TCAD-Assisted Deep-Learning Surrogate Modeling of a Dielectric-Modulated FinFET Biosensor for HER2-Equivalent Dielectric-State Sensing

Abstract This paper presents a TCAD-assisted deep-learning surrogate framework for a dielectric-modulated FinFET biosensor under a HER2-equivalent effective dielectric-state sensing condition. The first part establishes the device modeling of a three-dimensional silicon FinFET with an analyte-accessible nanocavity, where the nanocavity permittivity modulates the gate-to-channel electrostatic coupling and produces measurable changes in the drain-current transfer characteristics. The second part develops a multilayer-perceptron (MLP) surrogate model using logarithmic current and dielectric representations to reconstruct TCAD dielectric anchor states and predict the held-out HER2-equivalent k = 8 dielectric state. The framework is further extended to cavity-length and cavity-thickness screening , transconductance prediction, I ON /I OFF extraction, sensing-metric analysis, and computational-cost evaluation. For the surrogate-predicted k = 8 case, the extracted sensing parameters indicate an I ON /I OFF ratio of 1.72 × 10 9 , a threshold-voltage sensitivity of 0.26912, and a sensitivity of approximately 6300. Unlike direct clinical detection studies, the present work is restricted to device-level dielectric-to-electrical response prediction. It does not model HER2-antibody binding, concentration-dependent biomolecule loading, biological sample variability, or fabricated-device validation. Overall, the results show that the proposed surrogate can reproduce TCAD transfer characteristics, perform bounded interpolation at the HER2-equivalent dielectric state, and rapidly screen FinFET dielectric-geometry trends for biosensor design exploration.