高彬, 叶佐昌, 王燕. 基于反向传播的电路优化与模型参数提取方法[J]. 微电子学与计算机, 2019, 36(1): 79-84.
引用本文: 高彬, 叶佐昌, 王燕. 基于反向传播的电路优化与模型参数提取方法[J]. 微电子学与计算机, 2019, 36(1): 79-84.
GAO Bin, YE Zuo-chang, WANG Yan. Model Parameter Extraction with Back Propagation Algorithms[J]. Microelectronics & Computer, 2019, 36(1): 79-84.
Citation: GAO Bin, YE Zuo-chang, WANG Yan. Model Parameter Extraction with Back Propagation Algorithms[J]. Microelectronics & Computer, 2019, 36(1): 79-84.

基于反向传播的电路优化与模型参数提取方法

Model Parameter Extraction with Back Propagation Algorithms

  • 摘要: 传统的电路优化和模型参数提取方法假定仿真器为黑盒子, 通过修改参数或者优化变量得到仿真结果, 并用迭代的方式进行优化或模型参数提取.由于仿真器并未提供梯度信息, 因此只能采用不依赖于梯度的优化方法, 限制了优化和参数提取的性能.本文介绍了一种开源的电路仿真工具, 通过在给出仿真结果的同时计算出指定函数对指定参数的梯度, 为基于梯度的优化算法在电路参数提取、电路优化等领域的应用提供了便利.使用该工具实现的梯度下降算法在电感模型和变压器模型参数提取问题上的实验表现显著优于模拟退火等非梯度方法和同一算法不使用其梯度的实现的版本.实验结果证明, 本文提出的方法在电路优化和模型参数提取问题上的性能有很好的改进作用.

     

    Abstract: This work aimed at pushing forward the application on electronic fields of optimization algorithms which hire gradients. The authors developed an open source electronic simulating tool that gives out gradients of designed functions with respect to customized parameters of interest simultaneously when computing the desired state vectors of a circuit net list. These gradients power most gradient based optimization algorithms efficiently and relieve them from fetching the gradients expensively by secant methods. These optimization algorithms perform much better with the help of it, and also beat some non-gradient algorithms namely simulate anneal.

     

/

返回文章
返回