Optically Modulated Dual-mode Memristors Based On Core-shell CsPbBr₃@GDY Quantum Dots And Its Application In Fully Memristive Neuromorphic Hardware

Neuromorphic computing inspired by the human brain has shown unparalleled advantages in the field of data-centric artificial intelligence（AI）,and is considered as an effective means to break the von Neumann energy efficiency and arithmetic bottleneck in the post-Moore era.However,current neuromorphic AI chips are still based on traditional Complementary Metal Oxide Semiconductor technology,which require several devices to simulate functions of synapses and neurons.Therefore,there is an urgent need to develop novel conceptual devices with human-like synaptic and neuronal functions from the bottom up,as well as to develop genuine neuromorphic chips.With the advantages of simple structure,high integration,fast computing speed and low power consumption,memristors are ideal for constructing hardware artificial neural networks.Memristors have two modes,non-volatile resistive switching and volatile threshold switching,which can simulate the functions of synapses and neurons,respectively,creating the conditions for the construction of fully memristive neuromorphic hardware systems.However,the vast majority of memristors have only one mode of operation.Existing fully memristive artificial neural networks require two separate types of memristors to act as synapses and neurons,leading to a more complicated processing and integration system for the chip.The development of a dual-mode memristor,enabling flexible switching of the memristor between the two modes,is essential for the realization of reconfigurable fully memristive neuromorphic hardware systems.Compared to setting compliance current to control the operation mode of memristors,optical methods offer greater flexibility,energy efficiency,and reliability in modulation methods.In this thesis,a flexible optical-controlled dual-mode memristor array with Ag/polymethyl methacrylate/Cs Pb Br₃@Graphdiyne/polymethyl methacrylate/Ag structure has been proposed,which based on core-shell Cs Pb Br₃@Graphdiyne quantum dots.Under dark condition,this device shows typical non-volatile resistive switching behavior that can simulate synaptic functions.Under light conditions,it rapidly switches to a volatile threshold switching mode,which is useful for constructing neurons.This mode transition originates from the separation of photogenerated carriers at the Cs Pb Br₃/Graphdiyne interface,where photogenerated holes are transferred to the graphdiyne surface,oxidizing the nearby Ag to Ag⁺for breaking the conducting filament.The 64 memristors prepared in this paper have a 100%yield and reversibility in the transition between the two modes.In addition,compared to other metal halide perovskites based memristors,the Cs Pb Br₃ quantum dots encapsulated by graphdiyne used in this paper are able to greatly enhance the environmental stability of the devices.Further,this thesis proposes two fully memristive neuromorphic hardware systems based on this light-controlled dual-mode memristor.One is an injury information sensing and processing system that senses external injurious information,using the memristor as an injury receptor and spiking neuron.The other is a fully memristive spiking neural network consisting of 30×10 resistive-switching mode memristors and10 threshold-switching mode memristors.This system can perform hardware-level classification of digital images,and the trained neural network demonstrates high accuracy in exciting corresponding neurons when numbers 0-9 are input.This is the first fully memristive artificial neural network constructed from homogeneous memristors.