Study Of Neuromorphic Memristor Characteristics And Optical Modulation

In recent years,with the rapid development of Artificial Intelligence(AI),more stringent requirements have been put forward for computer speed and energy consumption.Neuromorphic computing has become the most potential candidate for future AI development due to its advantages of adaptive learning,high parallel computing,and low power consumption.However,whether a neuromorphic synaptic device with prominent biological behavior can be prepared has become the most important thing in the realization of neuromorphic computing.As the most potential candidate for neurosynaptic devices,memristors have made great progress in neuro-synapses.The development of AI requires the ability to perceive,memorize and process external information under the premise of rapid and low power consumption.For humans,over 70%of our access to external information is through the visual nervous system.Therefore,simulating the perception of light information by the human visual system has become the next important research stage of artificial intelligence.In addition to the advantages of purely electrical memristors,photoelectric memristors also have the advantages of high broadband,low crosstalk,low power and no delay that are difficult to achieve in electrical control.More importantly,they can simulate the most important retinal structure in the visual system.Therefore,adding light modulate methods to the research of memristors can effectively increase ability of the device to perceive information,realize contactless remote control,and greatly expand the application scenarios of memristors.However,the research on either purely electrically controlled memristors or photoelectrically controlled memristors is still at the basic stage,and different methods need to be used to optimize the device structure to achieve the improvement of device performance.This thesis will conduct research from four aspects:preparation process control,exploration of new conductive filaments,flexible high-temperature coexistence,and exploration of visual bionic prototype devices.The main content of this paper is as follows:1.Regulation of preparation process:Regulating the oxygen content of the film to improve the storage performance of neurosynaptic devices.a)As a traditional semiconductor material,SiO₂ is widely used in integrated circuits.The SiO₂ memristor device is fabricated by oxidation growth on p-type high-conductivity silicon using rapid thermal oxide growth equipment.And the electrical properties of the device are also tested.The SiO₂ memristor device exhibits the characteristics of a digital memristor after the electroforming process,that is,the device has two resistance states,high resistance state and low resistance state,and it can complete resistance conversion under+3.5 V and-2.5 V external voltage.The device has a high resistance state of 10⁶?and a low resistance state of 10⁴?,and the resistance ratio between the high resistance state and the low resistance state is 100.The neurosynaptic behavior is tested by applying a 5 V pulse sequence,and finally the device shows two important short-term excitatory and inhibitory neurosynaptic behaviors,EPSC and IPSC.The physical mechanism of trap-assisted tunneling(TAT)is analyzed by temperature-dependent characteristic testing and data analysis.b)On the basis of the research of SiO₂ digital type memristor,the anoxic type SiO_xmemristor is prepared by accurately controlling the rapid annealing condition of p-type high-conductivity silicon,which does not need electroforming operation.The continuous conductance change is shown in 5 consecutive positive 5 V voltage sweeps,proving that the device is an analog memristor.And the phenomenon can remain stable at room temperature,80?and 100?.Furthermore,the conductivity continuous modulation ability of the device is tested by applying different pulse sequences.It proves that the device has EPSC synapse phenomenon.After data analysis,the relaxation time change law of the device response under different pulse sequences is obtained.The relaxation time increases with the increase of the voltage,and increases with the increase of the pulse width,but,decreases with the increase of the interval.The relaxation time represents the forgetting rate,the larger the relaxation time,the slower the forgetting,the smaller the faster the forgetting.Finally,according to the behavior of the simulated memristor based on SiO_x,the physical mechanism model of the combined effect of variable-range hopping(VRH)and tunneling of the device is given.2.Exploration of new conductive filaments:Exploring the theory of carbon-based conductive filaments to improve the performance of synaptic devices.First,a layer of Pd bottom electrode is deposited on the mica substrate,and then 8 layers of graphene are transferred by wet transfer method.Then magnetron sputtering is used to deposit Al N functional layer,and finally two structures memristors with TiN/Al N/Graphene/Pd(GD)and TiN/Al N/Pd(GLD)structures devices are fabricated.By comparing the I-V curve characteristics,switching voltage uniformity and data retention ability of the two devices,it is found that GD device can realize more than 500 cycles of resistive switching times,while GLD appears several RESET failures in 60 times.The addition of two-dimensional material graphene improves the stability of the device by more than 8 times.In addition,through statistics of the switching voltage of the two devices,it is found that the switching voltage of the GD is more concentrated than that of the GLD.In terms of data retention capability,the GD device can achieve a retention capability of 10⁴ second at a high temperature of 200°C,which is significantly improved compared to GLD.Combining the temperature dependence of the device,HRTEM,EELS,and first-principles calculations to conduct a deep analysis of the physical mechanism,it is found that the conductive filaments of the GD device are composed of carbon atoms.The device also has neurosynaptic learning behavior,and tests related conductance regulation behaviors by applying different pulse sequences.This study is the first realization of the preparation of memristors based on carbon conductive filaments and provides strong evidence for the research of carbon conductive filaments.3.Exploring the coexistence of flexible and high temperature:Explore PET alternative substrate to improve the adaptability of neurosynaptic devices under extreme conditions.The layered mica is selected as the substrate,and the thickness is reduced to 10?m by mechanical peeling,which improves the flexibility of the device and also increases the light transmittance of the device,as well as its application potential in the field of flexible transparent electronics.A magnetron sputtering system is used to prepare a memristor device based on Zr_0.5Hf_0.5O₂ on a mica substrate.Through the test under its limit performance,it can be found that:in terms of high temperature stability,the IV characteristic curves of the device were measured at 20,80,140,and 300?respectively,and the results show that all of them can work normally.In terms of flexibility and stability,it was tested that the I-V curves of the device under 0,25,17.5,and 15 mm bending radii can still achieve normal switching characteristics.The resistance state stability of the device under a 15mm radius of 1000 times of bending,and the stability of high and low resistance state transitions at 300°C for 100 times were further tested,proving that the device can maintain stable switching performance under flexible and high temperature extreme conditions.In terms of neurosynaptic behavior,different pulse sequences are applied to test the response characteristics of the device.According to the design,the pulse intervals sent are 1,3,10,50 and 100 ms respectively.When the interval is less than50ms,the device exhibits dual pulse alienation(PPF),and when the interval is greater than50ms,the device exhibits dual pulse suppression(PPD).At the same time,another STDP synapse learning behavior related to pulse interval was also tested.The above test results show that the mica substrate improves the adaptability of neurosynaptic devices under flexible and high temperature extreme conditions.4.Visual bionic prototype device exploration:Exploring photoelectric memristor devices with light perception capabilities.The photoelectric memristor device of TiN/CeO_x/Zn O/Pd/Mica structure is fabricated by magnetron sputtering process.There are defects in the preparation process of CeO_x film,which increases the defect energy level in the film and expands its photosensitive range from ultraviolet to visible light area.The I-V characteristics of the device were measured by applying visible light with wavelength of 405nm and light intensity of 0,1.4,2.8,5.5 W/cm²,respectively.It is found that the structure of device can respond to 405nm light,and with the increase of light intensity,the device shows the characteristics of analog memristor with continuous change of conductance.Under the reading voltage of 0.05V,continuous application of 12 light pulses to stimulate the device to observe changes,as the number of light pulses increases,the measured device current also increases,showing the continuous adjustable characteristic of nerve synaptic conductance.And under different light intensities,the change in conductivity of the device is also different.Finally,a 4×4 photoelectric memristor array analog vision bionic system is developed by using the experimental data.The comparison of the weight change of the device simulated nerve synapses for the perception of light information under the same pulse number with different light intensity stimulation and the same light intensity with different pulse number are given.It is preliminarily proved that TiN/CeO_x/Zn O/Pd/Mica structure photoelectric memristor device can be used as an artificial retina to perceive,memorize and process light information in a visual bionic system.