Organic Memristive Platforms And Its Neuromorphic Computing With Activity Dependent-Mode Selection Characteristics

The advance of computation is the driving force of social development.The neuromorphic computing aims to improve the information processing ability of the computer by implementing the information processing mode in the brain's neural network,and enhance its cognitive ability.Constructing the neuromorphic structure using the basic device is increasingly attracted considerable attention.With simple structure,the memristor,which is known as the fourth basic element,can adaptively regulate its conductive state to memorize the input stimulating signals.And the stimulusresponse pattern is analogous to the plasticity of the elementary biological synapse,which contributes the implementation of the brain-inspired functions by memristors.Synaptic plasticity is the physiological basis of learning and memory function in neural network.However,at present,the emulated plasticity is limited to the plasticity in the framework of basic Hebb rules.The researching results in neuroscience have pointed out the limitations of the Hebb rules,and the researches in algorithm science also indicate the necessity of realizing comprehensive synaptic functions.Therefore,in order to construct the neuromorphic structure with comprehensive and effective functions,except for the synaptic plasticity based on the Hebb rules,it is essential to implement the comprehensive synaptic plasticity in memristors.In addition,from the perspective of device design and construction,the prevalent physical microprocesses for developing memristors are mainly based on shaping of the ion-type conductive filament under electric field,of which the smoothness and gradient of conductive state are highly sensitive to the layer composition and the bulk and interface components.Thus,it may be awkward to precisely control the growth location,orientation and rate of ionic filaments for inner imperfect structures in these devices,which result in the mutational,fluctuant characteristics of the conducting curves and nonhomogeneous operation state,namely low signal-to-noise ratio(SNR).These characteristics have adverse influences on the memristive performances for emulating the neural activities,because the regulatory neural states are characterized by smooth and gradual variations as a function of the overall stimulations.Based on the recent researching works of resistive memory and the deep understanding of the relationship between memory and memristor,we take the organic semiconductor-based optoelectronic devices as the researching object and take the multi-form memristive responses as the researching perspective to construct a serials of memristive platforms with excellent smooth and durative characteristics.During the researching processes,many strategies are adopted,such as adjusting the electron/hole processes,active layers and device structure,reconfiguring the device energy levels and regularly managing the ionic movements through functional molecules.As a result,a serials of organic memristive platforms were obtained with excellent smooth and durative memristive characteristics for comprehensive neuromorphic computing.And based on the regulated memristive behaviors,we intend to develop the novel neuromorphic computing functions,which are beyond the prevalently-emulated Hebb rules.Firstly,compared with the ionic migration with random characteristics,we fabricated the electrontype memristor using the stable organic copper phthalocyanine(Cu Pc)molecule,where the smooth memristive behaviors result from the gentle charge transfer and trapping processes.Based on this memristor,we implement the learning-memory performances in the framework of Hebb?s rules,homeostatic plasticity and habituation as well as sensitization plasticity beyond the Hebb?s rules.Consequently,the Cu Pc-based memristor provide us a multi-functional element for the neuromorphic computing architecture with self-stabilizing properties.In the Cu Pc-based memristor we further introduced a buffer layer-Li F at the anode,and obtained the electron-type memristor with rectifying characteristics through adjusting the device energy levels.Based on this,we constructed the rectifying electrical synapse and resistive electrical synapse complementary to the prevalently-emulated chemical synapse.Secondly,in the case of the ionic memristors,in order to optimize the randomness of the ionic transport in ionic memristors,we devised the liquid ionic memristor using the Pb I2 solution and the organic amine solution of perovskite raw materials.In these devices,under the buffering effects of the organic solvent the ionic movement results in the smooth memristive transition under more than 50 cyclic sweeps,embodying the advantage of the liquid ionic memristor.We also fabricated the planar device with graphene oxide(GO).The ionic transport in this device is smooth and easily controllable.We explored the stimulus amplitude,frequency,relaxation-dependent polarization and depolarization behaviours.Then we emulated the synaptic release and recovery processes and further emulated the controllable multiple synaptic depression,which is complementary to the prevalently-emulated synaptic excitation.Thirdly,considering the disadvantage of the two-terminal memristor that it is difficult to real-time transmit and regulate the memristive response,based an ambipolar transistor memory we fabricated the light-regulated neuromorphic platform with impressive smooth responses and long-term endurable characteristics.We implemented the light-inspired learning-memory functions in a synapse and devise the advanced synapse array.More importantly,we investigated the mutual regulation of excitation and inhibition,and implemented their sensitive mode transformations as well as the homeostasis property.Thus,these emulations demonstrate the advantages of the optoelectronic platform in implementing the brain-inspired functions in different neural structures.