Abstract:Bio-inspired neuromorphic computing has drawn increasing attention due to its potential for massively parallel, energy efficient and fault tolerant computation, e.g. nonstructured data processing, where memristors are considered as promising building blocks for the construction of neuromorphic networks. However, the lack of clear understanding on memristive switching dynamics (especially for oxide memristors), the undesirable nonlinearity in conductance modulation, and the inherent variations in present devices have hampered the implementation of neuromorphic device and functional networks. Here this chapter presents an approach to directly resolving ion transport dynamics in oxide based memristors via electrostatic force microscopy (EFM). Utilizing this method, unambiguous evidence of oxygen ion migration, accumulation, and conduction channel formation in HfO2 has been clearly observed, providing insights into the microscopic operation principles of oxide based memristors. These understandings are subsequently used to develop novel approaches to engineering the analog switching linearity and the number of weight states in memristive synapses. Moreover, both planar and vertical multi-terminal memristive devices capable of implementing heterosynaptic plasticity are demonstrated, enriching the functionality of memristive components. After incorporation into crossbar networks, the heterosynaptic plasticity endows the devices with facilely tunable learning rate that is highly desirable for achieving optimized learning scheme with accelerated learning and high accuracy at the same time. Finally, a fuzzy restricted Boltzmann machine (FRBM) network is proposed to tolerate device variation that is intrinsic to memristors based on ionic transport mechanism, thus paving the way for highly robust neuromorphic computing based on memristors.

In situ learning using intrinsic memristor variability via Markov chain Monte Carlo sampling

Reliable Memristor-based Neuromorphic Design Using Variation- and Defect-Aware Training

Exploiting the Short-term to Long-term Plasticity Transition in Memristive Nanodevice Learning Architectures

Leveraging Stochastic Memristor Devices in Neuromorphic Hardware Systems.

Memristors with Diffusive Dynamics As Synaptic Emulators for Neuromorphic Computing

Memristor-based Synaptic Networks and Logical Operations Using In-Situ Computing

Self-organization of an inhomogeneous memristive hardware for sequence learning

Tolerance of Intrinsic Device Variation in Fuzzy Restricted Boltzmann Machine Network Based on Memristive Nano-Synapses

Memristive Reservoirs Learn to Learn

Emulating long-term synaptic dynamics with memristive devices

Experimental demonstration of highly reliable dynamic memristor for artificial neuron and neuromorphic computing

High order synaptic learning in neuro-mimicking resistive memories

Building time-surfaces by exploiting the complex volatility of an ECRAM memristor

Neuromorphic Devices and Networks Based on Memristors with Ionic Dynamics

Memristors -- from In-memory computing, Deep Learning Acceleration, Spiking Neural Networks, to the Future of Neuromorphic and Bio-inspired Computing

Probabilistic Metaplasticity for Continual Learning with Memristors

Probabilistic metaplasticity for continual learning with memristors in spiking networks

Spiking Neurons with Neural Dynamics Implemented Using Stochastic Memristors

Memristors with Tunable Volatility for Reconfigurable Neuromorphic Computing

Nonlinearity in Memristors for Neuromorphic Dynamic Systems

Nonlinear Memristor Model with Exact Solution Allows for ex-situ Reservoir Computing Training and in-situ Inference