When traces become transmission lines
| Rule of thumb | Trace length > (t_r × 0.5) / (prop delay per unit) |
|---|---|
| Example | Signal with 1 ns rise time on FR-4 (~6 in/ns in outer): any trace > 3 in acts transmission-like |
| Characteristic Z_0 | Depends on trace width, height above plane, dielectric. Common: 50 Ω single-ended, 100 Ω diff |
Reflections
| Mismatch | Γ = (Z_L − Z_0) / (Z_L + Z_0) |
|---|---|
| Open end | Γ = +1 (full reflection, doubles voltage) |
| Short end | Γ = −1 (full inversion) |
| Termination | End at Z_0 for no reflection. Can also source-terminate. |
Termination styles
| Style | Where | Power | Use |
|---|---|---|---|
| Series | At source | Low | Point-to-point; DDR command lanes |
| Parallel | At far end | High | SSTL, LVDS (AC-coupled) |
| Thevenin | Split at far end | Medium | Bussed topology |
| AC | Cap + R at far end | Low | DDR4/5 differential |
| Differential | Across pair | Varies | USB, Ethernet, HDMI, MIPI |
Crosstalk
- Keep 3W spacing (3× trace width) between parallel traces.
- Avoid parallel runs > 10 mm on high-speed signals.
- Use guard traces or layer separation for critical signals.
- Break 90° turns into two 45° segments or arcs.
Power integrity
- Provide tight decoupling — cap close to IC power pins, continuous ground return.
- Avoid split planes under high-speed signals — return currents are forced to detour.
- Use bulk cap (10 µF) near regulators, distributed ceramics (100 nF) at ICs.
Worked examples
Critical length: on outer-layer FR-4 microstrip the delay is about 150 ps per inch. For a 1 ns edge, half the rise time corresponds to 0.5 / 0.15 = 3.3 inches, which matches the rule of thumb above. A more conservative design rule treats a trace as a transmission line once its delay exceeds one sixth of the rise time, which is 1 ns / 6 / 0.15 = 1.1 inches. Slowing the edge to 3 ns triples the length before termination is needed.
Reflection: a 50 ohm trace ending in a 75 ohm input has Γ = (75 − 50) / (75 + 50) = 0.2, so 20% of the incident step is reflected. Of a 3.3 V step, that is a 0.66 V echo.
Series termination: a driver with 20 ohms output resistance sends 3.3 V × 50 / (20 + 50) = 2.36 V onto a 50 ohm line. At an open (high-impedance) end the voltage doubles to 4.7 V, which exceeds the 3.3 V rail, and the part of the wave that reflects back sees Γ = (20 − 50) / 70 = −0.43 at the source and rings. Adding 30 ohms in series makes the source impedance 50 ohms. The line launches half the swing (1.65 V), the open end doubles it to the full 3.3 V, and the returning wave is absorbed at the source with no ringing.
How to check a design
- Measure with a short ground spring, not the long clip lead. A long ground lead creates ringing on the scope that is not on the board.
- Scope bandwidth must exceed the signal. To see a 1 ns edge faithfully you need about 350 MHz of probe-and-scope bandwidth (0.35 / rise time), and more for reliable overshoot readings.
- Check the return path. Whenever a fast trace changes layers, put a ground via next to the signal via so the return current has a short route. A missing stitching via is a common source of EMI failure.
- Simulate first when it counts. IBIS models and a field solver find problems that a rule of thumb misses, especially on DDR and multi-drop buses.