Advanced Time Series Data Processing Using Various Memristor‐Integrated Devices

摘要

Abstract Modern edge devices, from wearable health monitors to robot‐mounted sensors, generate continuous streams of time‐series data that require fast and energy‐efficient processing. Memristors, which combine memory and computation in a compact device, offer a promising solution for real‐time hardware‐based temporal data analysis. This perspective reviews how memristors are used in time‐series processors, focusing on two main approaches: conventional reservoir computing (RC) architectures and newly emerging temporal kernels (TKs). While RC systems often rely on volatile memristors and network structures to extract features from time‐varying signals, TKs enhance flexibility by combining memristors with circuit elements like resistors, capacitors, or semiconductor devices. This feature allows precise tuning of signal delays, thresholds, and nonlinear behaviors. Several representative TK designs and variations are introduced, each with strengths in capturing different temporal features. These are well suited for edge applications where quick and selective signal processing is essential. Future efforts must focus on optimizing kernel behavior and integrating TKs directly with sensors to move toward practical use. This direction can enable real‐time tasks such as on‐device health monitoring, biometric authentication, and adaptive sensing, positioning memristive TKs as a core technology for next‐generation intelligent edge systems.