Novel AI-Empowered Memristive Biosensing Computational Approach for Biomarker Detection

Nowadays, computational models of resistive random access memory (RRAM) structures are emerging as a pre-fabrication stage thanks to their capability for simulating device variables and dynamics. However, existing models face persistent challenges in multiphysics representation, readout reliability, device variability, and noise robustness. In this work a dynamic electrochemical simulation model of a metal-oxide memristive biosensor is presented and demonstrated for highly accurate bio-detection of C-reactive protein (CRP) and prostate-specific antigen (PSA). AI-tools are investigated to improve prediction of PSA biomarker levels. A multiphysics simulation framework was employed to model the coupled chemical and physical behaviour of a metal oxide (MoX) memristive biosensor. Bio-molecular immobilization was analysed by modelling partial differential equations (PDEs), and interfacial chemical reactions. The simulated electrical response to prostate specific antigen (PSA) was validated against previously reported experimental data and classified using a support vector machine (SVM) to reduce noise and improve accuracy. The obtained results show strong agreement with experimental measurements, achieving a mean absolute percentage error (MAPE) of 4.5%. The simulated biosensor exhibits a sensitivity of 7.34 × 107Ω · m3/g and a limit of detection (LOD) of 1.67 pM. Additionally, the proposed AI-based classification model attains an accuracy exceeding 99.5 %. This study demonstrates the suitability of AI-enhanced memristive platforms for highly accurate biosensing and motivating further development of advanced edge-biosensing architectures.