基于物理设计约束的模拟电路尺寸设计的优化算法GMSGA

Analog Circuit Sizing Optimization Algorithm GSMGA Based on Physical Design Rule

在模拟电路自动化尺寸设计方法中,基于仿真的方法精度高但耗时巨大.考虑到物理设计的约束,基于优化的模拟电路尺寸设计问题可以看作混合整数规划问题,然而利用传统的方法解决这类问题耗时巨大.为了解决该问题,本文提出一种新的优化算法:混合高斯采样贪婪算法(GMSGA).该算法首先基于混合高斯过程采样的方法获得约束侵犯和较小的良好的起始点,随后将混合整数规划问题近似地分解为整数规划问题和连续变量优化问题分别求解,其中:整数规划问题采用了一种贪婪算法,这极大地减少了仿真次数;连续变量优化采用序列二次规划算法,用来对结果进行局部优化.为验证算法的效率和可靠性,我们利用该算法了设计一个ADC中的运算放大器和E类功放.实验结果表明:相较于其他算法,该算法在相同的仿真次数下多次实验可以得到更好的优化结果.

For the optimization-based analog circuit sizing approach,simulation-based method features high precision,but at the cost of long CPU time.Considering physical design,optimization-based analog circuit sizing approach can be formulated as mixed-integer nonlinear programming problem,which is time-consuming with traditional MINLP algorithms.In this paper,a new algorithm called GMSGA(Gaussian Mixture Sampling Greedy Algorithm)is proposed.A Gaussian-mixture based sampling algorithm is first proposed to generate good start points with low sum of weighted constraint violations.Then the MINLP problem is approximately decompased into INLP and NLP.A greedy algorithm is proposed to solve the INLP to greatly reduce the number of simulation,SQP algorithm is used to optimize the continuous variables locally.To verily our algorithm’s efficiency and robustness,an operational amplifier in an ADC and a power amplifier are designed with our algorithm.The results show that compared with other algorithm,our algorithm can optimize analog circuits considering physical design with better results in many experiments using same number of simulation.