Advanced Design for Weakly Coupled Resonators by Automatic Active Optimization

ABSTRACT Advanced design for weakly coupled resonators (WCRs) based on micro disk‐shaped structures has been developed using an automatic active optimization (AAO) process with low coupling factors. Previous design methodologies have often used supervised learning (SL) with genetic algorithm for the optimization process but the limited initial design domain in the SL‐based predictor often results in local instead of global optimizations. This AAO scheme can generate superior designs in the neighborhood of initial high‐density dataset regions and add newly‐found designs to the original dataset to shift the data distribution toward a more generalized model to seek global optimal designs. Results show that after 10 to 15 epochs, the accuracy of the predictors for the first‐order mode and the second‐order mode coupling factor enhances significantly, with the root‐mean‐square errors dropping by 84.9% and 83.8%, respectively. Simultaneously, the optimized designs have resulted in reduced coupling factors of 0.033 and 5.7 × 10 −11 for the first‐ and second‐order modes, respectively, as compared to those in the initial datasets.