Research On Segmentation Algorithm For Large Scale Integrated Circuit Netlist

Intelligent automatic design of integrated circuits has always been a research hotspot in the field of electronic design.In recent years,with the implementation of the ”Electronics Resurgence Initiative” in the United States,it has further promoted the development of intelligent circuit design in the United States.However,the development of relevant industries in our country is still immature,and the research,development and introduction of technology are controlled by others.Among them,circuit intelligent design software(EDA)is one of the important tools in the electronic design industry,However,this technology has been monopolized by foreign giants at a high price,so the research and development of our country's circuit intelligent design software has important practical significance.In the circuit automatic layout and routing problem of circuit intelligent design,the size of circuit netlist determines the scale and efficiency of problem solving,and even affects the quality of circuit physical design.Therefore,large netlist often needs to be divided into some small sub modules for processing,that is netlist segmentation.Thesis focuses on circuit netlist segmentation in EDA.The main research contents are as follows:(1)Due to the lack of PCB netlist segmentation data set,thesis uses the methods of digital image processing,OCR character recognition and manual verification to make the netlist segmentation data set.The data set contains 11 netlists,and the number of segmentation sets is about 15 ~150.At the same time,thesis combined with the problem of netlist segmentation,the relevant indicators of target detection are adopted,which is used for the performance evaluation of subsequent netlist segmentation algorithms.(2)Starting from the geometric characteristics of circuit netlist connection,thesis uses Maxflow(Max flow and min cut)model to cut and model the netlist,puts forward the algorithms of contention mode netlist segmentation and super terminal mode netlist segmentation based on Maxflow model,and completes the weight design combined with the size and distance of core devices.The correct usage rate of this method achieved 11% ~54%.(3)After summarizing the shortcomings of netlist segmentation algorithm based on geometric characteristics,according to the electrical nature that the larger the number of pins and the closer the distance between the core devices in the circuit,the more peripheral circuit components are likely to have,thesis proposes a netlist partition algorithm based on gravity model.After summarizing the shortcomings of gravity algorithm,thesis further proposes a netlist partition algorithm based on network attributes,and reassigns the lost components that are easy to occur in the netlist partition algorithm based on network attributes when the circuit characteristics are summarized.On the segmented data set,the accuracy of gravity model partition algorithm is 42 % ~65 %,and the accuracy of network attribute partition algorithm can reach 81 % ~96 %.The main contributions of thesis are as follows:(1)starting from the geometric characteristics of netlist,two netlist segmentation algorithms based on maxflow model are proposed,which are netlist segmentation algorithms based on super terminal model and contention model.(2)On the basis of summarizing the circuit characteristics,the gravity division algorithm of netlist and the network attribute division algorithm of netlist are proposed.The network attribute partition algorithm can achieve high accuracy.This research can provide division result reference for engineers who have requirements related to circuit netlist division,so as to reduce their workload,and also lay a foundation for the automatic layout and routing of subsequent EDA software.