Prediction of Generated Single Event Transient Pulse Width Using Artificial Intelligence Methods

In this work, the use of artificial intelligence (AI) algorithms for the prediction of generated Single Event Transient (SET) pulse width in standard combinational cells is analyzed. The SET generation is characterized using the current injection approach in Spectre simulations. Based on conducted simulations, the dataset of SET pulse width in terms of gate driving strength, supply voltage, temperature and particle Linear Energy Transfer (LET) is created. The dataset is then used to train six standard machine learning models: linear regression, polynomial regression, support vector machine, decision tree, random forest, and multi-layer perceptron. For all models, the largest prediction error was observed for small SET pulse widths (< 100 ps). The multilayer perceptron model has shown the best accuracy in SET pulse width prediction, with the correlation coefficient R
²
> 0.99.