Machine Learning-Driven Structural Optimization of a Bistable RF MEMS Switch for Enhanced RF Performance.

Area of Science:

• Electrical Engineering
• Materials Science
• Computer Science

Background:

• The increasing demand for miniaturized, high-performance electronics necessitates advanced RF components.
• Radio-frequency microelectromechanical systems (RF MEMS) switches are vital for reconfigurable RF front-end systems, offering low-loss, high-isolation switching.
• Electrothermally actuated bistable lateral RF MEMS switches are favored for their energy efficiency.

Purpose of the Study:

• To enhance the RF performance of lateral RF MEMS switches through structural modification and machine learning (ML)-driven optimization.
• To investigate novel clamp configurations for improved high-frequency operation.
• To reduce the computational cost associated with optimizing RF MEMS switch designs.

Main Methods:

• Re-engineering the H-clamp structure into various lateral configurations, identifying the I-clamp as superior.
• Utilizing an eXtreme Gradient Boost (XGBoost) ML model for predicting optimal design parameters.
• Incorporating activation functions within the ML model to capture complex nonlinear relationships and improve prediction accuracy.

Main Results:

• The I-clamp configuration demonstrated superior RF characteristics compared to other lateral configurations.
• The ML-driven optimization significantly reduced design time by 87.7%.
• The optimized I-clamp switch achieved an insertion loss of -0.8 dB and an isolation of -70 dB at 10 GHz.

Conclusions:

• The proposed structural modifications and ML-driven optimization effectively enhance RF MEMS switch performance.
• The I-clamp design, optimized via XGBoost, offers a computationally efficient approach to achieving superior RF characteristics.
• This methodology provides a pathway for developing next-generation RF MEMS switches for advanced electronic applications.