SiC MOSFET with Integrated SBD Device Performance Prediction Method Based on Neural Network

Area of Science:

• Power Electronics
• Semiconductor Devices
• Machine Learning

Background:

• Silicon Carbide (SiC) MOSFETs with integrated Schottky Barrier Diodes (SBDs) offer excellent performance.
• Traditional TCAD simulations for these devices are complex and time-consuming.
• Neural networks show promise for predicting semiconductor device characteristics.

Purpose of the Study:

• To apply neural network machine learning for predicting static and dynamic characteristics of SiC SBD-MOSFETs.
• To develop a rapid and accurate method for SiC SBD-MOSFET performance prediction.
• To compare the effectiveness of Convolutional Neural Networks (CNNs) against traditional machine learning methods.

Main Methods:

• Modeled and simulated SiC SBD-MOSFET devices using Sentaurus TCAD.
• Generated 625 data sets for device structure and samples.
• Utilized neural networks, specifically CNNs, for performance prediction based on TCAD data.

Main Results:

• Achieved low Mean Square Error (MSE) values for key parameters: Vth (0.0051), BV (0.0031), R_on (0.0065), and E (0.0220).
• Demonstrated high accuracy in predicting static and dynamic characteristics.
• Found CNNs significantly outperformed traditional machine learning methods in accuracy.

Conclusions:

• Neural network-based prediction offers a fast and accurate method for SiC SBD-MOSFET performance.
• This approach accelerates the design process for devices meeting specific performance targets.
• The study highlights the potential of machine learning in advancing power semiconductor research.