Proof of concept: comparative study of machine learning models for optimization and performance evaluation of DM RSD JLNC-FinFET biosensor

Abstract This study presents a comparative machine learning (ML)-assisted optimization framework for a raised source-drain (RSD) dielectric-modulated (DM) junctionless (JL) Negative Capacitance (NC) FinFET biosensor, designed for detecting both charged and neutral biomolecules. The biosensor incorporates a ‘T-shaped’ metal support, providing adsorption regions on both sides to enhance biomolecule detection.Two ML algorithms, Random Forest Regression (RFR) and Artificial Neural Network (ANN), are employed to predict the sensor’s sensitivity with high accuracy. The performance of both models is evaluated using Root Mean Square Error (RMSE) and R 2 -score, with reliability ensured through K-fold cross-validation. The ANN model, featuring two hidden layers with 32 and 16 neurons, achieves a maximum R 2 -score of 0.9888 and a minimum RMSE of 0.031. The RFR model also demonstrates strong predictive capability, achieving an R 2 -score of 0.9815 and an RMSE of 0.0359. The training dataset is generated through 1140 device simulations in a TCAD-integrated ML framework by systematically varying key device parameters, including channel length (L g ), channel doping concentration (N CH ), fin thickness (T Fin ), and support metal work function (SMWF). Furthermore, the limit of detection (LoD) of the proposed biosensor is calculated using the International Union of Pure and Applied Chemistry (IUPAC) definition in conjunction with an ML approach. This comparative analysis provides valuable insights into the strengths and limitations of both models, guiding the selection of an optimal ML approach for biosensor performance prediction.