DCON26_PAPER_Track14_AcceleratingHigh-SpeedConnectorBreakoutwithPredictiveMachineLearningPhysics-GuidedInsi_46_52.pdf

1、 -Accelerating High-Speed Connector Breakout with Predictive Machine Learning and Physics-Guided Insights Kalyan Vaddagiri,Cisco Systems Sameer Joshi,Cisco Systems -Abstract This paper introduces a physics-aware machine learning framework to predict signal integrity metrics,including Return Loss(RL)

2、and Time-Domain Reflectometry(TDR),for connector via breakouts.Traditional high-frequency structure simulators such as Ansys HFSS,while accurate,require significant runtime and setup,limiting rapid design iteration.The proposed framework replaces repeated full wave runs with a surrogate model traine

3、d on electromagnetic simulation data,preserving the complex dependencies among dielectric constants,via geometry,and stub length.These models are integrated into a graphical interface that allows engineers to input design parameters,visualize predictions with confidence indicators,and perform prescr

4、iptive optimizations.This approach accelerates early-stage development by enabling immediate evaluation of geometric changes and can be extended to automate HFSS-driven workflows for diverse connector families.Author(s)Biography Kalyan Vaddagiri is a Senior Signal Integrity Engineer at Cisco Systems

5、,Bangalore,specializing in the convergence of Artificial Intelligence and Signal Integrity.He earned his masters degree from Delft University of Technology in 2012 and previously contributed to the development of high-speed(224 Gbps)connectors at Molex.His current work focuses on applying data-drive

6、n and physics-informed learning techniques to optimize SERDES and PCB interconnects.Sameer Joshi leads the Signal and Power Integrity(SI/PI)design team at Cisco Systems,Bangalore,driving advancements in high-speed PCB architecture and system-level performance.He brings over twenty years of experienc