An Enhanced Deep Reinforcement Learning-Based Global Router for VLSI Design

Algorithm 1

DDQN-based global router.

1: Decompose multi-pin nets with Prim algorithm

2: Encode two-pin nets

3: Initialize -network with random weights

4: Initialize experience replay buffer with router

5: Network training:

6: for episode : episodes do

7:  for two-pin net : two-pin nets do //Concurrent

8:   Get initial state code for two-pin net

9:   for : max step do

10:    Eliminate redundant actions

11:    With policy, select an action

12:    Take action in environment and get reward

   and state

13:    Update routing information

14:    Store experience

15:    Randomly sample training samples

16:    Set

17:    Perform a gradient descent step on MSE(,

  )

18:   end for

19:  end for

20: end for