基于神经网络的优化算法在EDA中应用研究进展  

Advances of optimization algorithm via neural network computing for EDA

作　　者：赵晨晖 贺珊 刘先明 郭东辉 Zhao Chenhui;He Shan;Liu Xianming;Guo Donghui(School of Electronic Science&Engineering,Xiamen University,Xiamen Fujian 361005,China;IC Design R&D Engineering Center of Fujian Province,Xiamen Fujian 361005,China)

机构地区：[1]厦门大学电子科学与技术学院,福建厦门361005 [2]福建省集成电路设计工程技术研究中心,福建厦门361005

出　　处：《计算机应用研究》2025年第1期1-10,共10页Application Research of Computers

基　　金：国家自然科学基金重点项目(61836010);福建省科技创新平台项目(2023-P-008)。

摘　　要：为了应对芯片设计复杂度的提升,电子设计自动化工具和方法也在不断进步。然而,EDA需要协调达到最佳的功率、性能和面积,通常其不能保证最优的解决方案。EDA工具在电路设计阶段包括逻辑综合、布局布线及验证等均属于多目标、多约束的非线性规划求解过程,且为了更好解决求解中的不确定性和易于出现局域极值等难题,基于神经网络的优化算法已被集成到EDA工具的设计流程中。首先对EDA中的优化问题、多目标优化计算及基于神经网络的优化算法进行了简要概述,继而详细梳理了基于神经网络的优化算法在逻辑综合、布局布线及验证等不同设计阶段中的优化求解方法,并阐述了当前研究所面临的挑战与机遇,希望为集成电路自动化设计及相关领域研究提供参考。In response to the increasing complexity of chip design,EDA tools and methods are also evolving.However,EDA needs to be coordinated to achieve optimal power,performance,and area,and it does not always guarantee an optimal solution.The application of EDA tools in the circuit design stage,including logic synthesis,layout and verification,belongs to the nonlinear programming solution process with multiple objectives and constraints.To better address the uncertainties of the solution and the problems such as the easy to appear local extreme values,optimization algorithms based on neural network had been integrated into the design process of EDA tools.This paper first gave a brief overview of the optimization problem,multi-objective optimization calculation and optimization algorithm based on neural network in EDA,and then sorted out the optimization solution methods of optimization algorithm based on neural network in different design stages such as logic synthesis,layout and verification,and expounded on the challenges and opportunities faced by the current research institute.It hoped to provide reference for automated integrated circuit design and related research.

关 键 词：电子设计自动化 非线性规划 多目标优化 神经网络 优化计算 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]