基于机器学习决策树模型的集成电路布图设计商用化价值研究  

Research on Commercialization of Integrated Circuit Layout Design Based on Machine Learning Decision Tree Model

作　　者：石英 于扬 李曦 刘庆莲 SHI Ying;YU Yang;LI Xi;LIU Qinglian(Zhongguancun Intellectual Property Promotion Center,Beijing 100009;Beijing Science and Technology Evaluation Center,Beijing 101117;Institute of Automation,Chinese Academy of Sciences,Beijing 100190)

机构地区：[1]中关村知识产权促进中心,北京100009 [2]北京科技审评中心,北京101117 [3]中国科学院自动化研究所,北京100190

出　　处：《中国发明与专利》2025年第3期11-19,共9页China Invention & Patent

摘　　要：[目的/意义]相较于专利等传统知识产权类型而言,目前集成电路布图设计的商用化研究尚有较大差距。通过分析集成电路布图设计的特点与商用化的关系,为实现其商用化提供借鉴。[方法/过程]通过建立机器学习决策树模型实现对一项布图设计能否被商用化的自动划分,进而为最初的芯片设计提供研发方向及可被商业化的预期。本研究在建立总体模型的基础上,进一步建立北京、上海、江苏等地的子模型以便比较不同地区模型的效果。[结果/结论 ]布图设计应用方向、创作人个数等指标成为影响商用化的关键因素;AI类芯片的布图设计相较于电源类布图设计更有可能被商用化;北京、上海、江苏三地区的决策树子模型f1-score在0.8以上,其预测结果贴近真实情况。[Purpose/Significance]Currently,there are still gaps in the commercialization research of integrated circuit layout design,compared to traditional intellectual property objects such as patent.The study analyzes the relationship between the characteristics of integrated circuit layout design and commercialization,so as to provide references for its commercialization.[Method/Process]By establishing a machine learning decision tree model,automatic partitioning of whether a layout design can be commercialized is achieved,thereby providing research directions and commercially viable expectations for the initial chip design.On the basis of establishing the overall model,the sub models are also provided for Beijing,Shanghai,Jiangsu and other regions to compare the effectiveness of the model in different areas.[Result/Conclusion]The study has found that the research and development direction of layout design,as well as the number of creators,are key factors affecting commercialization.The layout design of AI chips is more likely to be commercialized compared to power supply layout design.The f1-scores of decision tree sub models for Beijing,Shanghai,and Jiangsu are above 0.8,and their predicted results are close to the actual results.

关 键 词：机器学习 决策树模型 商用化 集成电路布图设计 

分 类 号：TN40[电子电信—微电子学与固体电子学]