Optimizing MMIC performance: The synergy of AI models and heterogeneous integration process

Abstract

This study presents a novel approach for optimizing the parameters of monolithic microwave integrated circuit (MMIC) functional units using machine-learning techniques and multi-objective optimization algorithms. We utilize advanced machine-learning methods, including random forest, artificial neural networks (ANNs), and recurrent neural networks (RNNs), to construct highly accurate models that predict the performance of these units. These models are subsequently integrated with a multi-objective optimization algorithm, specifically the multi-objective particle swarm optimization (MOPSO), to generate inverse design solutions for both the geometric designs of the units and the fabrication parameters of the heterogeneous integration process. Our approach, which has been validated through chip fabrication and testing, has demonstrated its robustness as a tool for achieving optimal MMIC designs. It not only reduces the design time but also enhances the manufacturability of MMICs, thereby opening new avenues in microwave and RF circuit design.