Perovskite photoelectric memristors with biological synaptic properties for neuromorphic computing

Abstract

The “Von Neumann bottleneck” of traditional computing architecture limits the speed of information processing and the physical size limit indicates the end of “More's Law”. Neuromorphic computing, a new computing architecture, is proposed to deal with the challenges. Memristors are potential in analogues of synapses and in neuromorphic computing. A synaptic device based on Au/CsPbI_3-xBr_x/GaAs memristor is fabricated. Typical synaptic plasticity of the synaptic device is investigated, including long-term potentiation (LTP), long-term depression (LTD) and paired-pulse facilitation (PPF) and the synaptic weight of the synaptic device is modulated by ultraviolet and completed the transition from short-term plasticity to long-term plasticity. Under the joint modulation of optical and electrical signals, the biological classical conditioned reflex of Pavlov's condition was achieved, proving that the device can perform associative learning. Furthermore, two artificial neural networks are constructed for modified National Institute of Standards and Technology (MNIST) data-set recognition to compare the accuracy of a single layer network and convolutional neural network (CNN).