Machine Learning Assisted Statistical Variation Analysis of Ferroelectric Transistors: From Experimental Metrology to Predictive Modeling

We proposed a novel machine learning (ML)-assisted methodology to analyze the variability of ferroelectric field-effect transistor (FeFET) with raw data from the metrology. Transmission Kikuchi diffraction (TKD) measurement was performed on grown Si-doped HfO
2
(Si:HfO
2
) thin film. An experimentally acquired polarization map was employed to generate the polarization variation of a ferroelectric gate stack. FeFETs with the multi-domains are simulated in TCAD to generate the training dataset. We trained a neural network using the polarization maps as inputs and the high/low threshold voltage, on-state current, and subthreshold slope as outputs. The trained model with 3,000 data points shows >98% of accuracy and is more than 10
⁶
times faster than performing TCAD to obtain statistics for 10,000 test samples.