What Happens to Differential Pair Signals at 1GHz High Speeds
Same stack up. Same length matching. Same impedance targets. Different routing path and a completely different eye diagram. This video proves why routing decisions on high-speed differential pairs like PCIe Gen 4 and Gen 5 can't be evaluated by rules alone. Using topology extraction in Allegro X with Sigrity X Aurora on a real Nvidia 10-layer HDI board, we compare two PCIe differential pairs routed on opposite sides of the PCB and test them against 1GHz, 2.5GHz, and beyond to see exactly where each path holds up and where it doesn't. At 20Gbps and above you're routing microwave fields through dielectric. Even a single via choice can change the outcome. Simulate before you fab. Try Allegro X free for 30 days: https://ow.ly/C4aW50YerAy Watch more videos on HDI Design: https://ow.ly/K7If50YKAHo Click here for more resources on SI/PI: https://ow.ly/UYE650YVeLO Subscribe to our Channel: youtube.com/@CadencePCBDesignandAnalysis?si=sRz2wbFXC5B0-dOX Follow us on social for the latest OrCAD X news! 👉 Follow Cadence PCB Design and Analysis on X: https://www.twitter.com/CadenceSDA 👉 Follow Cadence PCB Design and Analysis on LinkedIn: https://www.linkedin.com/company/cadencesda 👉 Follow Cadence PCB Design and Analysis on Facebook: https://www.facebook.com/CadenceSDA Disclaimer: This video may have been recorded prior to changes made to the product’s user interface, or the video may be based on an earlier release of the product. The concepts and workflows conveyed in this video still apply to the current release of the product. 00:00 — Why routing decisions matter more than rules at high speed 01:01 — Board overview: Nvidia 10-layer HDI PCB with PCIe on layer 3 02:13 — Setting up topology extraction for two PCIe differential pairs 04:09 — Viewing 3D routing topologies and via transitions 04:58 — How Sigrity X Aurora models dielectric, loss tangent, and copper 06:26 — Running eye diagram simulation at 1GHz 06:57 — Comparing results: D46 vs J6, J33, and U7 08:51 — Rerunning at 2.5GHz & how the eye closes 09:24 — Evaluating routing paths for your target data rate 09:53 — Why the same length and impedance still isn't enough 10:29 — Same board, different routing, different outcomes