Abstract:Memristive devices are commonly benchmarked by the multi-level programmability of their resistance states. Neural networks utilizing memristor crossbar arrays as synaptic layers largely rely on this feature. However, the dynamical properties of memristors, such as the adaptive response times arising from the exponential voltage dependence of the resistive switching speed remain largely unexploited. Here, we propose an information processing scheme which fundamentally relies on the latter. We realize simple dynamical memristor circuits capable of complex temporal information processing tasks. We demonstrate an artificial neural circuit with one nonvolatile and one volatile memristor which can detect a neural spike pattern in a very noisy environment, fire a single voltage pulse upon successful detection and reset itself in an entirely autonomous manner. Furthermore, we implement a circuit with only two nonvolatile memristors which can learn the operation of an external dynamical system and perform the corresponding time-series prediction with high accuracy.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to explore new methods for efficient time - series prediction and neural signal processing using dynamic memristors. Specifically, the paper proposes an information - processing scheme based on two dynamic memristors, one non - volatile and the other volatile. In this way, the researchers hope to achieve the following goals: 1. **Complex time - signal processing**: Design simple dynamic memristor circuits that can perform complex timing - information - processing tasks. For example, detect neural pulse patterns in a noisy environment and autonomously emit a single voltage pulse after successful detection, and then reset itself. 2. **Time - series prediction**: Use a circuit containing only two non - volatile memristors to learn the operation of an external dynamic system and perform time - series prediction with high precision. This requires fewer nodes and training weights than traditional deep neural networks (DNNs), thus improving computational efficiency. ### Key technical points - **Dynamic memristor characteristics**: Utilize the dynamic characteristics of memristors, such as the exponential dependence of resistance - switching speed on bias voltage, to achieve short - term/long - term memory functions. - **Negative bias voltage**: By introducing a negative bias voltage, "artificial forgetting" is simulated, causing the memristor to return to its initial high - resistance state when there is no input signal. - **Feedback mechanism**: By introducing a global feedback loop, automatic resetting of the circuit and output optimization are achieved. ### Experimental verification - **Signal - detection experiment**: The ability of the memristor to detect neural pulses in a noisy environment was verified through experiments, demonstrating its high detection accuracy in a high - noise background. - **Time - series - prediction experiment**: Using a circuit of two non - volatile memristors, high - precision time - series prediction of an external dynamic system was achieved. ### Conclusion This research shows the great potential of dynamic memristors in neural - information processing and time - series prediction, providing new ideas for developing more efficient and more compact neuromorphic - computing systems. By utilizing the dynamic characteristics of memristors, the researchers not only improve computational efficiency but also reduce hardware requirements, laying the foundation for future edge - computing and neuromorphic - computing applications.

Neural information processing and time-series prediction with only two dynamical memristors

Memristors with Tunable Characteristics Based on the Tellurene/nb-Doped MoS2 Heterojunction Toward Bio-Realistic Synaptic Emulation

Dynamical memristive neural networks and associative self-learning architectures using biomimetic devices

Single Neuromorphic Memristor closely Emulates Multiple Synaptic Mechanisms for Energy Efficient Neural Networks

Memristive neural network for on-line learning and tracking with brain-inspired spike timing dependent plasticity

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

A Memristors-Based Dendritic Neuron for High-Efficiency Spatial-Temporal Information Processing.

Spiking Neurons with Neural Dynamics Implemented Using Stochastic Memristors

Memristors‐based Dendritic Neuron for High‐Efficiency Spatial‐Temporal Information Processing

Low‐Voltage Oscillatory Neurons for Memristor‐Based Neuromorphic Systems

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

A Compact Memristor-Based Dynamic Synapse for Spiking Neural Networks

Memristor-based Neuromorphic Implementations for Artificial Neural Networks

An Efficient and Accurate Memristive Memory for Array-based Spiking Neural Networks

A neural network circuit with associative learning and forgetting process based on memristor neuromorphic device

Emulating long-term synaptic dynamics with memristive devices

Anticipative Objective Tracking with Short-Term Memristive Synapses

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

Nonlinearity in Memristors for Neuromorphic Dynamic Systems

Emulating Dynamic Synaptic Plasticity over Broad Timescales with Memristive Device

Fully memristive neural networks for pattern classification with unsupervised learning