Issue link: https://iconnect007.uberflip.com/i/1514189
42 DESIGN007 MAGAZINE I JANUARY 2024 In signal and power integrity (SI and PI), we would ultimately like to see a reasonable agreement between the predicted or simulated and the measured performance of our circuits. Real-world measurements will always con- tain errors and usually show a distorted rep- lica of the true behavior of the device under test (DUT). Real measurements always show more than just the behavior of the DUT. Even if we don't consider random noise and random errors, in the measured data we have contribu- tions from instrument, cable, and probe errors (just to name a few) that our calibration could not completely remove. Simulations work differently: ey always show less—an incomplete picture of the DUT—simply because there is no way for us to include every detail in our models practi- cally. Not to mention the details that we may not even know about. In simulations, our first question is how to select and set up the mod- els and how to adjust the simulation tool settings so that we get the correct and expected results. But sim- ulations will include only those details that we set up and specify in our simula- tion model. is is true for relatively well-known parameters, such as the dielectric con- stant and loss tangent of laminates, and for other The Effect on SI and PI Board Performance parameters that we tend to describe with a range of fitted models, such as surface rough- ness. Dimensional tolerances fall into this cat- egory. ey can be considered in simulations, for instance, by doing statistical margining of the dimensions. However, in high-speed inter- connects there are so many geometry details that determine the channel performance, that simulating a meaningful number of combina- tions becomes daunting. One interesting case is related to backdrilling the unwanted via stub on high-speed channels. If the backdrilled hole is not concentric to the via barrel that we want to remove, there is a chance that a sliver of the barrel remains (Figure 1). Figure 1a is a properly backdrilled via. In Fig- ure 1b, there is a visible misregistration of the backdrilled hole, leaving part of the original via barrel in place. e residual via barrel behaves similarly to a full via barrel; it creates a dip in the signal transmission. e frequency of the Quiet Power Feature Column by Istvan Novak, SAMTEC Figure 1: Photos of "good" and "bad" backdrilling. a b