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

Table 1

Wire length comparison between DQN-based and DDQN-based global routers.

No.	Grid size	Net number	Max step	Capacity	Wire length		Optimization rate
					DQN	DDQN
1		20	2	3	167	165	1.20%
2		20	2	3	170	168	1.18%
3		20	5	4	264	257	2.65%
4		20	5	4	250	249	0.40%
5		40	2	4	Fail	277	—
6		40	2	4	Fail	293	—
7		40	2	3	480	467	2.71%
8		40	2	3	Fail	521	—
9		40	5	5	949	937	1.26%
10		40	5	5	806	801	0.62%
Average							1.43%