We have proposed an ANN based model to predict current and terminal charge of Nanosheet FETs. The model is optimized by controlling output nonlinearity and range by logarithmic transformation with exponent loss function. Input basis is augmented with better correlated features to improve accuracy. Training on the dataset of 13 devices achieves an impressive test accuracy of −98% and −99.5% in drain current and terminal charge for a wide range of biases. The model predicts the characteristics in ~1000x lesser time compared to the industry standard TCAD tool and is accurate for circuit simulations.
We have proposed an ANN based model to predict current and terminal charge of Nanosheet FETs. The model is optimized by controlling output nonlinearity and range by logarithmic transformation with exponent loss function. Input basis is augmented with better correlated features to improve accuracy. Training on the dataset of 13 devices achieves an impressive test accuracy of −98% and −99.5% in drain current and terminal charge for a wide range of biases. The model predicts the characteristics in ~1000x lesser time compared to the industry standard TCAD tool and is accurate for circuit simulations.