GABP算法构建高精度CMOS电压自举采样开关性能预测模型

Performance Prediction Model of CMOS Voltage Bootstrap Sampling Switch Based on Genetic Algorithm Combined with BP Neural Network

模拟集成电路中器件的设计参数与性能指标具有非线性映射关系,同时繁复的设计参数相互制约,使模拟IC满足应用约束下的折中设计极为复杂,电路研发耗时费力。基于自适应学习的神经网络算法能够建立具有非线性映射关系的预测模型,同时具有宽解空间和易获取全局最优解的遗传算法可进一步弥补建模和求解的精度。采用BP神经网络结合遗传算法(GABP)的复合优化框架对CMOS电压自举采样开关的设计参数和性能指标进行精准建模并优化整体电路性能,建模结果与单BP神经网络模型进行了对比,结果表明,GABP复合建模精度高于BP神经网络,拟合相关度从0.73007有效提升到0.94596,模型可靠性有大幅提高,证明了GABP复合优化可有效应用于电路性能的高效预测和辅助优化设计。

Design parameters of devices and performance metrics of analog integrated circuits demonstrate a non-linear mapping relation-ship.As well,complicated design parameters restrict one another,so that the design tradeoff of analog ICs to meet application constraints is extremely complicated,and it is significant time-consuming and laborious in development of IC products.Neural network algorithms(NNA)based on adaptive learning can build a predictive model with nonlinear mapping relationships.At the same time,genetic algorithms(GA)with wide solution space and easy access to global optimal solutions can further compensate for modeling and solution accuracy.A composite optimization framework employing BP neural network combined with genetic algorithm(GABP)is used to accurately model the design parameters and performance metrics of a CMOS voltage bootstrap sampling switch to further optimize the overall performance.The modeling results are compared with that of the single BP neural network model,the results show that the accuracy of the GABP-based model is higher than that of BP neural network,and the fitted correlation(FC)value is effectively improved from 0.73007 to 0.94596,and this remarkable improvement proves that the GABP composite optimization framework can be effectively applied to the efficient prediction of circuit performance and optimized aided-design.