Application-oriented robustness optimization based on metamodels

Real-life systems should perform within tight limits even when affected by uncontrollable factors such as temperature and power supply variations. This paper outlines a methodology for robustness optimization based on the Design of Experiments and metamodelling concepts. The main idea is to derive metamodels of the mean response and its dispersion depending only on control factors; thus, the optimization can tackle more effectively the tasks of bringing the mean response to within a set range and minimize the response statistical dispersion. The proposed method was used to optimize the sizing of the external circuitry of a low dropout voltage regulator used in automotive applications; several parameters were improved, especially w.r.t. reducing their dispersion with temperature by up to 28.5%. The method makes no assumptions on the uncontrollable factors and can work with a large variety of metamodels employed; also, it can be easily extended to multiple-response optimization.