion loss against the budget
Dielectric and conductor loss separated, with the length the budget allows.
6.42 dB against a 6.0 dB budget.
- Loss at frequency
- 6.42dB
- Per inch
- 0.816dB/in
- Longest run
- 187mm
- Flight time
- 1,194ps
- Delay
- 5.970ps/mm
- Z0
- 50.90Ω
107% of the budget.
The length the whole budget buys.
Where the loss goes
| Term | dB/in | Share |
|---|---|---|
| Dielectric | 0.442 | 54% |
| Strip (roughened) | 0.255 | 31% |
| Return path | 0.119 | 15% |
| Roughness | 1.76× | skin depth 0.74 µm |
What the budget leaves out
Two vias, the connector, the package and the receiver's own parasitics are not in this number. A trace that spends 80% of the budget has almost nothing left for them, which is why the tool calls that tight rather than passing.
Get an Instant QuoteEmail a layout engineer
Free, no account. The engineer reviews the inputs before a fixed quote is issued.
Want the derivation? Jump to the math behind the calculator.
The math
The formulas, with the working shown
Each equation is rendered with KaTeX, which also emits a MathML twin for screen readers and answer engines, explained in plain language, and copyable as LaTeX from the button in its corner. The variable list says what every symbol means; the model-limit note says where the formula stops being trustworthy.
Insertion loss of a run
Loss is linear in length, so a budget turns directly into the longest run it allows.
- A_{\text{dB}}
- total insertion loss, dB
- \alpha_{\text{dB/mm}}
- attenuation per millimetre at the evaluation frequency
- L
- routed length, mm
Attenuation per unit length is a property of the cross-section; length is the only thing routing changes. That is why the same 0.15 dB per inch is fine at 200 mm and fatal at 800 mm.
The tool inverts the relation to report the longest length that still fits a budget, which is the number a router can act on.
Model limits
This counts the trace only. Two vias, the connector and the package each add their own loss, and on a short channel those can exceed the trace.
Dielectric attenuation
Dielectric loss grows linearly with frequency and with the laminate's loss tangent.
- f
- frequency, Hz
- \tan\delta
- loss tangent at that frequency
- q
- dielectric filling factor
The square root of the effective permittivity is how much the field is slowed; pi f over c turns that into a phase constant; the loss tangent is the fraction lost per radian.
The filling factor q is (εeff − 1)/(εr − 1): the part of the field that is in air has no dielectric loss, and a stripline buries all of its field so q is 1.
Model limits
Loss tangent is itself frequency dependent, and the calculator evaluates a causal wideband model anchored at the datasheet frequency rather than holding Df constant.
Conductor attenuation
Conductor loss rises with the square root of frequency, and roughness multiplies it.
- R_s
- surface resistance of copper at the frequency
- K
- roughness multiplier
- w_{\text{ret}}
- effective return-path width
Surface resistance is copper resistivity over skin depth, so it grows with the square root of frequency. The two reciprocal widths are the strip and the return path: both carry current and both dissipate.
The roughness factor is Hammerstad's multiplier, 1 + (2/pi) arctan(1.4 (Δ/δ)²). When the foil roughness equals the skin depth it is already 1.64, which is why rough copper costs 30–50% at 10 GHz.
Model limits
The return-path width is approximated as w + 2h for microstrip and w + b for stripline. A split plane or a thin neck is worse than this model says.
Skin depth
The depth the current occupies falls as the square root of frequency, which is why conductor loss grows that way.
- \rho
- copper resistivity, 1.724e-8 ohm metre
- \mu_0
- permeability of vacuum
- \delta
- skin depth, metres
At 1 GHz the skin depth in copper is about 2 µm; at 10 GHz it is about 0.66 µm. Once it is smaller than the foil's surface roughness the current has to follow the profile, and the effective resistance stops falling as fast.
The tool reports the skin depth at the evaluation frequency next to the roughness multiplier, so the two can be compared directly.
Model limits
The formula assumes a smooth good conductor. Above roughly 10 GHz roughness dominates and the smooth-copper depth overstates how much metal carries current.
Theory
Two mechanisms, two different slopes
Dielectric loss comes from the laminate's loss tangent: the field does work on the resin as it oscillates and the energy leaves as heat. It rises linearly with frequency, so it dominates a long channel at a high rate and it is the one term a better laminate actually fixes.
Conductor loss comes from the copper's surface resistance. At a few gigahertz the skin depth is below a micrometre, so the current runs in the surface profile of the foil: rough copper lengthens the path and adds loss. It rises with the square root of frequency, which is why the two terms swap dominance as the rate climbs.
- Dielectric loss is linear in frequency and set by the laminate's Df.
- Conductor loss is proportional to the square root of frequency and multiplied by roughness.
- The return path is a real term: on a thin dielectric it is a third of the copper loss.
- A wider trace cuts conductor loss but changes the impedance unless the dielectric grows with it.
Budget
The budget is what the receiver has left
A link closes when what the transmitter, the channel and the receiver can tolerate adds up to less than the margin the standard provides. The receiver's equalisation buys back a fixed amount, the connectors and vias spend some, and the trace gets the remainder. The tool calls 80% of the budget tight because two vias, a connector and a package are not in this number.
Length is the only input a layout can still change after the stackup is frozen, which is why the answer is a length as well as a loss. If the run exceeds the budget, the options are a better laminate, a wider trace on a thicker dielectric, or a different route.
When the channel has margin the closed form is enough. When it does not, the loss has to come from an extraction on the actual stackup; that is the SI/PI verification service engagement.
FAQ
Common questions
The Nyquist rate for a non-return-to-zero link: half the data rate. A 16 GT/s lane is judged at 8 GHz and a 32 GT/s lane at 16 GHz. Judging a channel at its clock or its bit rate instead of the Nyquist rate is the most common mistake in a loss budget, and it understates the loss by a large factor.
Want this checked on your board?
Send the stackup and interface list. We will check the loss budget against the real layout before the quote is fixed.