A Machine Learning-based Model for predicting Stochastic BTI Effects

BTI is a major reliability concern in nanoscale digital design, and addressing it during design space exploration in high levels of abstraction is essential to enhance reliability. Aging prediction model appropriate for these levels should have short runtime. In addition, the model must predict the new-observed stochastic effects of aging. A machine learning (ML)-based model for predicting stochastic aging effects is proposed in this paper. First, a large enough training set is obtained by Monte Carlo (MC) simulations, and then, the ML-based model is trained and developed to predict aging statistical characteristics. Various ML algorithms, such as Ridge, Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest, and stacked generalization are evaluated. Results show that ensemble algorithms have high efficiency in aging prediction. When compared to the MC-based approach, the proposed technique shows that the aging prediction runtime is reduced by more than 99%, while accurate prediction of the statistical properties of stochastic aging is obtained with an accuracy of up to 98%. This improvement is achieved by offline data collecting and model training which needs a noticeable runtime. However, it is a one-time offline task and has no impact on prediction runtime.