Prediction of Single-Event Effects in FDSOI Devices Based on Deep Learning

Area of Science:

• Semiconductor device physics
• Radiation effects in electronics
• Artificial intelligence in engineering

Background:

• Single-event effects (SEE) are critical for assessing radiation resistance in fully depleted silicon on insulator (FDSOI) devices.
• Traditional simulation methods (e.g., TCAD Sentaurus) for SEE analysis are computationally intensive, time-consuming, and complex.

Purpose of the Study:

• To develop a novel deep learning-based prediction method for SEE in FDSOI devices.
• To enable rapid and accurate characterization of SEE parameters under various particle incident conditions.

Main Methods:

• A deep learning model was trained to predict SEE characterization parameters.
• The model input includes particle incident conditions, and outputs include drain transient current pulses, peak current, and collected charge.
• Performance was evaluated against TCAD Sentaurus simulations.

Main Results:

• The deep learning model achieved a goodness of fit of 0.996 for drain transient current pulses.
• Prediction accuracy for peak drain transient current and total collected charge reached 94.00% and 96.95%, respectively.
• Simulation speed was enhanced by factors of 5.10 × 10^2 and 1.38 × 10^3 compared to TCAD Sentaurus.

Conclusions:

• The proposed deep learning method significantly reduces computational cost and improves simulation speed for SEE analysis in FDSOI devices.
• This provides a new, efficient, and feasible approach for studying single-event effects in advanced semiconductor devices.