Issue link: https://iconnect007.uberflip.com/i/1540184
22 DESIGN007 MAGAZINE I OCTOBER 2025 our circuits because getting to the other domain was straightforward and easy with the Fourier transform. Measuring PDN impedance at the milliohm level was readily doable with the two-port shunt-through scheme by probing the PDN at a dedicated power- ground plated through-hole pair with probes con- nected to the top and bottom. This is shown in a slide from my presentation at DesignCon East 2003 (Figure 1, right). ing current values, results in a fast-dropping target impedance. The recent boom of AI and ML has accelerated this trend, leading to another paradigm shift that is rapidly unfolding right before our eyes. The largest chips now require tens of microohms PDN imped- ance to feed them, which creates significant new challenges. Validating microohm AC impedance is very challenging using available instrumentation. Unfortunately, there is a bigger challenge: PDNs cannot be considered lumped anymore (even at low frequencies), because the series plane imped- ance becomes comparable to or even bigger than the required parallel PDN impedance. What are the ramifications of very low target impedance for the PDN? It can result in strong spa- tial variation of impedance and noise, regardless of whether we look at it in the frequency or time domain. In addition, the very low PDN impedance exposes 3D interactions between the probes and the PDN that were previously masked by the higher target impedance 3 . Figure 2 shows the probing pro- cess for a PDN in the footprint of a high-current chip. Note that PCBs using blind vias to connect the sur- face pads to the power planes below may not have F i g u re 2 : I l l u st rat i o n of 3 D i nte ra ct i o n b et we e n p ro b e t i p l o o p s . ▼ " " As the years pass, chips c ontinue to b e c ome more p ower-hungr y, clo ck fre quencies ke ep going up, supply voltages ke ep going down, and the maximum current draw and current transients are on the rise. As the years pass, chips continue to become more power-hungry, clock frequencies keep going up, supply voltages keep going down, and the max- imum current draw and current transients are on the rise. Lower supply voltages require us to set tighter absolute noise budgets, which, combined with ris-