A machine learning approach for optimizing and accurate prediction of performance parameters for stacked nanosheet transistor

Abstract

In this article, the possibilities of accurate prediction of wide range of parameters and optimizing the same through machine learning (ML) approach have been demonstrated for the multi stacked nanosheet transistor (NSFET). The machine is trained by the generated data of the tedious calibrated technology computer aided (TCAD) simulations. An innovative strategy is employed that combines ML with device simulations. Numerous devices are simulated with different geometric parameters like height, width, length and equivalent oxide thickness of the channel. The input, output, and CV characteristics are extrapolated from the simulation which is predicted by ML models. The DC, Analog and RF parameters are derived with a domain expertise approach. The device parameters are anticipated with actual values by the ML approach. Random forest regression, linear regression, polynomial regression and decision tree regression are employed for the prediction of the performance parameters. Random forest regression models provide significant R\textsuperscript{2} score with minimal error percentage. It indicates that TCAD-augmented ML can be considered an alternative to device simulation due to its reduction in computational cost. This work can also be treated as a benchmark for accurate prediction of the NSFET.

Performance optimization of AlGaAs and Al x Ga 1−x As based SM-TM-DG-JL-TFET for an analog/RF applications

“…As the device's dimensions drop, the length of the channel can be proportionally reduced to align with the device's. It may overlook the influence of short channel effects, worries about subthreshold leakage, and phenomena related to quantum tunnelling [3,4]. Most compound semiconductors are composed of a combination of elements from Group III and Group V.…”