Impact Ionization Coefficient Prediction of a Lateral Power Device Using Deep Neural Network

Area of Science:

• Semiconductor Physics
• Materials Science
• Electrical Engineering

Background:

• Avalanche breakdown theory traditionally uses a constant impact ionization coefficient derived from 1-D structures.
• This constant assumption is insufficient for complex 2D lateral power devices.

Purpose of the Study:

• To propose and extract a new impact ionization coefficient for silicon that accounts for 2D coupling and depletion effects.
• To evaluate the suitability of existing models for 2D devices and introduce a novel prediction method.

Main Methods:

• Extraction of a 2D-dependent impact ionization coefficient for silicon.
• Validation using Technology Computer-Aided Design (TCAD) simulations.
• Development of a deep neural network (DNN) for predicting breakdown location and impact ionization coefficient.

Main Results:

• The impact ionization coefficient in 2D lateral power devices is shown to be a function of 2D coupling effects, not a material constant.
• The conventional Fulop equation is inadequate for analyzing these 2D devices.
• The DNN prediction method achieved 97.67% accuracy for breakdown location and <6% average error for impact ionization coefficient prediction.

Conclusions:

• A new, 2D-dependent impact ionization coefficient model is essential for accurate analysis of silicon lateral power devices.
• Deep neural networks offer an efficient and accurate alternative for modeling complex semiconductor device behavior.
• This work advances the understanding and simulation of avalanche breakdown in modern power electronics.