Symbolic sensitivity analysis in the multi-objective optimization of CMOS operational amplifiers

When performing multi-objective optimization of analog integrated circuits (ICs) by applying evolutionary algorithms (EAs), one challenge is the proper selection of both the design variables and search spaces. This work proposes the use of symbolic sensitivity analysis to identify the most relevant design parameters in an optimization process. Therefore, search spaces are assigned to be larger to the most sensitive parameters and narrower for the parameters with low or zero sensitivity. It leads to a more efficient implementation of the optimization algorithm. The case of study is a CMOS two stages Miller compensated amplifier. We analyze the sensitivity of the magnitude and phase of the transfer function with respect to each design variable. Numerical simulation is done to rank the computed sensitivities and then to reduce the number of design variables and search spaces to perform circuit sizing by EAs. We highlight that sensitivity analysis is quite helpful to improve the sizing by applying the well-known EA: non-dominated sorting genetic algorithm (NSGA-II).