Self-powered Fully Light-controlled Synaptic Device And Neuromorphic Computation Based On Si/ZnO Heterojunction

With the rapid development of science and technology,artificial intelligence,cloud computing and brain-like computing based on big data have been widely applied in many fields.In order to meet people’s increasing information processing needs,the data processed by computers increases exponentially every year.However,the traditional computer based on the"von Neumann architecture"is physical separation of memory and processor when computing or processing massive data.The frequent transmission of data between different units not only lead to significant time delay,but also produce high power consumption,so it is difficult to meet people’s demand for real-time and efficient processing of massive data.The human brain has the advantages of high performance and low power consumption.The brain is a network composed of10¹¹ neurons and 10¹⁵ synapses through complex connections,among which the biological synapses are the most basic unit to realize the brain function.Inspired by the function and structure of the human brain,brain-like computing and network emerged.The construction of synaptic devices to simulate synaptic plasticity is the basis of brain-like computing.Early artificial synaptic devices mainly used electrical signals to regulate synaptic weight and simulate synaptic plasticity.However,devices regulated by electrical signals are limited by bandwidth and that can cause problems such as resistance capacitance delay and current crosstalk.Then the researcher introduced optical signals to regulate the weight of synaptic devices.However,in some devices,optical pulses only have unidirectional control over the weight of the device.Bidirectional and multistage control over the weight of synaptic devices needs to be achieved by means of electrical pulses,so as to realize brain-like calculation,which cannot completely solve the problems existing in the regulation of electrical pulses.Therefore,the regulation of synaptic device weight by all optical pulses is an effective way to solve the problems of bandwidth limitation,resistance capacitance delay and current crosstalk in the weight control of synaptic devices by electrical pulses.It is of great significance to promote the development of neuromorphic computing.Based on the above background,a p-Si/n-ZnO heterojunction device is prepared in this paper.The internal electric field of the heterojunction is used to drive the device to realize the function of self-power supply.The response of the device to visible light is used as the synaptic weight,and the bidirectional and multilevel regulation of the synaptic weight is achieved by ultraviolet and near-infrared irradiation.The device performance is used to update the weight,so as to realize the classification and recognition of handwritten numeral,and the convolution kernel based on the device performance realizes the image preprocessing.The research content of this paper mainly includes three parts:（1）Fabrication and basic properties of Si/ZnO heterojunction devices.ZnO nanowires were grown by hydrothermal method,and electrodes were prepared by magnetron sputtering and electron beam evaporation,so as to construct two ends of Si/ZnO heterojunction devices.After SEM,XRD,TEM and UV-VIS absorption spectra,it is shown that the ZnO nanowire arrays grow vertically and they have good crystallinity and abundant surface interface states,which is favorable for the control of device performance by light pulse irradiation.Subsequently,the effects of ultraviolet and near-infrared irradiation on the optical absorption properties,electrical properties and photoelectric response characteristics of the device were studied by spectral,I-V and I-t photoelectric characterization methods.The results show that ultraviolet and near-infrared irradiation can effectively increase and decrease the conductance of the device.（2）Performance and principle study of self-powered fully-lighted controlled synaptic devices based on Si/ZnO heterojunction.The responsivity of the device to visible light after ultraviolet and near-infrared light irradiation increases and decreases,respectively.The irradiation time and power density of ultraviolet and near-infrared light can realize the regulation of visible light responsivity.The fabricated device are responsive to light wavelength of 415-970 nm,and have a wide spectrum response from visible to near infrared.By studying the regulation principle,it is found that ultraviolet and near-infrared irradiation mainly regulate the surface and interface states of the material,thus regulate the resistance of the device and the height of the barrier on the ZnO side.After ultraviolet irradiation,the resistance of the device decreases with the effective barrier on the ZnO side,and the decrease of resistance plays a dominant role,leading to the increase of the photocurrent responsivity of the device.After near infrared light irradiation,the resistance of the device increases and the effective barrier on the ZnO side decreases,resulting in the decrease of the photocurrent responsiveness of the device.（3）Application based on fully-lighted controlled synaptic devices.LTP/LTD characteristic curves of synapses are obtained by ultraviolet and near-infrared light irradiation.The artificial neural network is constructed and parameters are extracted by weight updating formula,and it is found that LTP curves after ultraviolet irradiation have good linearity.When the power density of UV irradiation is 6.04 m W cm^-2 and near-infrared irradiation is non-uniform,the accuracy of handwritten numeral recognition of the feedback network constructed by the device performance is the highest,which is 78±2%.In addition,the image preprocessing,such as mean filtering,Gaussian filtering and image edge detection,can be realized by using the broad spectral response characteristics of the device.Our device can distinguish the response to irradiated light and detected light,and directly realize the response to optical signal.It is expected to build sensory-deposit-computing artificial network on the hardware and realize efficient and rapid image processing,which will promote the construction and development of new neuromorphic computing hardware network.