Neuromorphic Circuits with Redox-Based Memristive Devices

Abstract

AbstractThis chapter addresses opportunities for memristive devices in the framework of neuromorphic computing. Memristive devices are two-terminal circuit elements, comprising resistance and memory functionality. This simple and likewise ingeniously concept allows beneficial applications in numerous neuromorphic circuits. However, the electrical characteristics as well as the materials and technological framework of memristive devices need an optimization for each specific application. The chapter starts with a short overview of basic principles of biological data processing followed by a taxonomy of different bio-inspired computing architectures, divided into time-dependent and time-independent concepts. Furthermore, the requirements on particular memristive device properties, such as $$I\text {-}V$$ I - V linearity, switching time, retention, number of states, time-dependency, and device variability, are discussed. The results of tangible examples of digital and analog memristive switching devices used in a deep neural network based on CMOS-integrated resistive random access memory devices (RRAMs) for chronic obstructive pulmonary disease (COPD) detection, in stochastic learning, in bio-inspired analog learning, and, finally, in oscillatory computing are presented and discussed.