CN114358270A - Photon neural network of multidimensional multiplexing technology and implementation method - Google Patents
Photon neural network of multidimensional multiplexing technology and implementation method Download PDFInfo
- Publication number
- CN114358270A CN114358270A CN202210016124.3A CN202210016124A CN114358270A CN 114358270 A CN114358270 A CN 114358270A CN 202210016124 A CN202210016124 A CN 202210016124A CN 114358270 A CN114358270 A CN 114358270A
- Authority
- CN
- China
- Prior art keywords
- neural network
- light
- photonic
- mode
- nonlinear activation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 106
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000005516 engineering process Methods 0.000 title claims abstract description 16
- 239000000835 fiber Substances 0.000 claims abstract description 32
- 239000011159 matrix material Substances 0.000 claims abstract description 18
- 239000006185 dispersion Substances 0.000 claims abstract description 11
- 230000000694 effects Effects 0.000 claims abstract description 7
- 230000004913 activation Effects 0.000 claims description 29
- 230000003287 optical effect Effects 0.000 claims description 19
- 239000006096 absorbing agent Substances 0.000 claims description 16
- 230000001419 dependent effect Effects 0.000 claims description 7
- 239000013307 optical fiber Substances 0.000 claims description 7
- 239000000463 material Substances 0.000 claims description 6
- 230000005693 optoelectronics Effects 0.000 claims description 2
- 229910016001 MoSe Inorganic materials 0.000 claims 1
- 238000004364 calculation method Methods 0.000 abstract description 5
- 210000002569 neuron Anatomy 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 7
- 230000005540 biological transmission Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 229910021389 graphene Inorganic materials 0.000 description 1
- 239000012212 insulator Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- MHWZQNGIEIYAQJ-UHFFFAOYSA-N molybdenum diselenide Chemical compound [Se]=[Mo]=[Se] MHWZQNGIEIYAQJ-UHFFFAOYSA-N 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 238000002834 transmittance Methods 0.000 description 1
Images
Landscapes
- Optical Modulation, Optical Deflection, Nonlinear Optics, Optical Demodulation, Optical Logic Elements (AREA)
Abstract
The invention relates to the technical field of photon neural networks, and discloses a photon neural network of a multidimensional multiplexing technology and an implementation method thereof, wherein the photon neural network comprises the following steps: the laser generates pulse light, the pulse light is input into the multi-dimensional multiplexing module, after multiplexing of a mode, a fiber core and a wavelength, the pulse light is broadened in time due to a dispersion effect, and then is transmitted into the photonic neural network through the demultiplexer, the photonic neural network firstly completes multiplication of a matrix and then completes summation operation of the neural network, and the light after the operation is finally received by the photodiode; by introducing the mode and the fiber core space dimensionality, more space freedom degrees are provided for the photon neural network, serial calculation among neurons in the photon neural network is converted into parallel calculation, the running speed of the photon neural network is greatly improved, the hardware requirement of the system is reduced, the size of the system is further reduced, the overall stability of the system is improved, and the classification task of handwriting digital recognition and other data sets to realize pictures can be realized.
Description
Technical Field
The invention relates to the technical field of photonic neural networks, in particular to a photonic neural network of a multi-dimensional multiplexing technology and an implementation method thereof.
Background
The operation speed of the existing electrical neural network is limited by the bottleneck of electrical equipment, most of the power of the electrical neural network is used for data transmission between the processing unit and the memory module, and the power efficiency is relatively low. With the enlargement of the neural network scale, electronic devices with large bandwidth are urgently needed to accelerate the transmission rate of signals. However, these electrical devices are difficult to manufacture. In the photonic neural network, photons replace electrons to serve as a carrier for signal transmission, and no extra power is needed to be used for transmitting information between the memory module and the computing unit, so that the power efficiency of the photonic neural network is obviously higher than that of an electric neural network. Meanwhile, the existing high-speed optical fiber communication technology also supports high-speed transmission of information between different network layers of the photonic neural network, so that high-speed operation of the photonic neural network is realized.
However, the existing photon neural network mainly adopts serial calculation, has small construction scale, low operation speed and higher requirement on hardware.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a photon neural network of a multi-dimensional multiplexing technology and an implementation method.
In order to achieve the above purpose, the invention provides the following technical scheme:
a method for realizing a photon neural network of a multidimensional multiplexing technology comprises the following steps: the laser generates pulse light, the pulse light is input into the multi-dimensional multiplexing module, the pulse light is firstly subjected to multiplexing of a fiber core, a mode and a wavelength, the pulse light is broadened in time due to a dispersion effect, then the pulse light is transmitted into the photonic neural network through the demultiplexer, the photonic neural network firstly completes multiplication operation of a matrix and then completes nonlinear operation of the neural network, and the light after the operation is finally received by the photodiode.
In the present invention, it is preferable that a light source operating in the 980nm band is used as a pump to generate an ultra-short optical pulse by passive mode locking, and each of the photoelectric devices easily receives the generated light.
In the present invention, it is preferred that saturable absorbers based on the scientific two-dimensional material MoSe2 produce intensity-dependent losses, thereby producing narrowed pulses with pulse widths on the order of picoseconds and femtoseconds.
In the present invention, preferably, the multiplexed light is transmitted in parallel by a multicore few-mode fiber, and two spatial dimensions of a mode and a fiber core are provided for the photonic neural network, and simultaneously, the dispersion effect in the fiber broadens the light pulse in time.
In the present invention, preferably, the matrix multiplication operation of the photonic neural network is implemented by two waveform generators and two intensity modulators, the waveform generators are used for generating electric signals of arbitrary sequence, and the generated electric signals are modulated onto the light pulses by the intensity modulators.
In the present invention, it is preferable that the nonlinear operation of the neural network is performed by a nonlinear activation unit, and a nonlinear activation function in the nonlinear activation unit is provided by a saturable absorber having an intensity-dependent loss characteristic.
In the present invention, preferably, the light passing through the demultiplexer is modulated by two electro-optical modulators, X and W respectively, where X represents the input data of the photonic neural network and W represents the weight matrix of the neural network.
In the present invention, preferably, the intensity modulator in the photonic neural network functions as matrix multiplication to obtain the output Zi＝Wi·XiThe modulated light passes through an optical nonlinear activation unit to obtain the output X of the ith layer(i+1)＝fNL(Wi·Xi) Wherein f isNL(. cndot.) represents the operation of an optically nonlinear activation function.
In the invention, preferably, the neural network layers of the photonic neural network are connected by a loop ring, the output of the ith layer is used as the input of the (i + 1) th layer through the loop ring, after all the layers of operation are finished, the optical signal is received by the photoelectric detector, and after the gradient is updated, the optical signal is used for the forward operation of the next network.
A photon neural network of multidimensional multiplexing technology comprises a laser, a multidimensional multiplexing module, a neural network module and a photoelectric detector, wherein the laser is used for generating and transmitting ultrashort wave optical pulses to the multiplexing module,
the multi-dimensional multiplexing module comprises a plurality of multiplexers, a multi-core few-mode optical fiber and a plurality of demultiplexers, wherein the multiplexers are used for realizing multiplexing of wavelength, mode and fiber cores, the multi-core few-mode optical fiber provides two parallel operation dimensions of the mode and the fiber cores, the scale of the photon neural network is enlarged, meanwhile, optical pulses are spread by dispersion, and the demultiplexers are used for demultiplexing signals;
the neural network module comprises a neural network layer, an optoelectronic modulator, and a saturation circuitAnd the absorber, the intensity modulator and the nonlinear activation unit are arranged in the neural network layer for matrix multiplication to obtain an output Zi＝Wi·XiThe nonlinear activation unit operates the same to obtain the output X of the neural network layer(i+1)＝fNL(Wi·Xi) The saturable absorber provides a nonlinear activation function for the nonlinear activation unit, and outputs a classification result through iterative operation of a plurality of neural network layers.
Compared with the prior art, the invention has the beneficial effects that:
according to the method, by introducing the mode and the fiber core space dimension, more space degrees of freedom are provided for the photonic neural network, so that serial calculation among neurons in the photonic neural network is converted into parallel calculation, the running speed of the photonic neural network is greatly improved, the hardware requirement of the system is reduced, the volume of the system is further reduced, the overall stability of the system is improved, and the classification task of the handwritten number recognition and other data sets to the pictures can be realized.
Drawings
Fig. 1 is a schematic structural diagram of a photonic neural network of a multidimensional multiplexing technology according to the present invention.
Fig. 2 is a schematic structural diagram of a multidimensional multiplexing module of a photonic neural network of a multidimensional multiplexing technology according to the present invention.
Fig. 3 is a schematic diagram illustrating the input operation of the handwritten digit recognition data set of the photonic neural network of the multidimensional multiplexing technology.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It will be understood that when an element is referred to as being "secured to" another element, it can be directly on the other element or intervening elements may also be present. When a component is referred to as being "connected" to another component, it can be directly connected to the other component or intervening components may also be present. When a component is referred to as being "disposed on" another component, it can be directly on the other component or intervening components may also be present. The terms "vertical," "horizontal," "left," "right," and the like as used herein are for illustrative purposes only.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
Referring to fig. 1, a preferred embodiment of the present invention provides a method for implementing a photonic neural network of a multidimensional multiplexing technology, which provides more spatial degrees of freedom for the photonic neural network by introducing more spatial dimensions (modes and fiber cores), so that serial computation between neurons in the photonic neural network is converted into parallel computation, thereby greatly increasing the operating speed of the photonic neural network, reducing the hardware requirement of the system, further reducing the volume of the system, and improving the overall stability of the system, and includes the following steps: the method comprises the steps of generating pulsed light by using a laser, inputting the pulsed light into a multi-dimensional multiplexing module, firstly multiplexing a fiber core, a mode and a wavelength, widening the light pulse due to a dispersion effect, then transmitting the light pulse to a photon neural network through a demultiplexer, completing the multiplication operation of a matrix by the photon neural network, then completing the nonlinear operation of the neural network, and finally receiving the light after the operation by a photodiode, thereby completing the establishment of a large-scale photon neural network on the light.
As shown in fig. 1, in this embodiment, a passive mode-locked fiber laser provides ultrashort periodic pulses to provide light source signals for the following multidimensional multiplexing module, the passive mode-locked fiber laser adopts a light source working at a 980nm waveband as a Pump source (Pump), each photoelectric device easily receives light generated during working, and generates intensity-dependent loss based on a saturated absorber of a scientific two-dimensional material MoSe2, thereby generating narrowed pulses with pulse widths in picoseconds and femtoseconds.
Further, a polarization dependent isolator (PI-ISO) is connected between the saturable absorber and the Single Mode Fiber (SMF) for preventing the light source from being affected by back reflection noise or interference, so as to ensure stable operation of the laser.
As shown in fig. 2, in this embodiment, the multidimensional multiplexing module includes a multiplexer, a multi-core few-mode fiber, and a demultiplexer, a narrowed pulse emitted by a laser first passes through the multiplexer to multiplex a wavelength, a mode, and a fiber core, and then the multi-core few-mode fiber performs parallel transmission on the multiplexed light, the multi-core few-mode fiber is mainly used for parallel transmission of light pulses, and on the basis of wavelength division multiplexing, two spatial dimensions of the mode and the fiber core are provided for a photonic neural network, so that the scale of the neural network is greatly increased, and meanwhile, a dispersion effect in the fiber widens the light pulses in time, and the widened light pulses can be used for subsequent signal modulation.
In this embodiment, the matrix multiplication operation of the photonic neural network is implemented by two waveform generators (AWGs) and two intensity modulators, the waveform generators are configured to generate electrical signals of any sequence, the generated electrical signals are modulated onto optical pulses by the intensity modulators, the nonlinear operation of the neural network is performed by a nonlinear activation unit, a nonlinear activation function of the nonlinear activation unit is provided by a saturable absorber, the saturable absorber has an intensity-dependent loss characteristic, and as the peak light intensity of an incident pulse increases, the absorption coefficient of the saturable absorber gradually decreases, so that the light transmittance increases, and the saturable absorber can be implemented by using a tapered fiber and a two-dimensional material, where the two-dimensional material is selected from topological insulator materials such as graphene or molybdenum diselenide.
In the embodiment, ultrashort optical pulses sent by a passive mode-locked fiber laser are transmitted to a multidimensional multiplexing module, in the multidimensional multiplexing module, pulsed light with different wavelengths is used for space division multiplexing after passing through a wavelength division multiplexer, a multi-core few-mode fiber provides two parallel operation dimensions of a mode and a fiber core, the scale of a photonic neural network is enlarged, meanwhile, dispersion plays a role in broadening the optical pulses, and after passing through a demultiplexing device, each path of signals respectively realize the modulation of X and W through two electro-optical modulators, wherein X represents input data of the photonic neural network, and W represents a weight matrix of the neural network; the intensity modulator in the photon neural network plays the role of matrix multiplication to obtain output Zi＝Wi·XiThe modulated light passes through an optical nonlinear activation unit to obtain the output X of the ith layer(i+1)＝fNL(Wi·Xi) Wherein f isNLThe (h) represents the operation of the optical nonlinear activation function, the various neural network layers of the photonic neural network are connected by a loop, and the output of the i layer is used as the input of the i +1 layer through the loop. And after all the layer operations are finished, the optical signal is received by the photoelectric detector, and the gradient is updated to be used for the next network forward operation.
As shown in fig. 3, in the present embodiment, a photonic neural network based on a multidimensional multiplexing technique is applied to a classification task such as handwritten number recognition, a data set to be recognized of a handwritten number is input to an input end of the photonic neural network, picture data is modulated onto an intensity modulator 1 after being preprocessed, modulation of a weight W is realized by using the intensity modulator 2, multiplication of the handwritten number recognition data and the weight is completed, and a classification prediction of the neural network can be obtained through subsequent neural network processing, thereby realizing the recognition task of the handwritten number.
As shown in fig. 1, another preferred embodiment of the present invention provides a photonic neural network of multi-dimensional multiplexing technology, which can perform parallel computation and increase the operation speed of the photonic neural network, and comprises a laser, a multi-dimensional multiplexing module, a neural network module, and a photodetector, wherein the laser is used for generating and transmitting ultrashort optical pulses to the multi-dimensional multiplexing module,
the multi-dimensional multiplexing module comprises a plurality of multiplexers, a multi-core few-mode optical fiber and a plurality of demultiplexers, wherein the multiplexers are used for realizing multiplexing of wavelength, mode and fiber cores, the multi-core few-mode optical fiber provides two parallel operation dimensions of the mode and the fiber cores, the scale of the photon neural network is enlarged, meanwhile, optical pulses are spread by dispersion, and the demultiplexers are used for demultiplexing signals;
the neural network module comprises a neural network layer, a photoelectric modulator, a saturable absorber, an intensity modulator and a nonlinear activation unit, wherein the photoelectric modulator is used for respectively modulating signals input by the multidimensional multiplexing module by X and W, wherein X represents input data of the photonic neural network, W represents a weight matrix of the neural network, the intensity modulator and the nonlinear activation unit are arranged in the neural network layer and used for carrying out matrix multiplication to obtain output Zi＝Wi·XiThe nonlinear activation unit operates the same to obtain the output X of the neural network layer(i+1)＝fNL(Wi·Xi) The saturable absorber provides a nonlinear activation function for the nonlinear activation unit, and outputs a classification result through iterative operation of a plurality of neural network layers.
The above description is intended to describe in detail the preferred embodiments of the present invention, but the embodiments are not intended to limit the scope of the claims of the present invention, and all equivalent changes and modifications made within the technical spirit of the present invention should fall within the scope of the claims of the present invention.
Claims (10)
1. A method for realizing a photon neural network of a multidimensional multiplexing technology is characterized by comprising the following steps: the laser generates pulse light, the pulse light is input into the multi-dimensional multiplexing module, after multiplexing of a mode, a fiber core and a wavelength, the pulse light is broadened in time due to a dispersion effect, and then the pulse light is transmitted into the photonic neural network through the demultiplexer, the photonic neural network firstly completes multiplication operation of a matrix and then completes summation operation of the neural network, and the light after operation is finally received by the photodiode.
2. The method of claim 1, wherein a light source operating in the 980nm band is used as a pump to generate ultrashort light pulses through passive mode locking, so that each optoelectronic device can easily receive the generated light.
3. The method of claim 2, wherein the research two-dimensional material MoSe 2-based saturable absorber generates intensity-dependent loss, thereby generating narrow pulses with pulse widths on the order of picoseconds and femtoseconds.
4. The method as claimed in claim 3, wherein the photonic neural network is implemented by using a multi-core few-mode fiber to transmit the multiplexed light in parallel, and providing two spatial dimensions of mode and fiber core for the photonic neural network, and the dispersion effect in the fiber can broaden the light pulse in time.
5. The method of claim 1, wherein the matrix multiplication operation of the photonic neural network is performed by two waveform generators and two intensity modulators, the waveform generators are used to generate arbitrary sequences of electrical signals, and the generated electrical signals are modulated onto the optical pulses by the intensity modulators.
6. The method of claim 5, wherein the nonlinear operation of the neural network is performed by a nonlinear activation unit, and the nonlinear activation function in the nonlinear activation unit is provided by a saturable absorber having an intensity-dependent loss characteristic.
7. The method of claim 6, wherein the light passing through the demultiplexer is modulated by X and W through two electro-optical modulators, wherein X represents the input data of the photonic neural network and W represents the weight matrix of the neural network.
8. The method of claim 7, wherein the intensity modulator in the photonic neural network performs matrix multiplication to obtain the output Zi＝Wi·XiThe modulated light passes through an optical nonlinear activation unit to obtain the output X of the ith layer(i+1)＝fNL(Wi·Xi) Wherein f isNL(. cndot.) represents the operation of an optically nonlinear activation function.
9. The method of claim 8, wherein the neural network layers of the photonic neural network are connected by a loop ring, the output of the i-th layer is used as the input of the i + 1-th layer through the loop ring, and after all layer operations are completed, the optical signal is received by the photodetector and used for the next network forward operation after the gradient is updated.
10. A photonic neural network of multidimensional multiplexing technology, based on the realization method of the photonic neural network of multidimensional multiplexing technology of any one of claims 1 to 9, is characterized by comprising a laser, a multidimensional multiplexing module, a neural network module and a photodetector, wherein the laser is used for generating and transmitting ultrashort optical pulses to the multidimensional multiplexing module,
the multi-dimensional multiplexing module comprises a plurality of multiplexers, a multi-core few-mode optical fiber and a plurality of demultiplexers, wherein the multiplexers are used for realizing multiplexing of wavelength, mode and fiber cores, the multi-core few-mode optical fiber provides two parallel operation dimensions of the mode and the fiber cores, the scale of the photon neural network is enlarged, meanwhile, optical pulses are spread by dispersion, and the demultiplexers are used for demultiplexing signals;
the neural network module comprises a neural network layer, a photoelectric modulator, a saturable absorber, an intensity modulator and a nonlinear activation unit, wherein the photoelectric modulator is used for respectively modulating signals input by the multidimensional multiplexing module by X and W, wherein X represents input data of the photonic neural network, W represents a weight matrix of the neural network, the intensity modulator and the nonlinear activation unit are arranged in the neural network layer and used for carrying out matrix multiplication to obtain output Zi＝Wi·XiThe nonlinear activation unit operates the same to obtain the output X of the neural network layer(i+1)＝fNL(Wi·Xi) The saturable absorber provides a nonlinear activation function for the nonlinear activation unit, and outputs a classification result through iterative operation of a plurality of neural network layers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210016124.3A CN114358270A (en) | 2022-01-07 | 2022-01-07 | Photon neural network of multidimensional multiplexing technology and implementation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210016124.3A CN114358270A (en) | 2022-01-07 | 2022-01-07 | Photon neural network of multidimensional multiplexing technology and implementation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114358270A true CN114358270A (en) | 2022-04-15 |
Family
ID=81106460
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210016124.3A Pending CN114358270A (en) | 2022-01-07 | 2022-01-07 | Photon neural network of multidimensional multiplexing technology and implementation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114358270A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116957031A (en) * | 2023-07-24 | 2023-10-27 | 浙江大学 | Photoelectric computer based on optical multi-neuron activation function module |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0395060A2 (en) * | 1989-04-28 | 1990-10-31 | Nippon Telegraph and Telephone Corporation | Optical receiver |
| US5099114A (en) * | 1989-04-28 | 1992-03-24 | Nippon Telegraph & Telephone Corporation | Optical wavelength demultiplexer |
| US5321639A (en) * | 1992-03-02 | 1994-06-14 | Regents Of The University Of California | Dual-scale topology optoelectronic matrix algebraic processing system |
| US20190370644A1 (en) * | 2018-06-04 | 2019-12-05 | Lightmatter, Inc. | Convolutional layers for neural networks using programmable nanophotonics |
| WO2020020991A1 (en) * | 2018-07-25 | 2020-01-30 | Vrije Universiteit Brussel | Space division multiplexing method and system using speckle pattern recognition in multi-mode optical fibres |
| CN113642718A (en) * | 2021-09-01 | 2021-11-12 | 哈尔滨工程大学 | Optical fiber pulse neuron construction scheme |
| US20210357737A1 (en) * | 2018-11-12 | 2021-11-18 | Ryan HAMERLY | Large-Scale Artificial Neural-Network Accelerators Based on Coherent Detection and Optical Data Fan-Out |
| WO2022001002A1 (en) * | 2020-07-01 | 2022-01-06 | 浙江大学 | Photonic neural network on silicon substrate based on tunable filter, and modulation method therefor |
-
2022
- 2022-01-07 CN CN202210016124.3A patent/CN114358270A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0395060A2 (en) * | 1989-04-28 | 1990-10-31 | Nippon Telegraph and Telephone Corporation | Optical receiver |
| US5099114A (en) * | 1989-04-28 | 1992-03-24 | Nippon Telegraph & Telephone Corporation | Optical wavelength demultiplexer |
| US5321639A (en) * | 1992-03-02 | 1994-06-14 | Regents Of The University Of California | Dual-scale topology optoelectronic matrix algebraic processing system |
| US20190370644A1 (en) * | 2018-06-04 | 2019-12-05 | Lightmatter, Inc. | Convolutional layers for neural networks using programmable nanophotonics |
| WO2020020991A1 (en) * | 2018-07-25 | 2020-01-30 | Vrije Universiteit Brussel | Space division multiplexing method and system using speckle pattern recognition in multi-mode optical fibres |
| US20210357737A1 (en) * | 2018-11-12 | 2021-11-18 | Ryan HAMERLY | Large-Scale Artificial Neural-Network Accelerators Based on Coherent Detection and Optical Data Fan-Out |
| WO2022001002A1 (en) * | 2020-07-01 | 2022-01-06 | 浙江大学 | Photonic neural network on silicon substrate based on tunable filter, and modulation method therefor |
| CN113642718A (en) * | 2021-09-01 | 2021-11-12 | 哈尔滨工程大学 | Optical fiber pulse neuron construction scheme |
Non-Patent Citations (3)
| Title |
|---|
| PORTE XAVIER, ET AL: "A complete, parallel and autonomous photonic neural network in a semiconductor multimode laser", JOURNAL OF PHYSICS, 29 April 2021 (2021-04-29), pages 1 - 8 * |
| XU J, ET AL: "Opto-Electronic Neural Networks Based on Few-Mode Fiber", 2021 19TH INTERNATIONAL CONFERENCE ON OPTICAL COMMUNICATIONS AND NETWORKS (ICOCN), 19 October 2021 (2021-10-19), pages 1 - 5 * |
| 李强: "光学神经元及学习机制研究", 北京交通大学, 15 January 2020 (2020-01-15), pages 1 - 121 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116957031A (en) * | 2023-07-24 | 2023-10-27 | 浙江大学 | Photoelectric computer based on optical multi-neuron activation function module |
| CN116957031B (en) * | 2023-07-24 | 2024-05-24 | 浙江大学 | Photoelectric computer based on optical multi-neuron activation function module |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103678258B (en) | Method for improving data resolution ratio of silica-based optical matrix processor | |
| US20210099236A1 (en) | Quantum communication system having time to frequency conversion and associated methods | |
| CN114819132B (en) | Photon two-dimensional convolution acceleration method and system based on time-wavelength interleaving | |
| CN114358271A (en) | Time-wavelength interweaving photon neural network convolution acceleration chip | |
| CN115169542B (en) | Two-dimensional photon convolution acceleration system and device for convolution neural network | |
| US12001947B2 (en) | Ultra-wide data band optical processor | |
| JPH08507389A (en) | Optically encoded signal | |
| CN116432726B (en) | Photoelectric hybrid deep neural network operation device and operation method | |
| CN115222035B (en) | Photon neural network convolution acceleration chip | |
| CN114358270A (en) | Photon neural network of multidimensional multiplexing technology and implementation method | |
| CN110830249B (en) | Space division multiplexing continuous variable quantum communication encryption system and implementation method | |
| US11934943B1 (en) | Two-dimensional photonic neural network convolutional acceleration chip based on series connection structure | |
| Meng et al. | On-demand reconfigurable incoherent optical matrix operator for real-time video image display | |
| CN115130666B (en) | Two-dimensional photon convolution acceleration method and system | |
| CN104155721B (en) | Optical Sampling system based on quantum dot mode-locked laser | |
| Rabenandrasana et al. | Development of a metrological system for measuring the characteristics of single photon detectors based on an educational platform EMQOS 1.0 | |
| CN116316007A (en) | Ultra-high-speed arbitrary waveform generator and generation method based on synthesized dimension | |
| CN108011282A (en) | High speed real-time oscilloscope and its sample quantization method based on optical event stretching | |
| Qinggui et al. | PIN photodiode array for free-space optical communication | |
| Le et al. | All‐optical time‐domain demultiplexing of 170.8 Gbit/s signal in chalcogenide GeAsSe microstructured fibre | |
| Hui et al. | A new scheme to implement the reconfigurable optical logic gate in Millimeter Wave over fiber system | |
| Burgos et al. | Challenges in the path toward a scalable silicon photonics implementation of deep neural networks | |
| Yamamoto et al. | Monolithically integrated quantum dot optical modulator with Semiconductor optical amplifier for short-range optical communications | |
| CN110231746A (en) | The photon A/D conversion system and method compared based on full light | |
| Zhang et al. | Time-stretch optical neural network with time-division multiplexing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20230823 Address after: 230000, 3rd Floor, Building D4, Innovation Industrial Park, No. 800 Wangjiang West Road, High tech Zone, Hefei City, Anhui Province Applicant after: Hefei Photon Computing Intelligent Technology Co.,Ltd. Address before: Room 321-17, building 6-b, international enterprise R & D Park, No. 75, Tiansheng Road, Jiangbei new area, Nanjing, Jiangsu 210000 Applicant before: Nanjing Dingxin Photoelectric Technology Co.,Ltd. |