R
Member
I used to treat EM simulation as something you do when the schedule allows. After two painful respins on 28 GHz front-end boards, I changed my mind — but I also learned it doesn't pay off everywhere. Here's my current rule of thumb.
Where circuit models are fine: simple 50-ohm microstrip runs, well-behaved mitered bends, standard via transitions below ~6 GHz. A good coupled-line/MLIN model in your circuit simulator gets you within a few percent, and fab tolerance will eat the rest anyway.
Where I now insist on full-wave: connector launches (especially end-launch at mmWave — the launch discontinuity dominates the whole measurement), tightly coupled structures like Lange couplers, anything with a slot or defected ground, and transitions between dissimilar media (microstrip to waveguide, for example). On one project, a "looks fine in schematic" end-launch was 1.2 dB worse than modeled at 28 GHz; the EM model caught it in an afternoon, a respin would have cost three weeks.
Practical habits that made EM sim actually useful instead of decorative:
1. Simulate the discontinuity, not the whole board — port it, de-embed, drop the S-parameters back into the circuit sim.
2. Include soldermask and copper roughness if your stackup is above 10 GHz; they matter more than people admit.
3. Always sanity-check one structure against a measured coupon before trusting the rest of the model.
My heuristic: if the discontinuity is smaller than λ/20 at your top frequency, skip it. Bigger than that, simulate it.
Question for the bench: where do you draw the line? Any structure that surprised you — modeled fine but measured badly, or vice versa?
Where circuit models are fine: simple 50-ohm microstrip runs, well-behaved mitered bends, standard via transitions below ~6 GHz. A good coupled-line/MLIN model in your circuit simulator gets you within a few percent, and fab tolerance will eat the rest anyway.
Where I now insist on full-wave: connector launches (especially end-launch at mmWave — the launch discontinuity dominates the whole measurement), tightly coupled structures like Lange couplers, anything with a slot or defected ground, and transitions between dissimilar media (microstrip to waveguide, for example). On one project, a "looks fine in schematic" end-launch was 1.2 dB worse than modeled at 28 GHz; the EM model caught it in an afternoon, a respin would have cost three weeks.
Practical habits that made EM sim actually useful instead of decorative:
1. Simulate the discontinuity, not the whole board — port it, de-embed, drop the S-parameters back into the circuit sim.
2. Include soldermask and copper roughness if your stackup is above 10 GHz; they matter more than people admit.
3. Always sanity-check one structure against a measured coupon before trusting the rest of the model.
My heuristic: if the discontinuity is smaller than λ/20 at your top frequency, skip it. Bigger than that, simulate it.
Question for the bench: where do you draw the line? Any structure that surprised you — modeled fine but measured badly, or vice versa?