Analytical Separated Neuro-Space Mapping Modeling Method of Power Transistor

Area of Science:

• Electrical Engineering
• Materials Science
• Computational Modeling

Background:

• Accurate modeling of power transistors is crucial for electronic circuit design.
• Existing models face limitations in capturing complex DC and AC characteristics, especially for advanced semiconductor devices.
• Neural network-based approaches offer potential but often struggle with parameter interference and training efficiency.

Purpose of the Study:

• To propose a novel, analytically separated neuro-space mapping (Neuro-SM) model for power transistors.
• To enhance the accuracy of DC and AC characteristic modeling by mitigating internal parameter interference.
• To develop an efficient training methodology for accelerated optimization.

Main Methods:

• Development of an analytically separated neuro-space mapping (Neuro-SM) model.
• Introduction of two distinct mapping networks within the model to isolate DC and AC parameter influences.
• Derivation of novel analytical formulations for integrating mapping networks with a coarse transistor model.
• Implementation of an advanced training approach incorporating sensitivity analysis for faster optimization.

Main Results:

• The proposed Neuro-SM model accurately represents both DC and AC characteristics of power transistors.
• The separated mapping networks effectively prevent interference between internal neural network parameters.
• The advanced training approach significantly accelerates the model optimization process.
• Experimental validation using laterally diffused metal-oxide-semiconductor transistor data confirms the model's accuracy and efficiency.

Conclusions:

• The analytically separated Neuro-SM model offers a significant advancement in power transistor modeling.
• The model overcomes accuracy limitations of existing methods by enabling flexible transformation of terminal signals.
• The proposed method provides a simple yet powerful tool for accurate transistor characterization with efficient training.