Implementation of Convolutional Kernel Function Using 3-D TiOx Resistive Switching Devices for Image Processing

Abstract

We implemented a convolution operation in a convolutional neural network by introducing a 3-D integrated TiOx-based resistive switching device (RSD) array to act as the kernel. The 3-D integrated TiOx-based RSD exhibits a gradual SET/RESET behavior, which enables multilevel characteristics. A pair of layers in the 3-D structured array is used to achieve the positive and negative weights. Based on these results, we demonstrated the 3 x 3 Laplace kernel for edge detection using images from the Columbia Object Image Library data set in a simulator to validate the convolution operation.