Op-Amp sizing with large number of design variables using TuRBO

Bayesian optimization has been demonstrated to be an effective method in the design of analog integrated circuits. However, sizing of operational amplifier (op-amp) circuits with large number of design variables has not been achieved. This paper achieves the efficacy of Bayesian optimization on a CMOS op-amp circuit with 41 design variables, employing both normal Bayesian optimization method and the Trust Region Bayesian Optimization (TuRBO) method. The experimental results indicate that TuRBO can find the solution that achieved specified requirements. And, TuRBO performed better than normal Bayesian Optimization in difficult sizing task. Furthermore, the TuRBO method is shown to require fewer computational resources compared to normal Bayesian optimization. Moreover, TuRBO can be executed in parallel, and parallel TuRBO indicate fewer execution time than single TurbO.