What is crosstalk, and what are aggressor and victim nets?
From PDVerse PnR Interview Handbook · pdVerse Mentor Guide
Short Answer
Crosstalk is unwanted coupling between two wires that run next to each other. When one wire, the aggressor, switches quickly, the coupling capacitance between them pushes a small voltage onto its neighbour, the victim. If the victim is quiet, that shows up as a glitch; if the victim is also switching, it changes the victim's delay.
Technical Explanation
- Two parallel wires separated by a thin insulator form a capacitor. That coupling capacitance grows with how long they run side by side and shrinks as spacing increases.
- The aggressor is the net doing the switching, usually one with fast edges and high activity. The victim is the net being disturbed. A net can be an aggressor to one neighbour and a victim of another.
- On a quiet victim, a fast aggressor edge induces a glitch at the victim's receiver regardless of the victim's logic value. If the glitch crosses the receiver's switching threshold, a wrong value can propagate and even flip a register.
- On a switching victim, crosstalk changes delay. Opposite-direction switching slows the victim, which hurts setup. Same-direction switching speeds it up, which hurts hold.
- Modern metal is tall and narrow, so the sidewalls facing each other are large. That is why coupling is a bigger share of wire capacitance at advanced nodes than it used to be.
- Common fixes are more spacing, shielding with a power or ground wire, matching driver strengths, or moving one net to another layer. Most of these cost routing resources, so they are used on selected nets such as clocks.
- Crosstalk only becomes visible once real routing exists, because it depends on which wires ended up next to each other. It is analyzed on routed parasitics with signal integrity analysis enabled.
Formula Or Decision Rule
- Basic coupled noise peak:
V_n = Vdd · Cx / (Cx + Cv) Cx= coupling capacitance between the two wires,Cv= victim wire's own capacitance to ground,Vdd= the aggressor's switching swing.- This is the simple charge-sharing model. It overestimates noise because it ignores the victim driver holding its wire, but it shows the key point: more coupling, or less victim capacitance, means a bigger glitch.
Common Mistake
The Trap: Treating crosstalk as a noise-only problem and ignoring its effect on timing. A junior checks that no glitch crosses the threshold and signs off, but misses that the same coupling adds 20 ps of delta delay to a setup-critical path when the neighbour switches the opposite way.
Follow-up Question & Model Response
"Why does a stronger victim driver reduce crosstalk noise?"
Candidate Model Response: The simple model assumes the victim wire floats while the charge is shared. In reality the victim's driver is holding the wire at its logic level and pulls the glitch back. A stronger driver, meaning lower resistance, pulls it back faster, so the peak is smaller and shorter. The fuller model includes the ratio of aggressor to victim resistance for this reason, and it is why upsizing a victim driver is a standard crosstalk fix alongside spacing and shielding.
Practical Example
Tapeout Scenario: A victim net has 12 fF of capacitance to ground and runs 300 um next to a busy bus wire, giving 6 fF of coupling (illustrative). With a 0.8 V aggressor swing, the simple model gives V_n = 0.8 × 6 / (6 + 12) = 0.27 V. That is large enough to worry about on a quiet control net with a 0.4 V threshold. Spreading the two wires to double spacing roughly halves Cx to 3 fF, and the estimate drops to 0.8 × 3 / 15 = 0.16 V. The real glitch is lower because the victim driver fights back, but the direction of the fix is the same.
PnR Flow Mentor Guide
Master the Physical Design Implementation Flow
Read the complete 8-chapter PnR Flow Mentor Guide free on the web — library setup through placement, clock tree synthesis, routing, chip finishing, hierarchical implementation, and ECO, all the way to stream-out.
Continue practising