Research On Electronic Circuit Aided Design Based On Artificial Intelligence

With the printed circuit board(PCB)in intelligent electronic products becoming more and more miniaturized,and the density of components on the board is also increasing greatly,which resulting in the PCB routing work becoming more and more complex.However,the routing speed and rate of automatic routing function in current electronic design automation(EDA)tools are slow and low,respectively.The current PCB rout-ing work still mainly depends on manual,so that in the process of industrial PCB design still need to consume a lot of human resources.Therefore,intelligent routing algorithm is urgently needed to improve PCB design efficiency and save labor cost.In this paper,an intelligent routing algorithm based on deep reinforcement learning is proposed.The research goal is to realize the automatic routing algorithm with high routing rate,fast routing speed and can adapt to different routing scenarios,so that the intelligent routing algorithm can really replace the routing engineers.The core idea of the algorithm is to use reinforcement learning algorithm based on joint deep learning and Monte Carlo tree search(MCTS).The deep neural network can predict the prior probability of the decision-making action in the current state to improve the search efficiency of MCTS.On the other hand,the search results of MCTS can be applied to train the deep neural network,so that they can accumulate routing experience through continuous iteration and self-learning,realize the efficient and accurate automatic routing of Complex PCB,and significantly improve the efficiency of PCB design.And in the process of training,we can only rely on the reward function designed according to the electrical design rules without any human label information.Compared with the existing automatic routing algorithms,the proposed intelligent routing algorithm based on reinforcement learning has the following innovations:1)A wire order selection algorithm based on joint deep learning and MCTS is pro-posed.Compared with the existing automatic routing algorithm,this paper takes the routing order decision as an important part of the routing process,first applies the wire sort order selection algorithm to determine the routing order,and then uses the automatic routing algorithm for specific routing,so as to prevent the completed routing from excessively affecting the routing of subsequent pairs.Specifically,this paper maps the wire sort selection problem into a sequential decision search prob-lem.The core idea is to obtain the optimal routing order through MCTS search,and apply the action prior probability predicted by deep neural network to improve the search efficiency of MCTS.The experimental results show that the average routing rate of the automatic routing algorithm is improved by 14.76%after using the wire sort selection algorithm.2)An automatic routing algorithm based on joint deep learning and MCTS is proposed.Because the routing experience of routing engineers is difficult to be quantified,the proposed algorithm takes the electrical design rules as the reward function in re-inforcement learning.Through the reward and punishment,the routing algorithm can learn the routing rules automatically without any human routing rules.More-over,the deep neural network in the algorithm can continuously accumulate routing experience through training,just like routing engineers.The more routing work is completed,the predicted action prior probability is more accurate,which can contin-uously improve the search efficiency of MCTS and make the routing performance better and better.3)Compared with the existing algorithms,the experimental results show that the per-formance of proposed algorithm is improved.Firstly,under the same routing se-quence,the proposed algorithm has the highest routing efficiency and the shortest total routing length.Then,after combining the automatic routing algorithm with the wire sort selection algorithm proposed in this paper,the routing rate will also be improved,which proves the importance of wire sort selection in the routing pro-cess.Finally,in the real scene with the size of 200×200 and the number of pairs to be routed is 70,the proposed algorithm can complete all the pairs' routing,and then map the routing results back to the real routing EDA tool,which can meet the requirements of electrical design rules,and the routing efficiency is improved by 400%,compared with human.