EDA-Physical-Design-Router-via-Reinforcement-Learning

Reinforcement learning code to finish the routing works on a 2D grid world, as well as to visualize the result of the routing paths.

File list

• routing_gym.py: RL environment for routing on a 2D grid world.
• PPO_structure.py: RL agent implemented using PPO Algorithm.
• Net.py: Neural Networks of the PPO agent.
• IL_expert_alternate: Generate expert policies using A* algorithm.
• BC_train.py: Pre-training code using Behavior Cloning. Training runs on GPU by default if CUDA is available, can also train on CPU but much slower.
• PPO_train.py: Training code. Training runs on GPU by default if CUDA is available, can also train on CPU but much slower.
• PPO_test.py: Testing code. Results can be visualized on a grid map.

Requirements

• OpenAI Gym 0.26.2
• Pytorch 1.10.0+cu113
• Numpy 1.22.4
• Matplotlib 3.5.3