FPGA signal integrity · Diagnosis

Diagnosis: SSN or crosstalk?

The fastest way to separate SSN from crosstalk is to change switching count, adjacency, and edge rate one at a time.

Observable signatures

SSN points to the bank/package/PDN

Noise scales with the number of simultaneous outputs.
Multiple otherwise unrelated pins in the same bank look noisy.
Quiet pins move during a large output-switching event.
Lower drive or slower slew improves many victims at once.

Crosstalk points to a local aggressor

One victim follows one or a few neighboring nets.
Failure depends on physical adjacency or routing layer.
Specific data patterns such as 0101/1010 are worse.
Spacing, termination, or moving one net changes the symptom.

Test sequence

Classify the failing net: same I/O bank as the switching group, adjacent on PCB/package/connector, or both.
Run trigger patterns: walking-1, walking-0, all-toggle, pseudorandom, quiet-bus, and real traffic.
Measure with a real reference: use a short ground spring/coax probe. A long ground lead can create the apparent bounce.
Change one variable: slew, drive, bank placement, disabled neighboring outputs, or source resistance.
Check vendor SSO tables: device family, package, I/O standard, VCCIO, load, and bank usage all matter.

Interpretation

If the problem tracks total simultaneous switching count, suspect SSN first.
If the problem tracks one physical aggressor, suspect crosstalk first.
If both are true, treat it as combined bank-wide bounce plus local coupling.

Sources: Microchip AN4848; AMD/Xilinx XAPP689; Altium, “Is it Simultaneous Switching Noise or Crosstalk?”; UBC, “Crosstalk-Aware Routing in FPGAs”; Design & Reuse, “How to reduce simultaneous switching output noise with a stand-alone SerDes.”