Single event effects prediction of MOSFET device using deep learning.

Area of Science:

• Electrical Engineering
• Computer Science
• Materials Science

Background:

• Single Event Effect (SEE) is a critical reliability concern for integrated circuits.
• Traditional simulation methods (e.g., TCAD Sentaurus) for SEE analysis are computationally intensive and time-consuming.
• There is a need for faster and more efficient methods to study SEE in semiconductor devices.

Purpose of the Study:

• To propose and evaluate a novel deep learning-based method for researching SEE in 28 nm MOSFET devices.
• To demonstrate the capability of deep learning to predict key SEE parameters rapidly.
• To compare the simulation speed and accuracy of the deep learning method against traditional TCAD software.

Main Methods:

• Development of a deep learning model trained on parameters affecting SEE in 28 nm MOSFETs.
• Inputting relevant parameters to predict drain transient current pulse, peak current, and total collected charge.
• Utilizing metrics such as prediction accuracy and goodness of fit to evaluate the model's performance.

Main Results:

• The deep learning model achieved high prediction accuracy for transient current peak (96.95%) and total collected charge (97.53%).
• The model demonstrated a high mean goodness of fit (0.985) for predicting drain transient current pulse curves.
• Simulation speed was dramatically increased, by 5.89 × 10³ and 1.50 × 10³ times compared to TCAD Sentaurus for different parameters.

Conclusions:

• The proposed deep learning method provides a highly accurate and significantly faster approach for SEE research in MOSFETs.
• This method offers a viable and efficient alternative to conventional simulation techniques, enhancing the study of integrated circuit reliability.
• The findings open new possibilities for rapid and effective analysis of single event effects.