Pin Assignment Optimization of BGA Packages for EMI Mitigation Using Deep Reinforcement Learning

Pin assignment of ball grid array packages is a challenging engineering problem that must meet strict electromagnetic compatibility (EMC) standards. In this paper, we propose a novel pin assignment method for electromagnetic interference mitigation using deep reinforcement learning (DRL). Firstly, the pin assignment is converted to a Markov decision process that defines state, action, and reward. Then a classical dueling double deep Q network with a physics-inspired feature encoding scheme is trained to quickly optimize the pin assignment. The design case shows that the proposed DRL method only requires less than 100 episodes to find a final solution with excellent EMC performance. As a result, the proposed DRL method can aid in computer aided engineering software for intelligent packaging design.