Machine Learning Method for Accurate Analysis of Complicated Low Temperature Random Telegraph Noise

In this work, the Hidden Markov Model (HMM), and the machine learning methods, namely, K-Medoids clustering and Gaussian Mixture Model (GMM) were used for the data analysis of the complicated RTN signals in 18 nm Fully Depleted Silicon on Insulator (FDSOI) n-FET at low temperature. The differences of the three methods in fitting accuracy, efficiency and the defect location were compared. As compared with the conventional HMM model, the GMM model exhibits the highest fitting accuracy while the K-Medoids clustering shows the highest fitting efficiency. In general, K-Medoids clustering is more balanced in terms of extraction speed and accuracy. The relative locations of the traps within the gate oxide were also calculated, with HMM and K-Medoids show similar trap positions. Hence, the machine learning method provides a new solution for the accurate identification and localization of complicated RTN traps in cryogenic transistors.