kriging元模型构造方法及其在电路优化中的应用研究

发布时间：2018-03-08 10:17

本文选题：试验设计　切入点：元模型　出处：《西安电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：近年来,集成电路遵循摩尔定律朝着集成度不断提高、性能更优、功耗更低、易于携带的方面一步步发展,极大地便利了人们的生活。但是随着集成电路的制造工艺进入纳米时代,庞大地电路结构使得电路设计的难度不断增加,影响输出结果的因素越来越多。尤其对于模拟集成电路系统和混合信号系统来说,电路中各参数之间相互影响,参数和性能指标之间呈非线性变化,为了在输出端或者电路中某一节点获得设计者所需的性能指标值,在确定电路的相关参数时,往往需要花费大量的时间对电路参数进行优化,极大地增加了时间成本。尽管电路设计自动化(EDA)为大规模集成电路设计及优化提供了许多帮助,但是EDA软件之间相互独立,而且集成电路规模扩大使得EDA软件所消耗的参数优化时间越来越长,严重推迟了产品投放市场的时间。本论文首先使用了有效集共轭梯度法的方式对kriging模型进行建模,然后在分析灵敏度的基础上,针对带隙基准电压源设计中的性能指标要求对相应的关键节点进行了单目标和多目标的优化;最后,在分析TSV-衬底结构等效电路的基础上,对TSV的相关参数进行优化。本论文的主要研究成果可概括为:(1)从元模型的观点出发,将建立于物理与电学模型的电路等效成简单的电路关键节点参数与性能指标之间的数值模型,并比较分析了各个试验设计与建模方法的优劣,最终确定了拉丁超立方抽样与kriging模型结合的方法来进行建模可以得到对电路的最佳近似。(2)使用有效集共轭梯度法的方式确定空间相关函数参数θ。kriging模型中如何确定空间相关函数参数θ是建模过程中研究的一个重点。本论文使用了有效集共轭梯度法的方式确定空间相关函数参数θ,在与遗传算法的比较中可以发现,有效集共轭梯度法极大地提高了优化效率。(3)从电路的角度出发,针对带隙基准电压源的设计要求进行优化。在确定关键节点的基础上,利用拉丁超立方抽样方法对范围内的参数进行抽样,根据抽样数据结果建立基于有效集共轭梯度法的kriging模型,在验证模型精度以后,利用该模型对PSRR和TC进行单目标和多目标的优化并验证优化结果。(4)从器件结构的角度出发,针对硅通孔(TSV)-衬底结构的设计要求进行优化。在分析TSV-衬底结构的基础上对该结构进行相应的电路等效并确定电路的关键参数。针对TSV-衬底结构中的噪声耦合问题,利用该等效电路进行基于有效集共轭梯度法的kriging模型参数优化。
[Abstract]:In recent years, the integrated circuits follow Moore's law to improve integration, better performance, lower power consumption, easy to carry one step by step. It greatly facilitates people's lives. But as the manufacturing process of integrated circuits enters the nanoscale era, the huge circuit structure makes the design of circuits more and more difficult. Especially for analog integrated circuit system and mixed signal system, the parameters in the circuit interact with each other, and the parameters and the performance index show a nonlinear change, especially for analog integrated circuit and mixed signal system. In order to obtain the performance index value needed by the designer in the output or a certain node of the circuit, it takes a lot of time to optimize the circuit parameters when determining the relevant parameters of the circuit. Although circuit design automation (EDAA) provides a lot of help for LSI design and optimization, EDA software is independent of each other. Moreover, the expansion of integrated circuit scale makes the parameter optimization time of EDA software longer and longer, which seriously delays the time of product launch. In this paper, the effective set conjugate gradient method is used to model the kriging model. Then, on the basis of sensitivity analysis, the corresponding key nodes are optimized by single target and multi-objective according to the performance index in the design of bandgap voltage reference. Finally, the equivalent circuit of TSV-substrate structure is analyzed. The main research results of this paper can be summarized as: 1) from the point of view of meta-model, The circuit based on physical and electrical models is equivalent to a simple numerical model between the key node parameters and the performance index of the circuit, and the advantages and disadvantages of each experimental design and modeling method are compared and analyzed. Finally, the method of Latin hypercube sampling combined with kriging model is determined to model the circuit. The best approximation is obtained.) how to determine the parameters of spatial correlation function 胃. Kriging model by using the effective set conjugate gradient method. The parameter 胃 of spatial correlation function is one of the key points in the modeling process. In this paper, the parameter 胃 of spatial correlation function is determined by using the effective set conjugate gradient method, which can be found in comparison with genetic algorithm. The efficient set conjugate gradient method greatly improves the optimization efficiency. From the point of view of the circuit, the design requirements of the bandgap voltage reference source are optimized. The Latin hypercube sampling method is used to sample the parameters in the range. According to the sampling data, the kriging model based on the conjugate gradient method of the valid set is established, and the accuracy of the model is verified. The model is used to optimize PSRR and TC with single objective and multi-objective, and verify the optimization result. 4) from the point of view of device structure, Based on the analysis of the TSV-substrate structure, the equivalent circuit and the key parameters of the circuit are determined. The noise coupling problem in the TSV-substrate structure is discussed. The equivalent circuit is used to optimize the parameters of kriging model based on effective set conjugate gradient method.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN402

【相似文献】