New statistical analysis methodology to forecast the memory cell behavior before reliability test

In this paper, a machine learning method is proposed implementing the Principal Component Analysis to study the statistical EEPROM endurance degradation. This technique is firstly applied to an UV irradiated memory array. Then, the Density Based Spatial Clustering of Applications with Noise and the Gaussian Mixture Model are presented to extract the minority population of cells. The reliability test study demonstrated the ability of the proposed technique to correlate electrical parameters to forecast the quality and performance of a memory array. Compared to the classical threshold voltage (V th ) analysis, this method is more effective for predicting which population will experience greater degradation. • Endurance test performed on addressable 4 kb EEPROM memory array • Implementation of Principal Component Analysis of statistical data measurement • Electrical properties relationship is compared between DBSCAN and GMM clustering. • Prediction assessment of weak memory bit cells at t 0