Machine Learning-Based Modeling of Hot Carrier Injection in 40 nm CMOS Transistors

Abstract

This paper presents a machine-learning-based approach for the degradation modeling of hot carrier injection in metal-oxide-semiconductor field-effect transistors (MOSFETs). Stress measurement data have been employed at various stress conditions of both n- and p-MOSFETs with different channel geometries. Gaussian process regression algorithm is preferred to model the post-stress characteristics of the drain-source current, the threshold voltage, and the drain-source conductance. The model outcomes have been compared with the actual measurements, and the accuracy of the generated models has been demonstrated across the test data by providing the appropriate statistics metrics. Finally, case studies of degradation estimation have been considered involving the usage of machine-learning-based models on transistors with different channel geometries or subjected to distinct stress conditions. The outcomes of this analysis reveal that the established models yield high accuracy in such contexts.

Keywords

• Reliability
• hot carrier injection
• HCI
• machine learning
• ML
• Gaussian process regressions