Research On Resistive Switching Devices For Nonvolatile Memory And Neuromorphic Applications

Resistive switching device has drawn much attention in these years, and it can be used in many fields, such as nonvolatile memory and neuromorphic computing system. On one hand, traditional floating-gate memory suffers from some drawbacks, such as high operating voltage, low speed and poor endurance, making it difficult to meet the performance requirements and growing market of nonvolatile memory, and thus new memory devices are worth studying. Resistive switching device is one of the most competitive candidates for the next-generation nonvolatile memory because of its nonvolatility, simple structure, high-integration density, fast speed, low-power consumption and low-operating voltage. On the other hand, humans have been dreaming of building a machine that can work like the brain. Neuromorphic system is a brain-like computing system, which is inspired by the brain. In order to construct a large-scale neuromorphic system, the first thing should be addressed is how to effectively realize its basic information storage and processing blocks, i.e., synapse and neuron. Resistive switching devices can realize synapse- and brain-like functions, making it useful in neuromorphic systems. Aiming at researches on resistive switching device for nonvolatile memory and neuromorphic applications, the following works are carried out.1. For nonvolatile memory application, a bipolar resistive switching device based on Hf O2 is fabricated. Eight well distinguishable resistance states are achieved by current programming, and the resistance states obtained with this method have good retention characteristic. The Hf O2-based resistive switching device shows potential for multi-bit memory application which is a simple, effective, and low-cost method to increase storage density. In this thesis, gamma-ray total ionizing dose effects on the Hf O2-based resistive switching device are also carried out. The Hf O2-based resistive switching device shows good endurance, and can be switched more than 10000 cycles even after radiation with a total dose of 20 Mrad(Si). The small changes of set/reset voltages and high/low resistance states induced by gamma-ray radiation have little influence on the proper operation of the device. The Hf O2-based resistive switching device shows superior gamma-ray radiation resistance, and has potential for use in gamma-ray environment.2. For neuromorphic application, resistive switching devices based on Ni O are fabricated. And it is experimentally demonstrated that resistive switching device can realize behavior-level learning/memory/forgetting abilities, as well as synaptic plasticity. Behavior-level learning/memory/forgetting abilities of the brain are mimicked in a Ni O-based resistive switching device; Forgetting curve of brain is mimicked in a Ni O-based resistive switching device whose conductance decays over time. Synaptic long-term potentiation and paired-pulse facilitation are mimicked in Ni O-based resistive switching devices. And it is observed that the magnitude of paired-pulse facilitation in the resistive switching device decreases with the increasing of pulse interval, from which a rapid phase and a slower phase can be extracted by fitting.3. Based on the work of resistive switching device, neuromorphic application of resistive switching device is presented at circuitry level. Based on CMOS and resistive switching device, a synapse circuit that can realize spike timing-dependent plasticity is proposed. Based on the synapse circuit, an associative learning network consisting of 3 neurons and 2 synapses is designed. The network is then constructed with a Ni O resistive switching device and a peripheral circuit, and it can be used to implement associative learning and forgetting of association in the Pavlov’s dog. In addition, a reconfigurable Hopfield network based on Hf O2 resistive switching device is designed and constructed, and the network can be used to implement functions similar to the associative memory of the brain. Single-associative memory and multi-associative memory are demonstrated in the Hopfield network. This work provides a possible way for hardware implementation of neuromorphic systems to emulate memorization.4. As information processing block of neuromorphic system, the scheme to realize neuron concerns the scale of neuromorphic system. Neuron transistor which can realize neuron-like functions including weighted sum function and threshold function is also studied in this thesis. Coplanar neuron thin film transistors based on IGZO channel and Mo S2 channel are respectively fabricated. Function like a two-bead abacus, as well as “OR” and “AND” logic functions are realized in a single neuron transistor.