Deep Learning-Based Fast BSIM-CMG Parameter Extraction for General Input Dataset

A deep learning (DL) technique to extract the set of Berkeley short-channel IGFET model-common multigate (BSIM-CMG) compact model parameters directly from experimental capacitance–voltage ( ${C}_{\textit {gg}}$ – ${V}_{g}{)}$ and current–voltage ( ${I}_{d}$ ,– ${V}_{g}{)}$ measurements is presented in this article. The proposed technique uses a cascade of inverse and forward artificial neural networks (ANNs) to accurately compute an inverse of the compact model while avoiding the problem of non-uniqueness. It also accurately adjusts the BSIM-CMG compact model parameter values for any variation in the geometry, highlighting that the proposed technique successfully captures the uncertainties in device dimensions. The proposed model exhibits good generalizability and scalability with respect to the size of the sampled ${C}_{\textit {gg}}$ – ${V}_{g}$ and ${I}_{d}$ – ${V}_{g}$ datasets as well as the number of compact model parameters to be extracted.