Efficient Yield Estimation of RF-MEMS Using Vector Fitting Based Neuro-Transfer Models

This paper reports a novel approach for rapid and precise yield estimation of RF-MEMS using Vector Fitting (VF)-based neuro-transfer function (neuro-TF) models. This model offers rapid and accurate prediction of electromagnetic (EM) behavior of RF-MEMS. The advantages of this proposed method are exemplified through the analysis of a film bulk acoustic resonator (FBAR) as an illustrative example. It can predict EM behaviors of FBARs 2.4 × 10
³
times faster than Finite Element Analysis (FEA) simulation, with high accuracies of 98%. Consequently, it can potentially reduce the total computation time required for yield estimation by up to ~58.3% compared to the Monte Carlo (MC) method. Moreover, this method allows for the quantitative analysis of the most critical steps in the FBAR fabrication process, facilitating accelerated design and yield iteration cycles for the mass production of FBARs.