Abstract

Introduction The analog design assistance tool Adapt [5, 6] has been developed to help analog electronic circuit designers tuning design parameters, such that the functional design speci cations are met, given process technology constraints. Tuning is based on an optimization process, in which each iteration of the optimization loop implies the evaluation of the circuit by an analog circuit simulator. Considering the simulator as a black box tool, the choice of the optimization technique is restricted, because the simulator does not automatically supply derivatives of the design metrics and numerical noise is inherently present (for instance due to adaptive time stepping). This excludes optimization algorithms that adopt nite-dierence schemes to approximate derivatives. One of the two optimization algorithms available in Adapt is the Nelder{Mead (NM) method [7]. Adapt includes constraints by adding quadratic penalty terms to the cost function when using NM. The Nelder{Mead algorit

Citations