Characteristics Of Memristor Used For Vision System

For the biology,more than 70%information are learned from its vision system.Human vision system can recognize various objects and perceive the visual information in complex environment,which promotes the development of bionic vision system through electronic devices and provides a new way for future artificial vision.Although there are machine vision systems based on traditional CMOS devices,these systems couldn’t be accurate,efficient and energy saving to detect the external information liking biological vision system.In recent years,memristors have been emerged in neuromorphological learning,which makes it possible to build a full-memristors-based vision system with biological characteristics.Some researchers have used photoresists to regulate conductance in completing the perception and storage of the external shape by the vision system.However,the performance of memristor,which is used to realize retina,neuron and synapse functions,needs to be further optimized,and the bionic mechanism is not clear enough.This hinders the practical application and development of memristor based vision system.In this paper,we refer to the three important parts of human visual system:retina,neuron and synapse to fabricate the corresponding memristor.By optimizing materials,we can improve the performance of memristor,and explore the physical mechanism of memristor in neuromorphological application.The main contents of this Ph D thesis are shown in below for the three parts of visual system:a)The memristor for the retina:In the biological vision pathway,the retina plays a role in converting external information into neural signals,and then completes the transmission of external information to a higher visual center.In the preparation of memristors with retinal function,the nanoporous monoclinic BiVO₄ thin film was prepared by experimental methods using electrochemical deposition and hydrothermal treatment,which effectively increased the photosensitive area of the device.So that the photomemristor with TiN/BiVO₄/FTO structure are sensitive to the blue violet light with 408 nm wavelength and 11 pW/mm² power.The photocurrent could reach to 5.86mA large current in a constant voltage and reach 154.6 times of large light/dark current ratio in constant voltage mode, which effectively overcomes the problems of low response current and small switch ratio of the device.This device array could sense the“L”optical signal,which presents the potential for completing the function of the retina.And one optical memristor completed the“OR” logic by co-inputting optical signal and electrical signal,which could enlarge the logic integrate density.b)The memristor for neuron:The（120）M1 phase VO₂ was prepared on Nb-doped SrTiO₃ substrates and La_0.3Sr_0.7MnO₃ substrates by pulsed laser deposition system,and then the memristor was fabricated with Au/VO₂/substrates structure.Firstly,it could be found that the device character could be modulated by adjusting the Nb concentrations in the Nb:SrTiO₃ substrate.By comparing 0.05 wt%,0.1 wt%,and 0.5 wt%devices,the phase change couldn’t occur below 10mA current and 0.05 wt%concentration device.With the concentration increasing from 0.10 wt%to 0.5 wt%,the phase change voltage,effective barrier height,effective barrier width decreases from 3.17 V to 1.2V,0.579 eV to 0.387 eV,9.8×10^-8 m to 1.3×10^-9 m,respectively.This means the phase transition of VO₂ can be controlled by the generation of Joule heating which could be adjusted by the Schottky barrier.So,the phase change of the device is not controlled by the direction of the electric field and this lays a theoretical basis for the preparation of VO₂ based memristors.In the in-situ electric field TEM test,the phase transition of VO₂ was directly found by applying different voltages.In the in-situ CAFM test,the phase transition of VO₂ can occur in the entire test area,which indicates that the phase transition-based device can effectively overcome the localized and uncontrollable shortcomings of solid electrolyte-based conductive filaments.Moreover,the I-V device fabricated on the La_0.3Sr_0.7MO₃ bottom electrode showed good symmetry and bilateral threshold memristor,which laid a solid device foundation for the artificial neuron realized by a capacitor paralleling a memristor.c)The memristor for synapse:Since the signals of synapses are analog adjustable in calculation and storage,it is particularly important to prepare a nonvolatile memristor with analog resistance states.In order to increase the state values of the resistance state in the traditional resistance memory,two methods are used in this paper:（1）Quantum conductance states is used to provide multi-value resistance and quasi-continuous resistance state for the device.A memristor with Cu/Zr_0.5Hf_0.5O₂/Pt structure was prepared by magnetron sputtering system.By adjusting the formation of conductance filaments,a stable（1,2,3··,10）integer and（0.5,1.5,2.5,3.5,4.5...）semi-integer quantum conductance states appeared in the device,and these quantum conductance states with 1048steps present Gaussian distribution.The small-size devices with the sizes of 3mm×3mm and 500 nm×700 nm,which fabricated by lithography and electron beam lithography,also presents multi-level quantum conductance states.These quantum conductance states can effectively increase the storage states in one device.This quantum conductance device can simulate the biological synapse spike-timing-dependent plasticity（STDP）and paired-pulse facilitation（PPF）learning function by applied biological-like stimulation.This device also realizes the fast switching speed of 40 ns/60 ns and the low power consumption of 3 pJ/8pJ.More importantly,the linearity of low resistance state outputting of this device is close to 1 in the range of 0.6V,which makes the device realize high precision pattern recognition in the algorithm of linear multiplication.This quantum conductance device provides the basis for building synapses in the visual system.（2）Metal nanoclusters adjust the continuity of the devices conductance.In order to realize the bidirectional continuous adjustment of the resistance state of the memristor and reduce the switching voltage of the device,we use the method of doping metal nanoclusters in the inorganic metal oxide and organic material systems to verify the feasibility of this scheme.（Ⅰ）For inorganic oxide:By the method of co-sputtering of oxides and metal with magnetron sputtering system and gradient doping of metal Ag elements in inorganic oxide TiO₂,these doped metal Ag formed gradient distribution metal nanoclusters in TiO₂ thin films,and the I-V of the device was converted from abrupt change to bi-directional continuous changes in set and reset process.By using the bidirectional continuous conductance device,the potential and depression functions of biological synapses are well simulated.Under stimulation of biological signals,the devices can successfully achieve STDP,PPF,short-term memory to long-term memory（STM-LTM）and other learning and memory functions of biological.In testing STDP,the pulse interval is applied from 6ms to200 ns,which is faster 10⁵ times than biological synapse.After effectively doping Ag nanoclusters,the device also has a fast switching speed of 50 ns/50 ns and a low energy consumption of 26.0 pJ/22.9 pJ.（Ⅱ）For organic materials:By doping metal nanoclusters in the organic silk protein,the set voltage is reduced from 1.0 to 0.4 V and 0.12 V,and the reset voltage is reduced from-3V to-0.2V and-0.08V,respectively.The dispersions for set threshold voltage are reduced from 4.8 V to 0.55 V and 0.9 V,and reset threshold voltage are reduced from 3V to 0.40V and 0.70V,respectively.By doping metal nanoclusters,the dispersion of the device in the ratio of high/low resistance can be effectively reduced,and the continuity of the device resistance is increased at the same time.It also increases the continuity of device resistance change.This continuous conductance device with the Ag-doped also has the ability to mimic the potentiation and depression of synapses,and completes the STDP and PPF learning and memory rules of biological synapses,which provides a way to expand biological organic electronics.Silk protein films could be degraded after immersing in the solution of papain for 40 minutes,which provided the research direction for the preparation of degradable and environment-friendly green electronics.