Artificial Neural Network-Assisted Optimization of DM JLNC-FinFET Biosensor With Raised Source/Drain Architecture

This study presents an artificial neural network (ANN)-assisted optimization framework for the raised source–drain (RSD) dielectric-modulated (DM) junctionless (JL) negative capacitance (NC) FinFET biosensor, designed to detect both charged and neutral biomolecules. The biosensor features a “T-shaped” metal support, providing adsorption regions on both sides for enhanced biomolecule detection. To evaluate the ANN model’s performance, root mean square error (RMSE) and ${R}^{{2}}$ -score are analyzed. The network architecture comprises an input, followed by two hidden layers with 32 and 16 neurons, and an output layer with a single neuron. The model achieved a maximum ${R}^{{2}}$ -score of 0.9872 and a minimum RMSE of 0.0274 when trained on 90% of the dataset. Its reliability is further ensured through K-fold cross validation. The training and prediction processes utilize a TCAD-integrated ANN framework, where the dataset is generated by systematically varying key device parameters, such as channel length ( ${L}_{g}$ ), channel doping concentration ( ${N}_{\text {CH}}$ ), fin thickness ( ${T}_{\text {Fin}}$ ), and support metal work function (SMWF). A total of 1140 device simulations were performed, demonstrating the ANN capability to accurately predict the biosensor’s sensitivity to pyridine biomolecules with high precision.