Related Experiment Video
Updated: May 5, 2026
07:34
A Simple Stimulatory Device for Evoking Point-like Tactile Stimuli: A Searchlight for LFP to Spike Transitions
Published on: March 25, 2014
9.3K
A Universal Deep Learning Model for Predicting Detection Performance and Single-Event Effects of SPAD Devices
Yilei Chen1, Jin Huang1, Yuxiang Zeng1
1School of Microelectronics, Xidian University, Xi'an 710071, China.
Micromachines
|May 4, 2026
Summary
A new deep learning (DL) model rapidly predicts single-event effects (SEEs) in single-photon avalanche diode (SPAD) devices. This approach offers high accuracy and efficiency, overcoming limitations of traditional simulation methods for radiation reliability.
Area of Science:
- Electrical Engineering
- Materials Science
- Computer Science
Background:
- Single-event effects (SEEs) pose a significant threat to the radiation reliability of integrated circuits, particularly single-photon avalanche diode (SPAD) devices.
- Traditional analysis methods, such as Sentaurus Technology Computer-Aided Design (TCAD) simulations, are computationally intensive and time-consuming, hindering rapid device characterization.
Purpose of the Study:
- To develop a generalized deep learning (DL) model for the rapid and accurate prediction of critical parameters affected by SEEs in SPAD devices.
- To evaluate the model's prediction precision, robustness, and computational efficiency compared to conventional simulation techniques.
Main Methods:
- A deep learning (DL) model was developed using a silicon-based SPAD device with a double-junction double-buried-layer (DJDB) structure (180 nm CMOS process) as the subject.
- Key parameters influencing SEEs were integrated as model inputs for predicting transient current peaks and dark count rates.
- The model's generalization capability was validated by predicting the detection performance of the DJDB-SPAD device.
Main Results:
- The DL model achieved high prediction accuracies: 97.32% for transient current peaks and 99.87% for dark count rates.
- Prediction accuracies for four key performance parameters of the DJDB-SPAD device exceeded 97.5%, confirming model robustness.
- The proposed DL method demonstrated a 336-fold improvement in computational efficiency over conventional TCAD simulations.
Conclusions:
- The developed DL model provides an efficient and accurate solution for rapid characteristic analysis and reliability evaluation of SPAD devices under SEEs.
- This approach overcomes the computational burden of traditional methods, enabling faster design cycles and improved radiation hardness assurance.
- The study highlights the potential of deep learning in addressing complex challenges in semiconductor device reliability and performance analysis.
