当前位置:首页 > 报告详情

机器学习辅助射频集成电路设计实现:EDA 人工智能的新前沿.pdf

上传人: 芦苇 编号:651861 2025-05-01 26页 3.26MB

1、ML-Assisted RFIC Design Enablement:The New Frontier of AI for EDAHyunsu Chae1,Song Hang Chai1,Taiyun Chi2,Sensen Li1,and David Z.Pan11University of Texas at Austin,Austin,Texas,USA2Rice University,Houston,Texas,USA1ASPDAC 2025:Invited paperRFIC(Radio Frequency Integrated Circuit)RFIC components are

2、fundamental in wireless systemsRFIC=active+passive components2Behzad Razavi,RF MicroelectronicsActive ComponentsPassive ComponentsRFIC=Active+PassiveActive-Passive interactions influence the overall RFIC performance3Active ComponentsPassive ComponentsProvides the essential power and gainSignal gener

3、ation,amplification,Define the frequency responseImpedance matching,filtering,phase shifting,stabilization,harmonic manipulation,CThe majority of RFIC design efforts are focused on passive design and optimizationNeed to balance different design objectives across frequenciesMinimize loss,maximize ban

4、dwidth,address impedance transformationFollows active component design and can limit their performanceOccupy most of chip areae.g.,the passive component in an RF power amplifier(PA)occupies over 80%of the core chip areaRFIC Passive Design Challenges4Li+,JSSC24Zhang+,TMTT23Zhang+,JSSC23ActivesComputa

5、tionally expensive simulations IterativeHeuristically-constrainedHigh entry barriers for new designersRFIC Passive Design FlowFast evaluation utilizing ML modelsAutomatedNon-intuitive and beyond humanLowered entry barriers for new designersComputationally expensive simulations IterativeHeuristically

6、-constrainedHigh entry barriers for new designersRFIC Passive Design FlowFast evaluation utilizing ML modelsAutomatedNon-intuitive and beyond humanLowered entry barriers for new designersReplace EM simulation with ML surrogate models for faster performance evaluation ML Surrogate ModelingEM simulati

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文探讨了AI在射频集成电路(RFIC)设计中的应用,特别是机器学习辅助的RFIC设计。RFIC是无线系统的基础组件,包括主动和被动组件。传统上,RFIC设计主要集中在被动设计上,但主动-被动之间的相互作用会影响整体性能。文章指出,使用机器学习模型可以替代耗时的电磁模拟,实现快速性能评估,降低设计门槛。机器学习模型包括基于表格、图形和图像的方法。表格方法通过预定义的模板参数化实现,图形方法将金属迹线分割成图表示,图像方法将RFIC布局转换为像素化的图像。文章还讨论了机器学习模型在表示解决方案空间、数据集质量和数量方面的挑战,并提出了未来的研究方向,包括将物理定律整合到ML模型中,以及开发开放源代码的RFIC数据集。最后,文章提出了逆向设计优化的概念,通过快速合成满足给定目标性能的RF被动设计,以及混合设计空间优化的策略。
"AI如何助力RFIC设计?" "RFIC的主动和被动组件有何作用?" "未来RFIC设计将如何利用AI技术?"
客服
商务合作
小程序
服务号
折叠