Abstract

This paper proposes a hybrid algorithm combining reinforcement learning (RL) and a genetic algorithm (GA) for PDN decap optimization. The trained RL agent uses a graph convolutional neural network as a policy network and predicts the decap solution for a given PDN impedance and target impedance, which is seeded as an initial population to the GA. The trained RL agent is scalable regarding the number of decap ports. The main goal is to save computation time and find the near global minimum or global minimum. Generalization of the algorithm to different decap libraries is achieved through transfer learning, eventually reducing the training time of the RL agent. The proposed algorithm finds a decap solution satisfying target impedance twice as fast compared with genetic algorithms.

Department(s)

Electrical and Computer Engineering

Comments

National Science Foundation, Grant IIP-1916535

Keywords and Phrases

decap optimization; genetic algorithm; graph convolutional neural network; transfer learning

International Standard Serial Number (ISSN)

2158-1118; 1077-4076

Document Type

Article - Conference proceedings

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2024 Institute of Electrical and Electronics Engineers, All rights reserved.

Publication Date

01 Jan 2024

Share

 
COinS