24 results about "Optical neural network" patented technology

The application discloses a kind of diffractive light neural network computing systems and methods for realizing pure optical nonlinearity, belong to photonic computing and artificial intelligence hardware technical field.Its system includes coherent light source, linear light computing unit, nonlinear light activation unit and optical detection unit arranged in order along optical path, wherein linear light computing unit is programmable phase modulation using spatial light modulator, nonlinear light activation unit is programmable binary amplitude modulation using digital micromirror device, and both are directly coupled to form pure optical computing path without photoelectric conversion on optical path.Its method trains neural network through customized loss function, drives nonlinear activation function output to binary convergence, and maps the parameters obtained by training into phase map and binary mask respectively and loads to hardware.The application realizes true all-optical nonlinear computation, with the advantages of high energy efficiency, extremely low delay, compact structure and strong parallel processing capability.

The application discloses an optical matrix vector multiplier based on a unitary matrix-diagonal matrix decomposition. The multiplier comprises a Mach-Zehnder interferometer array for realizing a synthetic unitary matrix Omega and a modulation unit for realizing a corresponding diagonal matrix Sigma, wherein a target weight matrix M satisfies a decomposition form of M=ΩΣ. The method fuses two unitary matrices required for mapping in a traditional singular value decomposition into a single synthetic unitary matrix Omega, so that a unitary transformation can be completed by only one interferometer array of optical hardware, and a complete matrix multiplication is realized in cooperation with the modulation unit. Through algorithm-hardware collaborative design, the number of basic optical interference units required is significantly reduced, the hardware complexity, manufacturing cost and system power consumption of the optical neural network are greatly reduced while the computing performance is maintained, and an effective solution is provided for efficient optical computing acceleration of artificial intelligence.

This invention discloses a highly self-assembled dual quantum dot / graphene heterojunction broadband memristor, its fabrication method, and its applications, belonging to the field of inorganic semiconductor technology. The method includes the following steps: mixing highly self-assembled cerium oxide quantum dot ink and PbS quantum dot ink with a graphene dispersion and stirring until homogeneous to form a uniformly loaded quantum dot / graphene dispersion; then using the quantum dot / graphene dispersion to construct a dual quantum dot / graphene heterojunction film on a silicon wafer surface, thus obtaining the dual quantum dot / graphene heterojunction broadband memristor. This invention overcomes the inherent limitations of the photoresponse range of single materials, achieving broadband optoelectronic memristor behavior in the ultraviolet to near-infrared bands. Simultaneously, the memristor of this invention exhibits rich synaptic plasticity under light pulse modulation, providing core support for constructing a novel integrated optoelectronic system for broadband sensing, storage, and processing. It possesses strong technical adaptability and industrialization prospects in cutting-edge fields such as bionic vision, optical neural networks, and intelligent sensing.

A machine vision task, machine learning task, and / or classification of objects is performed using a diffractive optical neural network device. Light from objects passes through or reflects off the diffractive optical neural network device formed by multiple substrate layers. The diffractive optical neural network device defines a trained function between an input optical signal from the object light illuminated at a plurality or a continuum of wavelengths and an output optical signal corresponding to one or more unique wavelengths or sets of wavelengths assigned to represent distinct data classes or object types / classes created by optical diffraction and / or reflection through / off the substrate layers. Output light is captured with detector(s) that generate a signal or data that comprise the one or more unique wavelengths or sets of wavelengths assigned to represent distinct data classes or object types or object classes which are used to perform the task or classification.

The application discloses a large-scale and reconfigurable three-dimensional integrated optical neural network chip, and belongs to the field of optical neural networks, and comprises an optical information input layer, a reconfigurable hidden layer, a detection output layer and a feedback control structure; the reconfigurable hidden layer comprises a plurality of cascaded waveguide structures, wherein each waveguide structure comprises an independent modulation area and a continuous coupling area, the optical waveguides of the independent modulation area are independently transmitted and independently modulated by a light modulator; after reaching the continuous coupling area, the evanescent waves of the waveguide array are continuously coupled with each other in the transmission direction and are integrally modulated by the light modulator; and the feedback control structure adjusts a light modulator loading signal according to the detected output light intensity spatial distribution information, so that the reconfigurable hidden layer identifies the current input signal. The application can greatly improve the actual computing power of the current on-chip integrated optical neural network, and also provides a solution to the problems of low reconfigurability and large system size of the current three-dimensional spatial diffraction optical neural network.

The application relates to the technical field of artificial intelligence, and discloses an optical neural network device, a stress detection method thereof and a robot. The device comprises: a stress luminescent material array structure for emitting an optical matrix signal when subjected to a spatially distributed mechanical signal; a photodetector array for converting the optical matrix signal into a current signal; and a signal amplification unit for converting the current signal into a voltage signal, amplifying the voltage signal, and driving a driving unit according to the amplified voltage signal. The application utilizes the stress luminescence phenomenon to convert the mechanical signal into an optical signal, utilizes a neural network to convert the spatially distributed mechanical signal into a spatially distributed optical matrix signal, converts the optical matrix signal into a current signal, and then further amplifies the current signal to obtain a voltage signal as a driving signal of the driving unit, so that the device can quickly respond to a complex stress signal.

This invention discloses a multilayer nonlinear scalable optical neural network computing system based on 2D / 3D composite perovskite thin films, including a light source module, an optical convolution module, a nonlinear activation module, a photodetector module, and an electrical computing module. The nonlinear activation module utilizes 2D / 3D perovskite thin films formed from PEABr, CsBr, and PbBr₂, further cascaded to achieve high photoluminescence quantum yield and excellent film quality. It can efficiently convert the optical signal passing through the optical convolution module and achieve nonlinear activation. This invention employs an end-to-end joint optimization strategy to train optical masks combined with corresponding electrical back-ends to perform general visual processing tasks, including but not limited to image denoising and image classification. It can achieve high-precision image denoising and classification in a passive, lens-free manner, improving the computational depth and feature extraction capabilities of the photonic neural network.

The application relates to a method for enhancing the nonlinear optical performance of SnS2 nanosheets through electrostatic doping. First, SnS2 nanosheets are prepared on a fluorine-doped tin oxide (FTO) substrate by a chemical vapor deposition method, and then hydrogen ion intercalated SnS2-H + The electrostatic doping causes the band gap to shrink and a strong built-in electric field to be generated, so that the nonlinear optical performance of the material is enhanced. In addition, the saturated absorption response of the material enables the application to an optical neural network as a nonlinear activation function, and exhibits application potential in machine learning tasks. The SnS2-H + The application establishes a promising and convenient nonlinear optical material, and provides new insights for the design exploration and simple synthesis of high-performance nonlinear absorption materials.

This application belongs to the field of optoelectronic computing and artificial intelligence hardware acceleration technology, specifically disclosing an optical computing system and a method for implementing optical neural network operations. A digital micromirror chip is used to disperse a first femtosecond laser into beams of equal intensity but different angles. A dichroic mirror is used to combine a third femtosecond laser with the split beams. A two-dimensional material array performs operations on the CONV-BN layer of a convolutional neural network. A photodiode is used to collect the change in total transmitted light intensity after passing through the two-dimensional material array and perform photoelectric nonlinear conversion to obtain the relative transmittance change, realizing the nonlinear ReLU activation function. This application reduces inter-operator data transfer overhead by 60%, significantly reduces latency caused by the memory wall effect, and uses a DMD chip to encode the input feature map into an optical signal array. A single frame of input can process the entire feature map in parallel, achieving high parallelism.

The application discloses a kind of optical deep neural network chips based on time domain, belong to optical computing technical field.A kind of optical deep neural network chips based on time domain, including sequentially connected: optical signal input area, input data modulator area, convolution weight modulator area, first optical nonlinear unit area, pooling modulator area, fully connected weight modulator area, second optical nonlinear unit area, high-speed photoelectric detector area and time domain integrator area.Optical deep neural network chip includes the three kinds of functions of convolution, pooling and fully connected in classic convolutional neural network, covers input layer, hidden layer and output layer in depth network;Multiple optical nonlinear unit area cascades are used to realize multilayer depth optical neural network, and large-scale high-throughput optical computing is realized by time dimension.Multiple layers of optical deep neural network computing are realized, and are suitable for artificial intelligence big model training, semantic segmentation, image recognition and medical instrument modeling imaging and other optical computing fields.

The application provides an optical neural network module based on phonon polariton regulation and a design method, comprising a heterostructure of a waveguide, a phase change material and hexagonal boron nitride, layers of the neural network are connected through diffraction of phonon polaritons, each unit on the diffraction layer is a sub-wave source of a secondary spherical wave; input of a neuron of an arbitrary layer is output of all neurons of a previous layer, and the input is superimposed on the neuron after diffraction; weight of each neuron is defined as influence of a unit structure on the diffraction layer on phase and amplitude of the phonon polariton; input data is input from an input layer, propagation of waves between diffraction layers is calculated, and output results of an output layer are obtained; then, weight of a neuron of each diffraction layer is continuously optimized through an error back propagation algorithm, a trained neural network is obtained after several cycles, and the neural network result is written on the phase change material through laser with different wavelengths and powers.

The present application relates to the technical field of optical neural network computing, and provides an optical neural network computing system based on doped gain medium nonlinear activation, which comprises a signal input module, a core computing module and a signal output module.The signal input module is used for linearly mapping a preprocessed data matrix into optical power signals of different wavelengths, transmitting the signals to an optical fiber link after convergence, and avoiding interference of reflected light on the stability of a signal source.The core computing module comprises a doped optical fiber connected with a pump laser, and the doped optical fiber works in a gain near saturation region by injecting energy through the pump laser, and the gain saturation characteristics of the doped optical fiber are used to realize nonlinear activation conversion of optical signals.The signal output module is used for separating the multi-wavelength optical signals after nonlinear conversion, attenuating and regulating each channel optical signal based on training weights, and reading and outputting intensity after superimposing each optical signal, as a prediction result of a neural network.The present application realizes efficient computing with full optical domain, low delay, wide bandwidth and low power consumption.

This disclosure provides an optical neural network system and its training method and apparatus. The method may include: performing an initial optical configuration on a main optical modulation element, that is, initializing the phase distribution of each pixel block modulation unit to a basic phase, the phase distribution being used to determine the optical projection position of the corresponding pixel block modulation unit, the main optical modulation element including a permutation block, the permutation block including a pixel block modulation unit; determining the initial optical configuration as the current optical configuration, and performing the following first process: obtaining a modulated light field obtained by the main optical modulation element modulating the input light field according to the current optical configuration, updating the weight parameters of the neural network inference unit according to the modulated light field, if it is determined that the neural network inference unit has converged but does not meet the termination condition, adjusting the phase distribution of the pixel block modulation units in at least one permutation block to obtain the updated current optical configuration, and repeating the first process until convergence and the termination condition is met.

The present application relates to the technical field of optical computing. Provided are an optical neural network based on dispersive nonlinearity, and an optical neural chip. The optical neural network comprises: an optical encoding module, which comprises a tunable multi-wavelength light source, and is used for encoding data to be processed into optical input signals with different wavelengths; a tunable dispersion module, which uses a tunable dispersive material based on structural dispersion or material dispersion to modulate the intensities of the optical input signals on the basis of the wavelengths of the optical input signals, so as to obtain optical output signals, wherein there is a nonlinear functional relationship between the intensities of the optical output signals and the wavelengths of the optical input signals; and a fully connected layer module, which comprises a photodetector, and is used for detecting the intensities of the optical output signals, so as to obtain a prediction result of said data on the basis of the intensities of the optical output signals. The present application provides an optical neural network which has a simple structure, is capable of realizing arbitrarily tunable nonlinear activation functions, and is capable of implementing arbitrarily complex neural network functions.

Provided is a three-dimensional (3D) heterogeneous integrated optical computing chip, which belongs to the field of optical computing. The computing chip is formed by three parts: a spatial diffraction optical neural network chiplet, a planar matrix reconfigurable optical neural network component, and a planar switchable optical switch module. Through a 3D heterogeneous integration method, one or more spatial diffraction optical neural network chiplets, the planar matrix reconfigurable optical neural network component, and the planar switchable optical switch module are combined together to achieve efficient multi-task optical computing processing. The disclosure utilizes the combination of the spatial diffraction optical neural network and the planar matrix reconfigurable optical neural network, breaking through the conventional optical computing architecture, which has broad application prospects in the field of optical computing and fills the gap in the relevant technical field.

The application provides an integrated display system based on an optical neural network, and relates to the technical field of display, and comprises a perception layer, an optical neural network calculation layer and a display layer; the perception layer is used for perceiving external environment signals, modulating input light signals according to the external environment signals, and forming standard light input signals; the optical neural network calculation layer is used for receiving the standard light input signals through a preset weight parameter optical neural network, performing matrix multiplication operation in parallel in the optical domain based on the modulated light signals, performing nonlinear activation output of light field signals, and coupling the light field signals to each display unit of the display layer through an optical output structure; the display unit comprises a photoelectric conversion structure and a light-emitting control structure, the photoelectric conversion structure is used for converting the received coupled light signals into driving electric signals and sending the driving electric signals to the light-emitting control structure, so as to control the light-emitting brightness of the display unit. The integrated display system can effectively reduce the energy consumption and delay of the display system and improve the integration.

The application discloses a kind of multi-layer nonlinear optical neural network computing systems based on linear signal re-encoding, belong to optical neural network computing technical field, including: control circuit and optical processing chip system;The control circuit is used to realize the loading of input signal and the reading of output result;The optical processing chip system includes light source, integrated optical computing chip and detector, and the light carrier emitted by light source generates input optical signal after being modulated by input signal, and after being processed by multi-layer nonlinear activation in linear calculation area, linear re-encoding area and high-speed modulation area on integrated optical computing chip, output result is read by detector and returned to control circuit.The application utilizes pure linear device to realize all-optical nonlinear network, cooperates analog activation function by re-encoding and modulation, needs only single physical layer to construct depth architecture without photoelectric conversion, with the advantages of all-optical processing, CMOS compatible, high speed and low power consumption.

The application discloses an optical neural network device and an optical device, and belongs to the technical field of optical chips, and comprises a waveguide structure, the waveguide structure is used for propagating an optical signal; at least one first diffraction structure; at least one second diffraction structure; the first diffraction structure and the second diffraction structure are integrally formed with the waveguide structure, and the first diffraction structure and the second diffraction structure are used for adjusting at least one of phases, intensities, directions, polarizations and modes of the optical signal; the optical signal enters the waveguide structure through the first diffraction structure, the optical signal processed by the first diffraction structure is transmitted to the second diffraction structure, and the optical signal processed by the second diffraction structure is emitted from the waveguide structure. The application sets multiple diffraction structures on the waveguide structure, and the multiple diffraction structures are integrally formed with the waveguide structure, so that the optical path transmission of the optical neural network device is more stable.

The application relates to a low-threshold plasmonic graphene all-optical nonlinear activator, which comprises a substrate, a gold film is covered on the substrate, a device pattern is etched on the gold film, and a graphene film is covered on the gold film. The gold film pattern of the plasmonic structure can generate a very strong local field enhancement effect, significantly enhancing the interaction between light and graphene; by utilizing the Pauli blocking effect and the ultrafast carrier dynamics of graphene, combining the local field enhancement effect of the plasmonic structure to strengthen the light-matter interaction, low-threshold nonlinear activation is realized. The device activation threshold is as low as 7.03 nW, and the accuracy in MNIST handwritten digit recognition is 96.7%; by eliminating the photoelectric conversion link, a low-power-consumption, high-energy-efficiency optical domain processing scheme is provided, and the practicability of optical neural network calculation is significantly improved.

The present application relates to a kind of all-optical nonlinear neural network systems based on linear system, belong to optical neural network technical field.Solve the bottleneck problem that optical neural network relies on nonlinear material or photoelectric conversion to realize nonlinear operation.Techinical scheme includes input module, and input signal is loaded in annular resonant cavity resonance wavelength detuning amount;Coupling module controls interlayer weight;Neuron module passes through multilayer annular resonant cavity linear transmission signal;Output module loads output signal in probe light power.Input and output signal are loaded in different physical quantities, so that linear system realizes nonlinear function.Affordable effect includes breaking through optical nonlinear operation bottleneck, significantly reduce system complexity and energy consumption, enhance compatibility and practicality, improve parallel computing efficiency and system scalability.

This invention provides an optical neural network and optical neural chip based on dispersive nonlinearity, relating to the field of optical computing technology. The optical neural network includes: an optical encoding module comprising a tunable multi-wavelength light source for encoding data to be processed into light input signals of different wavelengths; a tunable dispersion module utilizing a tunable dispersion material based on structural or material dispersion to modulate the intensity of the light input signal according to its wavelength, thereby obtaining a light output signal; wherein the intensity of the light output signal and the wavelength of the light input signal have a nonlinear functional relationship; and a fully connected layer module comprising a photodetector for detecting the intensity of the light output signal to obtain a prediction result of the data to be processed based on the intensity of the light output signal. This invention provides an optical neural network with a simple structure, capable of implementing arbitrarily tunable nonlinear activation functions, and capable of realizing arbitrarily complex neural network functions.

This application provides a fiber optic integrated optical neural network sensing system, classification method, terminal, and medium. The sensing system includes a fiber optic sensing unit, an optical modulation unit, and an optical analysis unit. The fiber optic sensing unit acquires a speckle beam and transmits it to the optical modulation unit. The optical modulation unit receives the speckle beam, modulates it, and transmits it to the optical analysis unit. The optical analysis unit receives the modulated speckle beam, analyzes it, and obtains analysis results. The optical modulation unit and the optical analysis unit are communicatively connected. The optical analysis unit acquires the analysis results and transmits them to the optical modulation unit. When the optical modulation unit receives the analysis results, it modulates the received speckle beam based on the analysis results; otherwise, the optical modulation unit modulates the received speckle beam based on a preset initial modulation strategy. The sensing system provided by this application can achieve high-precision classification and integrates the functions of acquiring and processing information, achieving a unified sensing effect.

This application relates to the field of biomimetic neurotechnology and proposes a pluggable biomimetic optical neural network system with adaptive sensing capabilities. The system includes a laser source, a pinhole amplifier, a digital micromirror device, a beam splitter prism unit, a photodetector, a pluggable metasurface structure, a spatial light modulator (SLM), a mirror, and a digital camera with a charge-coupled device (CCD) image sensor. This application can adjust the state of the pluggable metasurface structure according to the contrast of the input image light, and adaptively extract high-frequency information of the object based on the contrast of the surrounding environment, thereby improving the accuracy of target recognition tasks. Furthermore, the activation level of neurons in the diffraction layer can be updated in real time based on the initial classification result of the input image light, improving the dynamic adaptive adjustment capability of visual attention, and thus achieving adaptive adjustment to the perceived target.