Optimization of analog circuits via simulation and a Lagrangian-type gradient-based method

We propose a new method for determining the physical sizes of components in an electrical circuit that maximize some primary performance measure while satisfying some conditions on the secondary performance measures. The proposed method is based on the observation that the performance measures are unimodal and smooth. Thus, it focuses on a local search and applies a Lagrangian method to search for a local optimum. The proposed method has advantages over existing methods because it does not rely on approximate formulas for the performance measures, like other equation-based methods do, and finds the “exact” optimal solution by calling an electrical circuit simulator, such as SPICE, at each iteration to evaluate the performance measures and to compute their gradients. The proposed method also enjoys fast convergence because it focuses on a local search rather than global searches. Numerical experiments illustrate the effectiveness of the proposed method in a one-stage operational amplifier.