Chapter 13 - Emerging memories and their applications in neuromorphic computing

Abstract

In the current era of big data, translating brain-like functionalities into hardware systems is the key to realizing artificial intelligence on a chip in a power-efficient and scalable manner. The neuromorphic computing paradigm offers intelligent systems that can be trained efficiently on unstructured big data through spiking or oscillatory neural networks. While spiking neural networks (SNN) utilize leaky integrator-and-fire neurons and spike-time-dependent synaptic plasticity to perform the training and inference on spatiotemporal data, oscillatory neural network (ONN) models neurons as coupled oscillators to solve combinatorial and computationally hard problems. An efficient implementation of SNN requires novel device concepts to process spatiotemporal data in-memory to overcome the von Neumann bottleneck. In this regard, emerging memory devices, such as ferroelectric field effect transistor (FeFET) and Resistive Random Access Memory (RRAM), come to the rescue. Here, we discuss the unique device properties of FeFET and RRAM and their applications as neurons and synapses to implement power-efficient neuromorphic computing systems. The application of FeFET as coupled oscillators for the implementation of ONN is also discussed.