Simplex-anchored regressors for fast global optimization-oriented miniaturization of microwave circuits.

Area of Science:

• Electrical Engineering
• Electromagnetics
• Computational Methods

Background:

• Miniaturization of high-frequency components is critical for modern applications.
• Optimizing microwave circuits for size and performance requires complex, global parameter adjustments.
• Electromagnetic (EM) simulations for parameter tuning are computationally expensive.

Purpose of the Study:

• To introduce an innovative, cost-effective framework for expedited miniaturization of microwave passive components.
• To treat performance metrics as constraints while prioritizing size reduction.
• To develop a method that significantly reduces the computational cost of EM simulations.

Main Methods:

• A three-stage framework: parameter space pre-screening, machine learning (ML)-driven globalized search, and gradient-based fine-tuning.
• Utilizing random sampling for initial parameter space exploration.
• Employing simplex-based regression models within the ML stage for computational efficiency.
• Implementing local optimization for final design refinement.

Main Results:

• The proposed algorithm demonstrates superior performance compared to benchmark methods on two planar devices.
• Achieved miniaturization with performance constraints met through an efficient optimization process.
• The method requires computational costs equivalent to only a few dozen full-wave EM analyses.

Conclusions:

• The developed framework offers a computationally efficient and straightforward approach to microwave passive component miniaturization.
• It provides a significant reduction in running costs and complexity compared to traditional methods.
• The technique successfully balances the need for miniaturization with stringent performance requirements.