AI Driven Power Integrity Compliant Design of High-Speed PCB

Modern high-speed PCB design places strict power integrity requirements on power distribution networks. More recently, artificial intelligence-based approaches have been presented to handle the increasing design complexity. In this paper, we improve upon existing deep learning implementations for real-time impedance prediction using convolutional neural networks and for automatic selection and placement of decoupling capacitors using deep reinforcement learning. Our approach relaxes restrictions on the number of different capacitor types and placement locations in prior surrogate models for fast impedance prediction. Our reinforcement learning implementation considers the impedance requirements of multiple active components simultaneously.