Abstract
Yield optimization aims at finding microwave filter designs with high yield under fabrication tolerance. The electromagnetic (EM) simulation-based yield optimization methods are computationally expensive because a large number of EM simulations is required. Moreover, the microwave filter design usually requires several performance objectives to be met, which is not considered by the current yield optimization methods for microwave filters. In this paper, an efficient yield-constrained optimization using polynomial chaos surrogates (YCOPCS) is employed for microwave filters considering multiple objectives. In the YCOPCS method, the low-cost and high-accuracy of polynomial chaos is used as a surrogate. An efficient yield-constrained design framework is implemented to obtain the optimal design solution. Two numerical examples demonstrate the performance of the YCOPCS method, including a coupling matrix model of a fourth-order filter with cascaded quadruplet topology and an EM simulation model of a microwave waveguide bandpass filter. The numerical results show that the YCOPCS method can obtain the filter designs with higher yield and reduce EM simulations by 80% compared to Monte Carlo-based yield optimization in all testing examples.
Highlights
Manufacturing variations in the structures of the microwave filters are unavoidable despite using advanced manufacturing techniques [1]–[3]
The computational cost of the Polynomial chaos (PC) method is significantly reduced for yield estimation than that of the Monte Carlo sampling (MCS) method
Compared with yield estimation using the MCS method, yield estimation using the PC model is more efficient because it adopts a low-cost PC surrogate model replacing the expensive
Summary
Manufacturing (process) variations in the structures of the microwave filters are unavoidable despite using advanced manufacturing techniques [1]–[3]. The manufacturing industry strives to increase the yield of microwave filter production to reduce the cost In this case, it is important to obtain the designs with high yield before fabrication. To reduce EM simulation costs, surrogate-based methods, using lowcost mathematical models to replace EM models, are often utilized in the yield estimation of microwave filters [17]–[20]. Machine learning techniques are adopted in the surrogate-based yield estimation, e.g. artificial neural network [18] and Gaussian process regression [19]. These methods still require lots of EM simulation samples for surrogate modeling.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
