WO2021074656A1 - Infroton type artificial neural network - Google Patents

Infroton type artificial neural network Download PDF

Info

Publication number
WO2021074656A1
WO2021074656A1 PCT/HU2020/000031 HU2020000031W WO2021074656A1 WO 2021074656 A1 WO2021074656 A1 WO 2021074656A1 HU 2020000031 W HU2020000031 W HU 2020000031W WO 2021074656 A1 WO2021074656 A1 WO 2021074656A1
Authority
WO
WIPO (PCT)
Prior art keywords
infroton
type
light
conical optical
type conical
Prior art date
Application number
PCT/HU2020/000031
Other languages
French (fr)
Inventor
János Dobos
Jozsef Dobos
Original Assignee
Dobos Janos
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Dobos Janos filed Critical Dobos Janos
Publication of WO2021074656A1 publication Critical patent/WO2021074656A1/en

Links

Classifications

    • G—PHYSICS
    • G02—OPTICS
    • G02F—OPTICAL DEVICES OR ARRANGEMENTS FOR THE CONTROL OF LIGHT BY MODIFICATION OF THE OPTICAL PROPERTIES OF THE MEDIA OF THE ELEMENTS INVOLVED THEREIN; NON-LINEAR OPTICS; FREQUENCY-CHANGING OF LIGHT; OPTICAL LOGIC ELEMENTS; OPTICAL ANALOGUE/DIGITAL CONVERTERS
    • G02F3/00—Optical logic elements; Optical bistable devices
    • G02F3/02—Optical bistable devices
    • G02F3/028—Optical bistable devices based on self electro-optic effect devices [SEED]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/067—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06E—OPTICAL COMPUTING DEVICES; COMPUTING DEVICES USING OTHER RADIATIONS WITH SIMILAR PROPERTIES
    • G06E3/00—Devices not provided for in group G06E1/00, e.g. for processing analogue or hybrid data
    • G06E3/001—Analogue devices in which mathematical operations are carried out with the aid of optical or electro-optical elements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks

Definitions

  • the technical field of the invention pertains to conical optical fibers, light waveguides, light collector fibers, light concentrators, light traps, photovoltaic cells, multiplexers, demultiplexers, optical logic gates, neural networks, photonic processors that operate in whispering gallery mode.
  • Optical networks usually use a cylindrical optical fiber to which light is introduced at the end that is perpendicular to its axis, and in which light propagates in accordance with the law of total reflection to the other end of the fiber.
  • the conical optical fibers used in photonic networks function in a similar way with the difference that in them, light propagation is unlimited only toward the thicker end of the cone, while it is not unlimited in the opposite direction because the narrowing segment of the cone, due to the continuous change of the angle of light reflection, reverses the light.
  • the inventors' primary starting point was the recognition that the light can also be introduced through the arched, external periphery of the fiber, if a suitable projecting light receiving surface formed on it, and in this case, the incoming light circulates in whispering gallery mode in the conical optical fiber while in axial direction toward the thicker end of the cone in accordance with the law of light reflection.
  • optical fibers of such type can be used to build light guides, light collectors, light concentrators, light traps, photovoltaic cells, multiplexers, demultiplexers, optical logic gates, neural networks, and photonic processors.
  • the US 20140166078A1 patent named "Light concentrator and a photovoltaic cell”, by Toshiba describes a thin light collector, light guide, light concentrator plate whose layers have different refractive indexes, and the light collected onto its surface is guided onto the photovoltaic cells placed on the side plates.
  • This light collector, light guide, light concentrator is an excellent solution, however it is the opinion of the inventors that a further development in accordance with this invention is necessary, namely, the further concentration and trapping of light is required to facilitate the use of even smaller photovoltaic cells hence increasing the efficiency.
  • Wavelength demultiplexer by Intel Corporation includes a wavelength division multiplexer and a demultiplexer optical solution, in which multiplexing and demultiplexing is performed by a semicircular diffraction grate (echelle grating) in accordance with the wavelength, and the optical units are placed one after the other linearly in a row, in two dimensions.
  • a semicircular diffraction grate echelle grating
  • the photon synapse consists of a cone-shaped waveguide with discrete islands of phase-change material (PCM) from the top optically connecting the presynaptic (preneuronal) and postsynaptic (postneuronal) signals.
  • PCM phase-change material
  • the use of purely optical means provides ultrafast operation speed, virtually unlimited bandwidth and no loss of electrical power on interconnects. It is significant that the synaptic weight can be randomly installed simply by changing the number of optical pulses that create a system with continuously changing synaptic plasticity, reflecting the true analog nature of the biological synapses.
  • the embodiments of the three dimensional, whispering gallery mode, INFROTON type conical optical fiber according to this invention, that is surrounded by the various embodiments of the INFROTON type conical optical spiral fiber is suitable for facilitating the further development of the above solutions, on the one hand, for decreasing the size, since, for example, a photonic neural node, or an "Or" type logic gate occupies an area of merely an area of merely 10 x 10/1 micromillimeter, so in theory, a 40 x 40 mm x 25 mm of photonic array can accommodate up to one hundred billion neurons and on the other hand, due to its small size, for decreasing the energy consumption.
  • This invention can be produced using traditional chip manufacturing (UV irradiation, chemical etching, laser fabrication) using single and double sided processing.
  • FIG 1. shows an embodiment of INFROTON type conical optical fiber, with a projecting light input surface, laid on its arched external periphery which has no light output surface and which, hence, is also a light guide, light collector, light concentrator and light trap, and it also produces electricity.
  • FIG. 2. shows two embodiment of element of INFROTON type conical optical fiber with phase change materials (Ge2Sb2Te5) embedded ring resonators, using waste light with produces electricity.
  • phase change materials Ga2Sb2Te5
  • FIG 3. shows a multiplexer where an embodiment of the INFROTRON type conical optical fiber and INFROTRON type conical optical spiral fiber is seen that is able to mix a combined light from the three different wavelength entering light and then to forward it to an embodiment of the INFROTON type conical optical fiber what has a light output surface where the combined light leaves the multiplexer.
  • FIG.4. shows three serially connected "OR" logic gates that can be formed into a photonic plate by double sided UV irradiation, chemical etching technology, and whose main elements are the INFROTON type conical optical fiber with input light and output light surfaces, the tunable IFROTON type conical optical spiral fiber with optical nonlinear elements.
  • FIG. 5 shows an array of plurality logic gates, where the logic gates, are connected by a projecting light output surface of INFROTON type conical optical spiral fiber and a projecting input surface of INFROTON type conical optical fiber, where the light propagates in all cases in whispering gallery mode.
  • FIG. 6. shows the time delayed mode of a photonic neuron node.
  • FIG. 7. shows the photonic neuron node with resonators.
  • FIG. 8. shows the photonic neuron node with echelle gratings.
  • FIG. 9. shows the logical structure and the implemented physical architecture of the photonic neuron node and photonic neuron node cluster.
  • FIG. 10 shows an INFROTON type artificial neural network, which is an array of four hidden layers of vertically and horizontally placed photonic neuron nodes connected with cross points to form a three-dimensional matrix.
  • FIG. 11. shows how to connect the INFROTON type artificial neural networks in four directions of the plane.
  • FIG.12. shows the logical structure and the perspective view of one embodiment of INFROTON type artificial neural networks.
  • Figure 1 shows an embodiment of 100 INFROTON type conical optical fiber laid on the 101 arched, external periphery that is also a light guide, light collector, light concentrator and light trap, and generates electricity through a 105 photovoltaic cell.
  • the 102 incoming light enters through the 103 projecting light input surface formed on the 101 arched, external periphery bordered by 107 light reflective walls into the 100 INFROTON type conical optical fiber, then it starts moving in whispering gallery mode in axial direction toward the thicker end of the 100 INFROTON type conical optical fiber.
  • the light that first starts moving toward the thinner end of the 100 INFROTON type conical optical fiber and circulating in whispering gallery mode in axial direction is then reversed due to the change in the angle of total light reflection, and then also starts moving toward the thicker end of the 100 INFROTON type conical optical fiber.
  • the 100 INFROTON type conical optical fiber can have light reflective surface, as well. Thus, it becomes evident for the professionals of the field that all collected 102 incoming light, due to the light guiding in accordance with the invention, is concentrated and trapped at the thicker end of the 100 INFROTON type conical optical fiber, and it circulates there in whispering gallery mode.
  • the 100 INFROTON type conical optical fiber can be bordered by 107 light reflective walls whose material may be air gap, mirror, or an electric conductor mirror surface.
  • FIG. 2. shows two embodiment of element of 100 INFROTON type conical optical fiber, with 110 waveguide, with 124 phase change materials (Ge2Sb2Te5) embedded 122 ring resonator, using 137 waste light generates electricity through a 105 photovoltaic cell.
  • 124 phase change materials Ga2Sb2Te5
  • Introducing a 124 phase change materials element on top of the 122 ring resonator waveguide allows us to control 121 various wavelength input light signal propagation through the ports by merely changing the state of the 124 phase change materials element.
  • the 123 weighted wavelength light signals passing through the 122 ring resonator waveguide get evanescently coupled to the 124 phase change materials element and gets differentially absorbed by the 124 phase change materials in its low-loss amorphous state and high-absortion crystalline state.
  • a 105 photovoltaic cell is placed in the path of the 137 waste light the light trap formed at the thicker end of the 100 INFROTON type conical optical fiber, electric current can be produced.
  • Figure 3 shows a 200 multiplexer where an embodiment of the 109 INFROTON type conical optical spiral, with 110 waveguide, that is able to mix a 108 combined light from the three different wavelength, 102 incoming light in its 101 arched, external periphery, and then to forward them to the 103 projecting light input surfaces from where it starts propagating in 100 INFROTON type conical optical fiber in 104 whispering gallery mode in axial direction toward the 106 projecting light output surfaces located at the thicker end of the 108 combined light cone where, finally, it leaves the 200 multiplexer.
  • Figure 4. shows serially connected 300 logic gates, which are "OR" type logic gates which can be produced into an 111 photonic plate by using double sided UV irradiation, chemical etching technology, and one of the main elements of which is an embodiment of the 100 INFROTON type conical optical fiber that has a 103 projecting light input surface at its thinner end two 106 projecting light output surfaces at its thicker end. While its other main element is an embodiment of the 109 INFROTON type conical optical spiral fiber that that has four 103 projecting light input surfaces and a 106 projecting light output surfaces where the 110 waveguides are separated by 107 light reflective walls.
  • the 115 CW1 laser impulse and 116 CW2 laser impulse operated optical switches, 112 optical nonlinear elements are built into the 110 waveguide, which, depending on their refractive indexes support only the transmittance of certain wavelength lights and ban the transmittance of other wavelength lights.
  • FIG. 5. shows the 300 logic gates are connected by the 106 projecting light output surface of 100 INFROTON type conical optical fiber and the 103 projecting light input surface of 109 INFROTON type conical optical spiral fiber, where the light propagates in all cases in 104 whispering gallery mode and in axial direction, toward te thicker end of the 100 INFROTON type conical optical fiber, that can be produced into the two sides of the 111 photonic plate by UV irradiation, chemical etching.
  • FIG. 6. shows the continuous time delayed mode of a 500 photonic neuron node with the discrete islands of 124 phase change materials (Ge2Sb2Te5) embedded 109 INFROTON type conical optical spiral fiber, and with the 124 phase change materials (Ge2Sb2Te5) embedded 100 INFROTON type conical optical fiber.
  • the 500 photonic neuron node retained the biological concept of artificial neurons, the 121 various wavelength input light signals is assigned a weight that represents its relative importance, and 123 weighted wavelength light signals combine the input with their internal state (activation) and an optional threshold using an activation function, and produce output 136 spike signal using, an output function with the 124 phase change materials (Ge2Sb2Te5) embedded 100 INFROTON type conical optical fiber.
  • the 121 various wavelength input light signals is assigned a weight that represents its relative importance
  • 123 weighted wavelength light signals combine the input with their internal state (activation) and an optional threshold using an activation function, and produce output 136 spike signal using, an output function with the 124 phase change materials (Ge2Sb2Te5) embedded 100 INFROTON type conical optical fiber.
  • Weighting operation is based 124 phase-change materials, which can modify the propagating optical mode in a controlled manner. If the integrated power of the 123 weighted wavelength light signals surpasses a certain threshold, the 124 phase-change foil on the 100 INFROTON type conical optical fiber, the thicker end of which acts as a resonator switches and an output pulse 136 spike signal is generated.
  • the 123 weighted wavelength light signals passing through the 110 waveguide of 500 photonic neuron node get evanescently coupled to the 124 phase change materials element and gets differentially absorbed by the 124 phase change materials in its low-loss amorphous state and high-absortion crystalline state. It is significant that the synaptic weight 123 weighted wavelength light signals can be randomly installed simply by changing the number of optical pulses that create a system with continuously changing synaptic plasticity, reflecting the true analog nature of the biological synapses.
  • the 100 INFROTON type conical optical fiber in whispering gallery mode work, and obey the properties behind constructive interference and total internal reflection.
  • the through a 128 light splitter the 123 weighted wavelength light can be split into multiple (sub) rays.
  • the 128 light splitter has a 129 start node and a 130 destination node.
  • the 123 weighted wavelength light enters through the start node and traverses the 127 optical waveguide of different length and different refractive index until it reaches the destination.
  • the FIG.6. shows, in case of three 129 start nodes we expect fluctuations in the eight continuous time delayed intensity of the signal.
  • the continuous time delayed mode 500 photonic neuron node makes time-shifted copies of 123 weighted wavelength light signals, thus, it continuously emits 136 spike signal.
  • the 500 photonic neural node is less sensitive to 121 various wavelength input light signals changes, because time-shifted 123 weighted wavelength light signals continuous give almost the same 136 spike signal, so 500 photonic neural node can generalize, so it can be used to build a shift invariant neural network.
  • the 109 INFROTON type conical optical spiral has a 132 conical waveguide which reverses the direction of light, so which reverses 121 various wavelength input light signals, or 136 spike signal, or 123 weighted wavelength light signals, if they do not follow the operating direction of the 700 INFROTON type artificial neural network.
  • FIG. 7. shows the 500 photonic neuron node with the 124 phase change materials (Ge2Sb2Te5) embedded 122 resonators, and with the 124 phase change materials (Ge2Sb2Te5) embedded 100 INFROTON type conical optical fiber.
  • Weighting operation is based around the 124 phase change materials embedded 122 resonator, which as previously written can modify the propagating optical mode in a controlled manner.
  • the 124 phase change materials embedded 122 resonator perform both linear and nonlinear transformations for the 121 various wavelength input light.
  • the first step is to the resonator selects the 121 various wavelength input light according to the wavelength, then the 124 phase change materials will perform weighting operation, then the 122 resonator transfer 123 weighted wavelength light signals to multiplexing, it to the 124 phase change materials embedded 100 INFROTON type conical optical fiber, where after the threshold is exceeded generates the 136 spike signal.
  • Optical 122 resonators work on the principles behind total internal reflection, constructive interference, and optical coupling, functions as a filter, as switce.
  • FIG. 8. shows the a 500 photonic neuron node with 126 echelle gratings with the discrete islands of 124 phase change materials embedded 100 INFROTON type conical optical spiral fiber, and with the 124 phase change materials embedded 100 INFROTON type conical optical fiber.
  • the 110 waveguides were bordered by 107 light reflective walls.
  • Weighting operation continue discrete islands of 124 phase- change materials, which can modify the propagating optical mode in a controlled manner.
  • the 500 photonic neural node has a switch with 120 tunable threshold value, which allows 123 weighted wavelength light signals to pass when it exceeds the threshold value.
  • the 136 spike signal generation is as described previously.
  • the 109 INFROTON type conical optical spiral has a 132 conical waveguide which reverses the direction of light, so which reverses 121 various wavelength input light signals, or 136 spike signal, or 123 weighted wavelength light signals, if they do not follow the operating direction of the 700 INFROTON type artificial neural network.
  • FIG. 9. shows the logical structure and the implemented physical architecture of the 500 photonic neuron node and 600 neuron node cluster.
  • the 121 various wavelength input light signals (in the drawing Xl+X2+...Xn), multiplexed by 110 waveguide and then modulated by a 131 Mach-Zehnder interferometer, it is conveyed to the 109 INFROTON type conical optical spiral fiber, where it receives weights, and it is conveyed to the 100 INFROTON type conical optical fiber, where, when it exceeds the threshold value, a 136 spike signal is generated (in the drawing Y).
  • the 600 photonic neuron node cluster consists of four 500 photonic neuron nodes full interconnected. So, they also receive from themselves 136 spike signal (information).
  • FIG. 10. shows an 700 INFROTON type artificial neural network, which is an 139 array of four 134 hidden layers of vertically and horizontally placed 500 photonic neuron nodes connected with 138 cross points to form a three-dimensional matrix.
  • the 500 photonic neuron nodes receive 136 spikes from elsewhere in the network. When received 136 spikes signal accumulate for a certain period of time and reach a set threshold, the 500 photonic neuron node will fire off its own 136 spikes signal to its connected another 500 photonic neuron node.
  • the short and long-term memory of the 700 INFROTON type artificial neural network i.e. the learning can be ensured in two ways: on the one hand, by the fact that the weighted 121 various wavelength input light signals and 136 spikes signal in the 700 INFROTON type artificial neural network circulate, thus the weights are changed in its favour and the same 123 weighted wavelengths signal and 136 spikes is generated at the next operation, and on the other hand, by using, similarly to other solutions, advantageous phase 124 change materials.
  • the 700 INFROTON type artifical neural network very deep residual network, because through the 140 cross point, passing 136 spikes from one layer to a later layer as well as the next layer. Basically, it adds an identity to the solution, carrying the older input over and serving it freshly to a later layer.
  • One motivation for skipping over layers is to avoid the problem of vanishing gradients, is to avoid the 136 spikes, the information disappearance, by reusing activations from a previous layer until the adjacent layer leams its weights.
  • the 700 INFROTON type artifical neural network one “capsule network”, because the 500 photonic neuron nodes are connected with multiple weights instead of just one weight. This allows 500 photonic neuron nodes to transfer more information than simply which feature was detected, such as where a feature is in the picture or what colour and orientation it has.
  • lower level 500 photonic neuron nodes capsules send its input to higher level 500 photonic neuron nodes.
  • a capsule is a set of for example four 500 photonic neuron nodes 600 photonic neuron cluster that individually activate for various properties of a type of object, such as position, size and hue.
  • a cluster causes the higher capsule to output a high probability of observation that an entity is present. Higher-level capsules ignore outliers, concentrating on clusters. Routing by agreement of algorithm.
  • the 700 INFROTON type artifical neural network one long short term memory is an artificial recurrent neural network (RNN) architecture, because has feedback connections.
  • a long short-term memory the 600 photonic neuron cluster, because common LSTM unit is composed of a cell, an input gate, an output gate and a forget gate.
  • FIG. 11. shows how to connect the 700 NFROTON type artificial neural networks in four directions of the plane.
  • This design allows the 700 NFROTON type artificial neural networks to scale out to many other 700 NFROTON type artificial neural networks in the four planar directions.
  • FIG.12. shows the logical structure and the perspective view of one embodiment of 700 INFROTON type artificial neural networks.
  • This embodiment also showed that it is 500 photonic neuron node of 700 INFROTON type artifical neural network one very deep residual network, because 136 spike signals passing are from one layer to a later layer.
  • This embodiment also showed that the 133 input layer communicates to one or more 134 hidden layers, the 134 hidden layers then link to an 135 output layer.
  • the 100 INFROTON type conical optical fiber according to the invention is a low cost, excellent light guide, light collector, light concentrator and light trap, and can generate electricity.
  • Figure 3 shows a 200 multiplexer produced by femtosecond laser processing from borosilicate glass.
  • the experimental sample piece was able to mix a 108 combined light from the three different wavelength 102 incoming light.
  • the 120 tunable threshold value optical switch emitted the 136 spike output signal only after the threshold value was exceeded.
  • 700 NFROTON type artificial neural networks can already solve simple image recognition tasks.
  • more complex images can be processed and more difficult tasks, such as letter (or digit) recognition or language identification can be solved using the same basic approach.
  • the 700 INFROTON type artifical neural network updates its weights on its own and in this way adapts to a certain pattern over time, without the need for an external supervisor. If an 121 various wavelength input light signals arrives just before an output 136 spike signal was generated, that 121 various wavelength input light signals is to have contrib-uted to reaching the firing threshold and the corresponding weight will be increased.
  • the synaptic weight will be decreased.
  • the 700 INFROTON type artifical neural network adapts to it over time, until finally the neuron has learned this pattern without any inter-vention from an external supervisor.
  • the 700 INFROTON type artifical neural network promises access to high speed and high bandwidth inherent to optical systems, thus enabling the direct processing of optical telecommunication and visual data in the 700 INFROTON type artifical neural network.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Nonlinear Science (AREA)
  • Optics & Photonics (AREA)
  • Neurology (AREA)
  • Optical Modulation, Optical Deflection, Nonlinear Optics, Optical Demodulation, Optical Logic Elements (AREA)

Abstract

The invention is an INFROTON type artifical neural network (700), that is comprised of two main elements operating in whispering gallery mode, the INFROTON type conical optical fiber (100) and the INFROTON type conical optical spiral fiber (109), into which elements, light enters through the arched, external perifery (101) trhough a projecting light receiving surface (103). Using the various embodiments of the INFROTON type conical optical fiber (100) and the INFROTON type conical optical spiral fiber (109) light guides, light collectors, light concentrators, light traps, photovoltaic cells, multiplexers, demultiplexers, optical logic gates, neural networks, and photonic processors can be built.

Description

Name of invention: INFROTON TYPE ARTIFICIAL NEURAL NETWORK
DESCRIPTION
FIELD OF THE INVENTION
The technical field of the invention pertains to conical optical fibers, light waveguides, light collector fibers, light concentrators, light traps, photovoltaic cells, multiplexers, demultiplexers, optical logic gates, neural networks, photonic processors that operate in whispering gallery mode.
BACKGROUND OF THE INVENTION
Optical networks usually use a cylindrical optical fiber to which light is introduced at the end that is perpendicular to its axis, and in which light propagates in accordance with the law of total reflection to the other end of the fiber. The conical optical fibers used in photonic networks function in a similar way with the difference that in them, light propagation is unlimited only toward the thicker end of the cone, while it is not unlimited in the opposite direction because the narrowing segment of the cone, due to the continuous change of the angle of light reflection, reverses the light.
The inventors' primary starting point was the recognition that the light can also be introduced through the arched, external periphery of the fiber, if a suitable projecting light receiving surface formed on it, and in this case, the incoming light circulates in whispering gallery mode in the conical optical fiber while in axial direction toward the thicker end of the cone in accordance with the law of light reflection. Secondarily, they concluded that optical fibers of such type (called INFROTON type optical fibers by the inventors) can be used to build light guides, light collectors, light concentrators, light traps, photovoltaic cells, multiplexers, demultiplexers, optical logic gates, neural networks, and photonic processors.
The US 20140166078A1 patent, named "Light concentrator and a photovoltaic cell", by Toshiba describes a thin light collector, light guide, light concentrator plate whose layers have different refractive indexes, and the light collected onto its surface is guided onto the photovoltaic cells placed on the side plates. This light collector, light guide, light concentrator is an excellent solution, however it is the opinion of the inventors that a further development in accordance with this invention is necessary, namely, the further concentration and trapping of light is required to facilitate the use of even smaller photovoltaic cells hence increasing the efficiency.
Number US 20190158209 patent, named "wavelength demultiplexer" by Intel Corporation includes a wavelength division multiplexer and a demultiplexer optical solution, in which multiplexing and demultiplexing is performed by a semicircular diffraction grate (echelle grating) in accordance with the wavelength, and the optical units are placed one after the other linearly in a row, in two dimensions.
Number US20100290749A1 patent named "all-optical logic gates using nonlinear elements" by Covey Tech co. describes optical logic gates that generate binary logic level optical output signal using nonlinear elements placed linearly in a row, in two dimensions. Number WO 2016/182537 A1 patent named “optical logic gates” by Hewlett Packard Co. describes logic gates that include such optical elements and photon crystal resonators that have such dedicated continuous wave input that regulates the refractive index of the system adjusted to a switching threshold. The optical elements are placed linearly in a row in this system, as well.
Number WO 2018/213399 A1 patent by the university of Maryland, USA describes a neural, reservoir computer network that contains integrated optical tools with programmable, tunable nodes, where the optical elements are place linearly in a row, in two dimensions.
A photonic chip containing 70 photon synapses was demonstrated in 2017 by a team from the universities of Oxford, Munster and Exeter. The recording, erasure and reading of information in this case are carried out completely by optical methods. The photon synapse consists of a cone-shaped waveguide with discrete islands of phase-change material (PCM) from the top optically connecting the presynaptic (preneuronal) and postsynaptic (postneuronal) signals. The use of purely optical means provides ultrafast operation speed, virtually unlimited bandwidth and no loss of electrical power on interconnects. It is significant that the synaptic weight can be randomly installed simply by changing the number of optical pulses that create a system with continuously changing synaptic plasticity, reflecting the true analog nature of the biological synapses.
These photonic chips are excellent inventions, however, they require further development in order to facilitate the decrease of their size, and a resulting increase of their efficiency.
SUMMARY OF THE INVENTION
The embodiments of the three dimensional, whispering gallery mode, INFROTON type conical optical fiber according to this invention, that is surrounded by the various embodiments of the INFROTON type conical optical spiral fiber is suitable for facilitating the further development of the above solutions, on the one hand, for decreasing the size, since, for example, a photonic neural node, or an "Or" type logic gate occupies an area of merely an area of merely 10 x 10/1 micromillimeter, so in theory, a 40 x 40 mm x 25 mm of photonic array can accommodate up to one hundred billion neurons and on the other hand, due to its small size, for decreasing the energy consumption. This invention can be produced using traditional chip manufacturing (UV irradiation, chemical etching, laser fabrication) using single and double sided processing.
The invention can be understood on the basis of the drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG 1. shows an embodiment of INFROTON type conical optical fiber, with a projecting light input surface, laid on its arched external periphery which has no light output surface and which, hence, is also a light guide, light collector, light concentrator and light trap, and it also produces electricity. FIG. 2. shows two embodiment of element of INFROTON type conical optical fiber with phase change materials (Ge2Sb2Te5) embedded ring resonators, using waste light with produces electricity.
FIG 3. shows a multiplexer where an embodiment of the INFROTRON type conical optical fiber and INFROTRON type conical optical spiral fiber is seen that is able to mix a combined light from the three different wavelength entering light and then to forward it to an embodiment of the INFROTON type conical optical fiber what has a light output surface where the combined light leaves the multiplexer.
FIG.4. shows three serially connected "OR" logic gates that can be formed into a photonic plate by double sided UV irradiation, chemical etching technology, and whose main elements are the INFROTON type conical optical fiber with input light and output light surfaces, the tunable IFROTON type conical optical spiral fiber with optical nonlinear elements.
FIG. 5. shows an array of plurality logic gates, where the logic gates, are connected by a projecting light output surface of INFROTON type conical optical spiral fiber and a projecting input surface of INFROTON type conical optical fiber, where the light propagates in all cases in whispering gallery mode.
FIG. 6. shows the time delayed mode of a photonic neuron node.
FIG. 7. shows the photonic neuron node with resonators.
FIG. 8. shows the photonic neuron node with echelle gratings.
FIG. 9. shows the logical structure and the implemented physical architecture of the photonic neuron node and photonic neuron node cluster.
FIG. 10. shows an INFROTON type artificial neural network, which is an array of four hidden layers of vertically and horizontally placed photonic neuron nodes connected with cross points to form a three-dimensional matrix.
FIG. 11. shows how to connect the INFROTON type artificial neural networks in four directions of the plane.
FIG.12. shows the logical structure and the perspective view of one embodiment of INFROTON type artificial neural networks.
DETAILED DESCRIPTION
Figure 1. shows an embodiment of 100 INFROTON type conical optical fiber laid on the 101 arched, external periphery that is also a light guide, light collector, light concentrator and light trap, and generates electricity through a 105 photovoltaic cell. The 102 incoming light enters through the 103 projecting light input surface formed on the 101 arched, external periphery bordered by 107 light reflective walls into the 100 INFROTON type conical optical fiber, then it starts moving in whispering gallery mode in axial direction toward the thicker end of the 100 INFROTON type conical optical fiber. The light that first starts moving toward the thinner end of the 100 INFROTON type conical optical fiber and circulating in whispering gallery mode in axial direction is then reversed due to the change in the angle of total light reflection, and then also starts moving toward the thicker end of the 100 INFROTON type conical optical fiber.
The 100 INFROTON type conical optical fiber can have light reflective surface, as well. Thus, it becomes evident for the professionals of the field that all collected 102 incoming light, due to the light guiding in accordance with the invention, is concentrated and trapped at the thicker end of the 100 INFROTON type conical optical fiber, and it circulates there in whispering gallery mode.
In case a 105 photovoltaic cell is placed in the path of the light that is circulating in 104 whispering gallery mode and propagating in axial direction into the light trap formed at the thicker end of the 100 INFROTON type conical optical fiber, electric current can be produced. The 100 INFROTON type conical optical fiber can be bordered by 107 light reflective walls whose material may be air gap, mirror, or an electric conductor mirror surface.
FIG. 2. shows two embodiment of element of 100 INFROTON type conical optical fiber, with 110 waveguide, with 124 phase change materials (Ge2Sb2Te5) embedded 122 ring resonator, using 137 waste light generates electricity through a 105 photovoltaic cell. Introducing a 124 phase change materials element on top of the 122 ring resonator waveguide allows us to control 121 various wavelength input light signal propagation through the ports by merely changing the state of the 124 phase change materials element. The 123 weighted wavelength light signals passing through the 122 ring resonator waveguide get evanescently coupled to the 124 phase change materials element and gets differentially absorbed by the 124 phase change materials in its low-loss amorphous state and high-absortion crystalline state. In case a 105 photovoltaic cell is placed in the path of the 137 waste light the light trap formed at the thicker end of the 100 INFROTON type conical optical fiber, electric current can be produced.
Figure 3. shows a 200 multiplexer where an embodiment of the 109 INFROTON type conical optical spiral, with 110 waveguide, that is able to mix a 108 combined light from the three different wavelength, 102 incoming light in its 101 arched, external periphery, and then to forward them to the 103 projecting light input surfaces from where it starts propagating in 100 INFROTON type conical optical fiber in 104 whispering gallery mode in axial direction toward the 106 projecting light output surfaces located at the thicker end of the 108 combined light cone where, finally, it leaves the 200 multiplexer.
Figure 4. shows serially connected 300 logic gates, which are "OR" type logic gates which can be produced into an 111 photonic plate by using double sided UV irradiation, chemical etching technology, and one of the main elements of which is an embodiment of the 100 INFROTON type conical optical fiber that has a 103 projecting light input surface at its thinner end two 106 projecting light output surfaces at its thicker end. While its other main element is an embodiment of the 109 INFROTON type conical optical spiral fiber that that has four 103 projecting light input surfaces and a 106 projecting light output surfaces where the 110 waveguides are separated by 107 light reflective walls. The 115 CW1 laser impulse and 116 CW2 laser impulse operated optical switches, 112 optical nonlinear elements are built into the 110 waveguide, which, depending on their refractive indexes support only the transmittance of certain wavelength lights and ban the transmittance of other wavelength lights.
In case of the "OR" type logic gate seen in Figure 4., two input logic signals arrive, 113 DATA 1 and 114 DATA 2, which are guided by the 110 waveguide to the first 112 optical nonlinear element that can be switched on and off by 115 CW1 laser impulse and that operates as a "NOT" logic gate, then the signals with a new logic level is guided to the second 112 optical nonlinear element that can be switched on and off by 116 CW2 laser light and that operates as a "NOT" logic gate, which, in turn, generates a signals with a yet newer logic level. It is evident for a professional of the field that: if the amplitude of any of the two, 113 DATA 1 and 114 DATA2 input optical signals is high (that is: it has a high logic level), then there are two copied (with identical, high logic level) 117 OUTPUT 1 and 118 OUTPUT 2 output signal, and if the amplitude of both, 113 DATA 1 and 114 DATA 2 input optical signals is low (that is they have a low logic level), then there is no output signal it is also evident for the professional of the field from the structure of the 300 logic gate that any type of logic gate can be built based on the invention.
FIG. 5. shows the 300 logic gates are connected by the 106 projecting light output surface of 100 INFROTON type conical optical fiber and the 103 projecting light input surface of 109 INFROTON type conical optical spiral fiber, where the light propagates in all cases in 104 whispering gallery mode and in axial direction, toward te thicker end of the 100 INFROTON type conical optical fiber, that can be produced into the two sides of the 111 photonic plate by UV irradiation, chemical etching.
FIG. 6. shows the continuous time delayed mode of a 500 photonic neuron node with the discrete islands of 124 phase change materials (Ge2Sb2Te5) embedded 109 INFROTON type conical optical spiral fiber, and with the 124 phase change materials (Ge2Sb2Te5) embedded 100 INFROTON type conical optical fiber.
The 500 photonic neuron node retained the biological concept of artificial neurons, the 121 various wavelength input light signals is assigned a weight that represents its relative importance, and 123 weighted wavelength light signals combine the input with their internal state (activation) and an optional threshold using an activation function, and produce output 136 spike signal using, an output function with the 124 phase change materials (Ge2Sb2Te5) embedded 100 INFROTON type conical optical fiber.
Weighting operation is based 124 phase-change materials, which can modify the propagating optical mode in a controlled manner. If the integrated power of the 123 weighted wavelength light signals surpasses a certain threshold, the 124 phase-change foil on the 100 INFROTON type conical optical fiber, the thicker end of which acts as a resonator switches and an output pulse 136 spike signal is generated.
The 123 weighted wavelength light signals passing through the 110 waveguide of 500 photonic neuron node get evanescently coupled to the 124 phase change materials element and gets differentially absorbed by the 124 phase change materials in its low-loss amorphous state and high-absortion crystalline state. It is significant that the synaptic weight 123 weighted wavelength light signals can be randomly installed simply by changing the number of optical pulses that create a system with continuously changing synaptic plasticity, reflecting the true analog nature of the biological synapses.
The 100 INFROTON type conical optical fiber in whispering gallery mode work, and obey the properties behind constructive interference and total internal reflection.
The through a 128 light splitter the 123 weighted wavelength light can be split into multiple (sub) rays. The 128 light splitter has a 129 start node and a 130 destination node. The 123 weighted wavelength light enters through the start node and traverses the 127 optical waveguide of different length and different refractive index until it reaches the destination. The FIG.6. shows, in case of three 129 start nodes we expect fluctuations in the eight continuous time delayed intensity of the signal.
The continuous time delayed mode 500 photonic neuron node makes time-shifted copies of 123 weighted wavelength light signals, thus, it continuously emits 136 spike signal. One skilled in the art will recognize the 500 photonic neural node is less sensitive to 121 various wavelength input light signals changes, because time-shifted 123 weighted wavelength light signals continuous give almost the same 136 spike signal, so 500 photonic neural node can generalize, so it can be used to build a shift invariant neural network. The 109 INFROTON type conical optical spiral has a 132 conical waveguide which reverses the direction of light, so which reverses 121 various wavelength input light signals, or 136 spike signal, or 123 weighted wavelength light signals, if they do not follow the operating direction of the 700 INFROTON type artificial neural network.
FIG. 7. shows the 500 photonic neuron node with the 124 phase change materials (Ge2Sb2Te5) embedded 122 resonators, and with the 124 phase change materials (Ge2Sb2Te5) embedded 100 INFROTON type conical optical fiber.
Weighting operation is based around the 124 phase change materials embedded 122 resonator, which as previously written can modify the propagating optical mode in a controlled manner.
The 124 phase change materials embedded 122 resonator perform both linear and nonlinear transformations for the 121 various wavelength input light. In the linear operation process, the first step is to the resonator selects the 121 various wavelength input light according to the wavelength, then the 124 phase change materials will perform weighting operation, then the 122 resonator transfer 123 weighted wavelength light signals to multiplexing, it to the 124 phase change materials embedded 100 INFROTON type conical optical fiber, where after the threshold is exceeded generates the 136 spike signal.
Optical 122 resonators work on the principles behind total internal reflection, constructive interference, and optical coupling, functions as a filter, as switce.
FIG. 8. shows the a 500 photonic neuron node with 126 echelle gratings with the discrete islands of 124 phase change materials embedded 100 INFROTON type conical optical spiral fiber, and with the 124 phase change materials embedded 100 INFROTON type conical optical fiber. We use a 126 echelle grading to demultiplex the 121 various wavelength input light coming from the 110 waveguide. So, the 139 rough weighting happens first. The 110 waveguides were bordered by 107 light reflective walls.
Weighting operation continue discrete islands of 124 phase- change materials, which can modify the propagating optical mode in a controlled manner. The 500 photonic neural node has a switch with 120 tunable threshold value, which allows 123 weighted wavelength light signals to pass when it exceeds the threshold value. The 136 spike signal generation is as described previously. The 109 INFROTON type conical optical spiral has a 132 conical waveguide which reverses the direction of light, so which reverses 121 various wavelength input light signals, or 136 spike signal, or 123 weighted wavelength light signals, if they do not follow the operating direction of the 700 INFROTON type artificial neural network.
FIG. 9. shows the logical structure and the implemented physical architecture of the 500 photonic neuron node and 600 neuron node cluster.
The 121 various wavelength input light signals (in the drawing Xl+X2+...Xn), multiplexed by 110 waveguide and then modulated by a 131 Mach-Zehnder interferometer, it is conveyed to the 109 INFROTON type conical optical spiral fiber, where it receives weights, and it is conveyed to the 100 INFROTON type conical optical fiber, where, when it exceeds the threshold value, a 136 spike signal is generated (in the drawing Y).
The 600 photonic neuron node cluster, consists of four 500 photonic neuron nodes full interconnected. So, they also receive from themselves 136 spike signal (information).
FIG. 10. shows an 700 INFROTON type artificial neural network, which is an 139 array of four 134 hidden layers of vertically and horizontally placed 500 photonic neuron nodes connected with 138 cross points to form a three-dimensional matrix.
The 500 photonic neuron nodes receive 136 spikes from elsewhere in the network. When received 136 spikes signal accumulate for a certain period of time and reach a set threshold, the 500 photonic neuron node will fire off its own 136 spikes signal to its connected another 500 photonic neuron node.
The short and long-term memory of the 700 INFROTON type artificial neural network, i.e. the learning can be ensured in two ways: on the one hand, by the fact that the weighted 121 various wavelength input light signals and 136 spikes signal in the 700 INFROTON type artificial neural network circulate, thus the weights are changed in its favour and the same 123 weighted wavelengths signal and 136 spikes is generated at the next operation, and on the other hand, by using, similarly to other solutions, advantageous phase 124 change materials.
These materials preserve the data during the crystallisation process at the phase change in the dynamics of the crystallisation and re-thawing processes. In this case it is evident that the operations take place in the memory, that is inside the phase change material, therefore the calculation within the memory is realised, and the result of this calculation is forwarded by the phase change material, but it also records them in the dynamics of its crystallisation. The 700 INFROTON type artifical neural network very deep residual network, because through the 140 cross point, passing 136 spikes from one layer to a later layer as well as the next layer. Basically, it adds an identity to the solution, carrying the older input over and serving it freshly to a later layer. One motivation for skipping over layers is to avoid the problem of vanishing gradients, is to avoid the 136 spikes, the information disappearance, by reusing activations from a previous layer until the adjacent layer leams its weights.
It is obvious to one skilled, the 700 INFROTON type artifical neural network one “capsule network”, because the 500 photonic neuron nodes are connected with multiple weights instead of just one weight. This allows 500 photonic neuron nodes to transfer more information than simply which feature was detected, such as where a feature is in the picture or what colour and orientation it has. In this process of routing, lower level 500 photonic neuron nodes capsules send its input to higher level 500 photonic neuron nodes. A capsule is a set of for example four 500 photonic neuron nodes 600 photonic neuron cluster that individually activate for various properties of a type of object, such as position, size and hue. A cluster causes the higher capsule to output a high probability of observation that an entity is present. Higher-level capsules ignore outliers, concentrating on clusters. Routing by agreement of algorithm.
It is obvious to one skilled, the 700 INFROTON type artifical neural network one long short term memory (LSTM) is an artificial recurrent neural network (RNN) architecture, because has feedback connections. A long short-term memory the 600 photonic neuron cluster, because common LSTM unit is composed of a cell, an input gate, an output gate and a forget gate.
FIG. 11. shows how to connect the 700 NFROTON type artificial neural networks in four directions of the plane.
This design allows the 700 NFROTON type artificial neural networks to scale out to many other 700 NFROTON type artificial neural networks in the four planar directions.
FIG.12. shows the logical structure and the perspective view of one embodiment of 700 INFROTON type artificial neural networks.
This embodiment also showed that it is 500 photonic neuron node of 700 INFROTON type artifical neural network one very deep residual network, because 136 spike signals passing are from one layer to a later layer.
This embodiment also showed that the 133 input layer communicates to one or more 134 hidden layers, the 134 hidden layers then link to an 135 output layer.
Experiments: We produced the 100 INFROTON type conical optical fiber shown in Figure 1. from a commercially available optical fiber using pressing tool and hot shaping. We fixed the 100 INFROTON type conical optical fibers onto a type of Panasonic light collector plate that is also commercially available using optical glue. The area of the light collector plate is 32,000 mm2. We also used "Azur" type, small size (5.5 x 5.5 mm = 30.25 mm2) commercially available 105 photovoltaic cell cell that were coated with antireflective coating material.
We glued the 105 photovoltaic cell onto the thicker end of the 100 INFROTON type conical optical fiber as it is shown in Figurel. Based on the ratio of the light collector area, 32,000 m2 and the area of the photovoltaic cell (30,25 mm2), we could ensure light concentrations over 1000 times.
Because of the conical shaping, the light reflected from the 105 photovoltaic cell was trapped and moved again and again toward the 105 photovoltaic cell. We performed the measurements between 11 AM and 2 PM in August. The 105 photovoltaic cell cell continuously gave a performance of 12 and 14 W at 1000 times concentration in accordance with the manufacturing data.
The experiments provided clear proof that as proven by the simulations, the 100 INFROTON type conical optical fiber according to the invention is a low cost, excellent light guide, light collector, light concentrator and light trap, and can generate electricity.
Figure 3. shows a 200 multiplexer produced by femtosecond laser processing from borosilicate glass. The experimental sample piece was able to mix a 108 combined light from the three different wavelength 102 incoming light.
According to Figure 4., the geometry of three pieces of 300 logic gate was also cut using femtosecond laser into a borosilicate glass plate from two sided. We glued commercially available 112 optical nonlinear elements, photonic crystals, into the nests formed in the 110 light guide channels using optical glue. The three pieces of serially connected "OR" type logic gates showed excellent functioning according to the simulations.
The as shown in FIG 10, but with substantially enlarged dimensions, we made it the two pieces of 500 photonic neuron nodes, on the prototype level, also cut using femtosecond laser into a borosilicate glass plate.
In accordance with the simulation, the 120 tunable threshold value optical switch emitted the 136 spike output signal only after the threshold value was exceeded.
Using only two 500 photonic neuron nodes, 700 NFROTON type artificial neural networks can already solve simple image recognition tasks. By increasing the number of inputs per 500 photonic neuron nodes and the number of 500 photonic neuron nodes, more complex images can be processed and more difficult tasks, such as letter (or digit) recognition or language identification can be solved using the same basic approach.
In an unsupervised approach, the 700 INFROTON type artifical neural network updates its weights on its own and in this way adapts to a certain pattern over time, without the need for an external supervisor. If an 121 various wavelength input light signals arrives just before an output 136 spike signal was generated, that 121 various wavelength input light signals is to have contrib-uted to reaching the firing threshold and the corresponding weight will be increased.
If the 121 various wavelength input light pulse arrives after the output 136 spike signal occurred, the synaptic weight will be decreased.
When the input pattern is repeated, the 700 INFROTON type artifical neural network adapts to it over time, until finally the neuron has learned this pattern without any inter-vention from an external supervisor.
The experiments clearly confirmed the expected results, the dispersion of the light can be prevented by the light moving in whispering gallery mode in the arched peripheries, hence a significant decrease of the brilliance is avoided. This way, it evident for professionals of the field that there is no need for optical amplifier, the dimensions can be decreased, and as a result, the energy consumption is more efficient.
The use of purely optical means provides ultrafast operation speed, virtually unlimited bandwidth.
The vanishing gradient problems and information loss did not occur during the experiments.
The 700 INFROTON type artifical neural network promises access to high speed and high bandwidth inherent to optical systems, thus enabling the direct processing of optical telecommunication and visual data in the 700 INFROTON type artifical neural network.

Claims

1. An INFROTON type artificial neural network (700) comprising, an array (139) of pluarity layers of INFROTON type photonic neuron nodes (500), wherein the INFROTON type photonic neuron nodes (500) and the layers are interconnected by cross points (140) to form a three-dimensional matrix. where the main parts of the photonic neuron nodes (500) comprising, an INFROTON type conical optical fiber (100), that has an arched, external periphery (101), that is interrupted at least at one place by a projecting light input surface (103) starting from the thinner end of the cone, an INFROTON type conical optical spiral fiber (109) which has one or more such waveguides (110) that are coiled around and axially, in a constant or in continuously changing distance, onto the INFROTON type conical optical fiber (100), that has arched, external periphery (101), that is interrupted by, at least at one place, by a projecting light input surface (103), and, at least at one place, by a projecting light output surface (106),
2. The one embodiment of INFROTON type artificial neural network (700) of claim 1, characterised in that the INFROTON type conical optical fiber (100) and INFROTON type conical optical spiral fiber (109) comprising, resonators (122), Mach-Zehnder interferometers (131), switches with tunable threshold value (120), echelle gratings (126), phase-change materials (124), optical nonlinear elements (112) photovoltaic cells (105) light splitters, (128) optical waveguide of different length and different refractive index (127),
3. The other embodiment of INFROTON type artificial neural network (700) of claim 1, characterised in that the INFROTON type conical optical fiber (100) has an arched external periphery (101) interrupted by a projecting light input surface (103) at least at one place starting from the thinner end of the cone, and, interrupted by a projecting light output surface (106) at least at one place starting from the thicker end of the cone.
4. The a third embodiment of INFROTON type artificial neural network (700) of claim 1, characterised in that the INFROTON type conical optical fiber (100) has an arched external periphery (101) that is not interrupted by projecting light input surface (103), and not interrupted by projecting light output surface (106).
5. The INFROTON type artificial neural network (700) of claim 1. characterised in that according to the fourth embodiment, the logic gates (300) are connected by the projecting light output surface (106) of INFROTON type conical optical fiber and the projecting light input surface (103) of INFROTON type conical optical spiral fiber (109).
6. The INFROTON type conical optical spiral fiber (109) of INFROTON type artificial neural network (700) of claim 1. characterised in that it has one or more conical waveguide which reverses the direction of light (132)
7. The INFROTON type artificial neural network (700) of claim 1., 2., 3., 4., and 5., characterised in that the cone angle of INFROTON type conical optical fiber (100) and INFROTON type conical optical spiral fiber (109) may be zero.
8. The INFROTON type artificial neural network (700) of claim 1., 2., 3., 4., 5., 6., and 7., characterised in that the INFROTON type conical optical fiber (100) and INFROTON type conical optical spiral fiber (109) according to their arched, external periphery (101) may consist of such linear line segment and planar surfaces that approximate the shape of the arched, external periphery (101).
PCT/HU2020/000031 2019-10-14 2020-10-08 Infroton type artificial neural network WO2021074656A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
HUP1900359 2019-10-14
HU1900359A HUP1900359A1 (en) 2019-10-14 2019-10-14 Artificial neural network

Publications (1)

Publication Number Publication Date
WO2021074656A1 true WO2021074656A1 (en) 2021-04-22

Family

ID=89992997

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/HU2020/000031 WO2021074656A1 (en) 2019-10-14 2020-10-08 Infroton type artificial neural network

Country Status (2)

Country Link
HU (1) HUP1900359A1 (en)
WO (1) WO2021074656A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113839715A (en) * 2021-09-18 2021-12-24 苏州浪潮智能科技有限公司 Optical neural network and chip

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
MATJAI GOMIL?EK, WHISPERING GALLERY MODES (SEMINAR), November 2011 (2011-11-01) *
XING LIN, YAIR RIVENSON, NEZIH T. YARDIMCI, MUHAMMED VELI, YI LUO, MONA JARRAHI, AYDOGAN OZCAN: "All-optical machine learning using diffractive deep neural networks et Xing Lin all: All-optical machine learning using diffractive deep neural networks", SCIENCE, vol. 361, no. 6406, 7 September 2018 (2018-09-07), pages 1004 - 1008, XP055816942 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113839715A (en) * 2021-09-18 2021-12-24 苏州浪潮智能科技有限公司 Optical neural network and chip
CN113839715B (en) * 2021-09-18 2023-08-04 苏州浪潮智能科技有限公司 Optical neural network and chip

Also Published As

Publication number Publication date
HUP1900359A1 (en) 2021-06-28

Similar Documents

Publication Publication Date Title
Bai et al. Photonic multiplexing techniques for neuromorphic computing
Shastri et al. Photonics for artificial intelligence and neuromorphic computing
Xu et al. A survey of approaches for implementing optical neural networks
US10514498B2 (en) Waveguide architecture for photonic neural component
CN111683304B (en) All-optical diffraction neural network and system realized on optical waveguide and/or optical chip
El Srouji et al. Photonic and optoelectronic neuromorphic computing
Mengu et al. At the intersection of optics and deep learning: statistical inference, computing, and inverse design
US20220044100A1 (en) Parallel architectures for nanophotonic computing
US10451798B2 (en) Waveguide architecture for photonic neural component with multiplexed optical signals on inter-node waveguides
Peserico et al. Integrated photonic tensor processing unit for a matrix multiply: a review
WO2021074656A1 (en) Infroton type artificial neural network
Guo et al. StarLight: a photonic neural network accelerator featuring a hybrid mode-wavelength division multiplexing and photonic nonvolatile memory
Yuan et al. Ultra‐Compact and NonVolatile Nanophotonic Neural Networks
Xu et al. Scalable networks of neuromorphic photonic integrated circuits
Brückerhoff-Plückelmann et al. Event-driven adaptive optical neural network
CN116029343A (en) Photon pulse neural network implementation method based on MRR and phase change material
US20230306253A1 (en) 3D Photonic Neural Network
CN116027481A (en) Ultra-low power consumption photon chip and neural network architecture
Tait et al. Demonstration of a silicon photonic neural network
Wright et al. Integrated phase-change photonics: A strategy for merging communication and computing
CN113839715A (en) Optical neural network and chip
Zhou et al. 2D neuromorphic photonics
WO2021211125A1 (en) Photonic tensor core matrix vector multiplier
CN118297125A (en) Photonic neural network chip architecture
Goi et al. Perspective on photonic neuromorphic computing

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20875672

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20875672

Country of ref document: EP

Kind code of ref document: A1