High-Performance and Intelligent Digital Active Sensor System for Epilepsy Information Acquisition and Classification

Abstract

In this article, we present a performance-enhanced digital active sensor system consisting of a quantum-enhanced active sensor (that frontend micro-electro-mechanical-systems (MEMS) electrode coupled with readout circuits) and an artificial intelligence (AI) empowered engine, which is used for detecting dual-mode neural signals sensitively and classifying about epilepsy disease accurately, respectively. First, digital active sensors with improved sensitivity and reliability include: 1) the active sensor’s bare microelectrode array (MEA) manufactured from the perspective of quantum mechanics; 2) the high-resolution active sensor’s information quantizer for weak neural electrophysiological signal quantization with noise depression techniques; and 3) transmitting power of the radio transmitter reduces by employing bitstream compression with fast rolling-in hash table method. Second, structure and subsystem re-synthesizable AI engine with flexible and scalability abilities. For bare microelectrodes, on the one side, an optimized stack Si₃N₄/SiO₂ dielectric thin film of the bare microelectrode can prevent leakage caused by Fowler-Nordheim (FN) tunneling and stress-induced leakage-current (SILC) effects, also solve the microelectrode peeling problem caused by mechanical stress. In addition, nanomaterial Pt with nanostructure modified on the surface of Pt/Ti alloy of the bare microelectrode for reconstructing local Fermi level, increasing charge injection, and boosting electrocatalytic with continuous-descent-energy-step (CDES) approach. For the neurotransmitter microelectrode manufactured based on the bare microelectrode, the tested linear range is 0.998° and sensitivity is 0.0072 pA/ $\mu $ m². Energy consumption of the entire digital active sensor is less than 70 mW under a 1.8-V power and electrophysiological spike resolution is $0.93~\mu $ V. The recognition accuracy of the resynthesizable AI engine arrives at 99.74% using a random forest model, which is appropriate for epilepsy disease classification.