AI-based Optimization of Power Delivery Networks on Printed Circuit Boards

In this paper a reinforcement learning (RL) based approach to determine optimal design strategy for power delivery networks (PDNs) is presented. This methodology uses the deep deterministic policy gradient (DDPG) algorithm and demonstrates capabilities in optimizing both, the placement of decoupling capacitors (Decaps) and the dimensioning of the power-plane on a printed circuit board (PCB). The results show that by properly configuring the DDPG algorithm, the RL agent can efficiently identify optimal solutions even for complex problem scenarios.