Artificial neuromodulator-synapse mimicked by a three-terminal vertical organic ferroelectric barristor for fast and energy-efficient neuromorphic computing

Abstract

Novel structures for synaptic devices and innovative array configurations are crucial for implementing fast and energy-efficient neuromorphic electronics. We introduce a three-terminal vertical organic ferroelectric barristor equipped with synaptic functions based on Schottky barrier height modulation to implement a neural network with parallel concurrent execution. The barristor can be extended to a diagonal neural network array while sustaining a crossbar array with nondestructive cell programming given the vertical stacking of layered gate line patterning on top. The array enables fast and energy-efficient operation of a diagonal convolutional neural network (CNN) that performs simultaneous weight update of cells sharing a kernel matrix. One-step convolution and pooling can be achieved, omitting sequential convolution for extracting and storing feature maps. The energy for vector-matrix multiplication on the MNIST and Clothes datasets using the diagonal CNN can be reduced by 75.80% and 71.79%, respectively, compared with the use of a conventional CNN structure while reducing the number of image sliding operations to one-fourth and achieving similar recognition accuracy of similar to 91.03%.