Deep-learning-based precise characterization of microwave transistors using fully-automated regression surrogates.

Area of Science:

• Electrical Engineering
• Materials Science
• Computational Science

Background:

• Accurate transistor models are crucial for microwave device design, particularly low-noise amplifiers.
• Data-driven transistor modeling is complex due to intricate relationships between characteristics, parameters, bias, and frequency.
• Artificial neural networks (ANNs) offer flexibility but face challenges like hyperparameter tuning and overtraining.

Purpose of the Study:

• To develop a novel, automated methodology for transistor surrogate modeling.
• To overcome limitations of standard ANN-based methods in transistor characterization.
• To enhance the accuracy and reliability of transistor models for microwave applications.

Main Methods:

• Proposes the Fully Adaptive Regression Model (FARM) methodology.
• Utilizes Tree Parzen Estimator for automatic determination of network components and processing functions.
• Comprehensively validates the FARM technique on three microwave transistor examples.

Main Results:

• FARM demonstrates competitive edge over state-of-the-art methods.
• Achieves superior modeling accuracy for transistor scattering and noise parameters.
• Effectively addresses issues of data-network matching and potential overtraining in ANNs.

Conclusions:

• FARM provides a robust and automated solution for transistor surrogate modeling.
• The proposed method enhances the design process of microwave devices.
• FARM represents a significant advancement in data-driven device modeling.