Machine Learning-Based Modeling and Generic Design Optimization Methodology for Radio-Frequency...

Area of Science:

• Electrical Engineering
• Materials Science
• Mechanical Engineering

Background:

• Radio Frequency Micro-Electro-Mechanical Systems (RF-MEMS) technology has advanced significantly, with efforts focused on novel designs and materials for extreme performance.
• However, the design optimization aspect of RF-MEMS devices remains underexplored, limiting their full potential.

Purpose of the Study:

• To develop a computationally efficient, generic design optimization methodology for RF-MEMS passive devices.
• To address the need for an approach applicable to various RF-MEMS components, not limited to a single device type.

Main Methods:

• Coupled finite element analysis (FEA) to model both electrical and mechanical aspects of RF-MEMS design.
• Generation of a comprehensive design space dataset using FEA models.
• Development of machine-learning-based surrogate models to predict device behavior.
• Application of a genetic algorithm-based optimizer to extract optimal device parameters.

Main Results:

• Validated the methodology on RF-MEMS inductors and electrostatic switches, achieving simultaneous multi-objective optimization.
• Successfully extracted optimal trade-offs (Pareto fronts) among conflicting design objectives.
• Demonstrated the generic applicability of the optimization approach across different RF-MEMS passive devices.

Conclusions:

• The proposed methodology offers a computationally efficient and generic solution for optimizing RF-MEMS passive devices.
• This approach enables simultaneous optimization of electrical and mechanical performance, leading to improved device design.
• The study provides valuable insights into design objective conflicts and trade-offs in RF-MEMS.