Dynamic memristor based high-efficiency reservoir computing

A dynamic memristor-based reservoir computing system with controllable parameters shows the capability of performing high-accuracy temporal signal processing tasks efficiently.

Like Comment
Read the paper

The rapid development of artificial neural networks (ANNs) in the past decade has empowered artificial intelligence in many applications such as image recognition, natural language processing, autonomous driving, and security, etc [1]. While feedforward neural networks, in which neurons are separated into layers and the signal only goes forward, are widely used to recognize static images, neural networks with recurrent connections (e.g., recurrent neural network, or RNN) are usually needed to process temporal signals. However, the training of RNN is usually very difficult and requires extensive computational power, mainly due to the problem of exploding or vanishing gradients in recurrent structures.

Derived from standard RNN, reservoir computing (RC) is designed as a highly efficient network suitable for processing temporal signals with low training cost and simple hardware implementation. RC has been widely used in dynamic system identification, time-series forecasting and many other applications. Recent research has revealed that a recurrent network with complex connections can be equivalently replaced by a nonlinear dynamic system, which enables a complete RC system to be realized using just a few components.

Memristor, as a new type of information processing device, has the advantages of analog resistive switching and in-memory computing. In recent years, significant progress has been made in the implementations of memristor-based ANNs, mostly feedforward neural networks such as multi-layer perceptron (MLP) and convolutional neural network (CNN) [2], [3]. Interestingly, there is a special class of memristors, called dynamic memristor [4], whose inherent dynamic properties and nonlinearity make it have great potential to realize compact and efficient RC systems.

Fig. 1 Single dynamic memristor as complex recurrent network. (a) Device structure and IV hysteresis curves. (b) The experiment exploring the dynamic characteristics of device. (c) The input voltage after the mask process and the corresponding current response of the dynamic memristor, where the time step is much larger than the characteristic time of the device. In this case, the device rapidly saturates to a state that is independent of previous inputs. (d) As a result, the extracted virtual nodes are independent of each other, and each node is only coupled with itself at the previous time step. (e) The input voltage after the mask process and the corresponding current response of the dynamic memristor, where the time step is smaller than the characteristic time of the device. In this case, the device does not have enough time to reach a saturated state. (f) As a result, the extracted virtual nodes can be coupled with their neighbours efficiently, and hence a functional RC system can be implemented.

In this work, a single dynamic memristor, with a device structure of Ti/TiOx/TaOy/Pt, is employed to realize the complete reservoir function. By developing a controllable mask process to generate rich reservoir states, a parallel RC system based on multiple dynamic memristors is further designed. The state richness and feedback strength of the RC system can be well controlled by adjusting the mask length, so the system performance can be optimized. In order to verify the feasibility of the RC system, temporal signal processing tasks including spoken-digit recognition and time-series prediction task of the Hénon map are demonstrated with extremely low word error rate (WER = 0.4%) and prediction error (NRMSE = 0.046) respectively. Compared with previous work [5], the operating power of our memristor-based RC system is much lower owing to the mask process, and the energy consumption can be further reduced by reducing the input voltage pulse width.

Fig. 2 System architecture and performance of the dynamic memristor-based RC. (a) Schematic of a dynamic memristor-based parallel RC system, where the mask sequences are different for every single memristor RC unit. (b) The performance of our RC system in spoken-digit recognition task, where the word error rate is as low as 0.4%. (c) Word error rate as a function of the mask length M, where the total reservoir size (M×N) remains constant at 400. (d) The performance of our RC system in Hénon map prediction task, where the prediction error (NRMSE) is as low as 0.046. (e) The NRMSE changes with the reservoir size in different RC systems including the standard ESN, memristor-based parallel RC and software-simulated one.

It should be noted that the current RC system is based on a single reservoir layer as a proof-of-concept demonstration. Nevertheless, it points out a feasible pathway to construct a high-efficiency memristor-based RC system that can handle complex temporal tasks in real time. In the future, a more sophisticated RC system based on multiple reservoir layers can be constructed, which would have much improved performance because of its richer reservoir states and stronger memory capacity.

These results were recently published in Nature Communications:

"Dynamic Memristor-based Reservoir Computing for High-Efficiency Temporal Signal Processing", Nature Communications XXX, XXX–XXX (2021).

 https://www.nature.com/articles/s41467-020-20692-1

 Reference:

[1]         LeCun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521, 436-444, (2015).

[2]         Prezioso, M. et al. Training and operation of an integrated neuromorphic network based on metal-oxide memristors. Nature 521, 61-64, (2015).

[3]         Yao, P. et al. Fully hardware-implemented memristor convolutional neural network. Nature 577, 641-646, (2020).

[4]         Li, X. et al. Power-efficient neural network with artificial dendrites. Nature Nanotechnology 15, 776-782, (2020).

[5]         Moon, J. et al. Temporal data classification and forecasting using a memristor-based reservoir computing system. Nature Electronics 2, 480-487, (2019).

Jianshi Tang

Assistant Professor, Tsinghua University