Performance And Physical Mechanism Of Quantum Dot Doped Ga₂O₃ Memristor

With the advent of the era of big data and the growing demand for portable mobile electronic devices,nonvolatile memory has been widely studied.The traditional nonvolatile memory represented by flash memory is faced with many technical challenges and some physical limitations,such as high threshold voltage,poor retention,low speed,proportional limitation of tunneling oxides and so on.In order to overcome these problems,many new nonvolatile memories have emerged.Because of its simple structure,fast speed,strong retention and low power,memristor has attracted people's attention.However,due to the random formation of the conductive filaments,the switching voltage of the memristor presents a very wide distribution.It needs a lot of artificial neural calculation training time to simulate the devices with poor repeatability of the resistance state,which further increases the probability of neural network programming errors,so it is particularly important to solve the problem of the memristor switching voltage dispersion.In addition,the application of neural synapse Bionics in memristor and the problem of overcoming the sneak current of memristor in integrated circuit are also hot spots.we fabricated three types of MDs,namely,i)Ag/Ga₂O₃/Pt[i.e.,device I],ii)Ag/Ga₂O₃/isolated QDs/Pt,[i.e.,device II]and iii)Ag/Ga₂O₃/networked QDs/Pt.[i.e.,device III]Among these three types of devices,the MDs of type?iii?show the best electrical performance.For the device III,we achieved a reduced threshold voltage,centralized distribution of the SET and RESET voltages,robust retention,fast response time,and low switching power consumption.We also achieved comprehensive biosynaptic functions and plasticity.At the same time,this paper makes a physical mechanism analysis of this performance enhancement.Due to the ordered arrangement of PbS QDs in the device,the local electric field around the quantum dot is enhanced,which can effectively guide the conductive filament in the smallest area.In addition,the conductance of devices III can be continuously tuned by changing the pulse parameters,which provides an important basis for spike-timing-dependent plasticity?STDP?.After effectively demonstrating that quantum dots can enhance the local electric field and guide the growth of conductive filaments,it has also been found that random conduction in the memristor is caused by the high diffusivity of Ag⁺ions,and the low nucleation position and probability of nucleation.Deformation of the silk.It is difficult to effectively control the dispersion range of the switching voltage within 10%.Therefore,we have proposed a novel memristor based on carbon QDs demonstrating good performance due to low diffusivity of carbon cluster ions and high nucleation sites and nucleation probability.In particular,the switching voltage distribution has been significantly centralized,which the variation???in the SET and RESRT voltages is only 1.76%and 5%.TEM images confirm that carbon CF exists between the two electrodes.Four types of STDP learning and Pavlovian associative learning rules are faithfully emulated in CMDs.On the basis of experimentally measured LTP/LTD characteristics of the synaptic device,and a recognition accuracy achieved by SPL based artificial neural network?ANN?can reach 92.63%for digit recognition.Our realization of carbon CFs model synaptic memristor provides a promising physical mechanism of carbon filament model for bio-inspired neuromorphic computing.Because of its simple structure,the memristor is very conducive to the three-dimensional integration of semiconductor sneak circuits.However,due to the influence of parasitic currents between device resistances,especially the interaction between low-resistance devices during the integration of three-dimensional arrays,the power consumption of the entire circuit chip is increased,and even the fixed memory during data read.The possibility of misreading when reading the resistor status.We have shown exceptional selection characteristics in MoS₂ QDs-based threshold switching devices.SA seep slope of<2 mV/dec,high selectivity of 10⁶ were achieved simultaneously in a device.The feasibility of integrating such a selector with a memristor has been demonstrated by fabricating 1S1R integrated devices,which could benefit numerous applications with large crossbar arrays,including nonvolatile memory and neuromorphic computing.