Simulations should match measurements. Otherwise, what good are they? When doing signal integrity simulations, that starts with comprehensive stackup modeling. I quite often hear that stackup editor calculations don’t match measurements, but when I probe deeper, I usually find that the stackup has not been modeled properly. More often than not, the calculated impedance ends up being much lower than the measured impedance. Why is that? The usual culprit is dielectric constant, Er. Most folks will put in their dielectric heights and a value of 4.0 for the dielectric constant for a FR4 board. However, dielectric constant varies with dielectric height, usually within a range of 3.5 to 4.5. The thinner dielectrics typically have a dielectric constant closer to the 3.5 value, and since most stackups have a lot of 4-mil and 5-mil dielectric layers, that means that 4.0 is too high, and that the simulated stackup is showing too low of an impedance.
Dielectric constant is also very important in determining timing numbers. With many of today’s modern busses being source-synchronous, skew is more important than actual flight time so this is becoming less of a concern, but there are still plenty of common-clock interfaces that require accurate prediction of signal flight time.
So, talk to your manufacturer and make sure you are including the right dielectric constant in your stackup calculations. It is a crucial first step to making sure your simulations are accurate.
You can read more about accurate stackup modeling in my recent article in InCompliance magazine: