Structure and preparation method of a-si-based optical readout neurosynaptic device

Abstract

The present invention provides an a-Si-based optical reading neural synapse device structure and a preparation method thereof, comprising a "metal / a-Si / metal" surface plasmon waveguide and an "upper electrode / double resistive switching layer / lower electrode" embedded therein Electrode" memristor; surface plasmon waveguide has a vertical three-layer structure of "second metal layer / dielectric layer / first metal layer" from top to bottom; memristor has "upper electrode / second resistive switching layer" from top to bottom / First resistive switching layer / lower electrode" vertical four-layer structure, the first resistive switching layer and the second resistive switching layer of the memristor are connected horizontally with the dielectric layer of the surface plasmon waveguide as the optical signal propagation channel; the present invention realizes the neural synapse The weighted optical reading makes the optical reading neurosynaptic device with the amplitude and phase of the optical signal as the synaptic weight, which has the incomparable advantages of the traditional synaptic device with the resistance as the synaptic weight. Surface plasmon waveguide can make the light The signal is transmitted beyond the diffraction limit, which is conducive to the further reduction of the device size.

Description

technical field [0001] The invention belongs to the field of silicon-based photonic integrated devices and neuromorphic chips, and in particular relates to an a-Si-based light-reading neurosynaptic device and a preparation method thereof. Background technique [0002] For a memristor with a "metal / dielectric layer / metal" sandwich structure, if different bias voltages are applied, the resistance of the device will show a nonlinear change. This non-linear change in resistance is caused by the formation or disappearance of conductive channels in the dielectric layer under different bias voltages. Interestingly, the connection strength of this nanoscale filamentous conductive channel varies with the magnitude and duration of the bias voltage. This property is very similar to the working mechanism of synapses connecting different neurons in the biological nervous system. It is this similarity between memristors and biological system synapses that makes them very suitable as syn...

AI Technical Summary

Problems solved by technology

However, so far, in the published patents of synaptic devices based on the memristive effect, all are based on electronic media to read synaptic weights, which can be called "electrically modulated electrical reading", and do not use photons. medium to read the weight of synapses (reference basis: CN 104916313 A, CN 10378055A, CN105287046A, CN105304813A, CN 104934534 A, CN104376362 A)

Images