Artificial Neural Network for Response Inference of a Nonvolatile Resistance-Switch Array.

Guhyun Kim,Dohun Kim,Doo Jeong,Inho Kim,Vladimir Kornijcuk,Cheol Hwang

doi:10.3390/mi10040219

Abstract

An artificial neural network was utilized in the behavior inference of a random crossbar array (10 × 9 or 28 × 27 in size) of nonvolatile binary resistance-switches (in a high resistance state (HRS) or low resistance state (LRS)) in response to a randomly applied voltage array. The employed artificial neural network was a multilayer perceptron (MLP) with leaky rectified linear units. This MLP was trained with 500,000 or 1,000,000 examples. For each example, an input vector consisted of the distribution of resistance states (HRS or LRS) over a crossbar array plus an applied voltage array. That is, for a M × N array where voltages are applied to its M rows, the input vector was M × (N + 1) long. The calculated (correct) current array for each random crossbar array was used as data labels for supervised learning. This attempt was successful such that the correlation coefficient between inferred and correct currents reached 0.9995 for the larger crossbar array. This result highlights MLP that leverages its versatility to capture the quantitative linkage between input and output across the highly nonlinear crossbar array.

Highlights

An artificial neural network (ANN) is a layered graph of nodes and edges, offering an immensely versatile hypothesis for various types of data description and different training methods [1]
For the small crossbar array (10 × 9), a network including a single hidden layer (O = 1) loaded with 100 rectified linear unit (ReLU) units could successfully be trained with the 500,000 training examples (Figure 2a,b)
The error histogram for each case is plotted in the inset, indicating a root mean squared error (RMSE) of 0.313 μA and 17.8 μA, respectively

Summary

Introduction

An artificial neural network (ANN) is a layered graph of nodes (activation units) and edges (nonzero connection weights), offering an immensely versatile hypothesis for various types of data description and different training methods [1]. The scope of tasks (other than conventional tasks mentioned above) within the capability of ANN has been markedly expanding, including quantum mechanical problems such as estimation of quantum mechanical ground state given a two-dimensional potential distribution [7] and modelling a mechanical system in presence of noise [8] These examples highlight the neural network as a versatile hypothesis and the capability of backpropagation for supervised learning as a widely applicable training method. Its current response to an applied voltage array naturally captures the multiply-accumulate (MAC) operation so that crossbar arrays have often been used for physical implementation of the matrix–vector product [12,13,14] The benefit of this approach is obvious in comparison to the digital MAC operation: high speed due to the fully parallel operation and energy-efficiency due to no need for data transference during the operation. Our new method may offer a new feasible means of crossbar circuit simulations as an alternative to conventional circuit simulation methods

Description of Model System

Description of Artificial Neural Network

Training and Test Datasets

Training Results

Conclusions