Integrating Device Simulation with Machine Learning for Improved Characteristic of Halo Doped Gate Stack Triple Metal Gate all around (HD-GS-TM-GAA) MOSFET

In this paper, a machine learning method is proposed for analysing variations in device parameters for HD-GS-TM-GAA MOSFET. The Random Forest Regressor (RFR) algorithm is used to envisage the fluctuation in the device parameters of the analysed device. The five input parameters of HD-GS-TM-GAA MOSFET are considered and processed by the ML-RFR model. The proposed ML-RFR method achieves the same accuracy level as device simulation. Python is used for implementing and model training. The root mean squared error (RMSE) has been calculated for the explored device. The findings obtained from the algorithm are quite promising. Moreover, the proposed approach may facilitate rapid device-level assessment for memory-oriented transistor platforms and related circuit exploration where threshold-voltage management and low computational overhead are of prime importance. It mitigates the simulation computational cost by 99% and the error rate is less than 1%. Most of the test samples produce a prediction error within the acceptable range. Proper training and optimization of the device ensure the accurate prediction of device parameters with low computational cost.