CN118192010A - Silicon-based light-operated micro-ring resonator for in-situ training of optical neural network - Google Patents
Silicon-based light-operated micro-ring resonator for in-situ training of optical neural network Download PDFInfo
- Publication number
- CN118192010A CN118192010A CN202410143193.XA CN202410143193A CN118192010A CN 118192010 A CN118192010 A CN 118192010A CN 202410143193 A CN202410143193 A CN 202410143193A CN 118192010 A CN118192010 A CN 118192010A
- Authority
- CN
- China
- Prior art keywords
- optical
- resonant cavity
- waveguide
- silicon
- neural network
- 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.)
- Granted
Links
- 230000003287 optical effect Effects 0.000 title claims abstract description 84
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 44
- 229910052710 silicon Inorganic materials 0.000 title claims abstract description 39
- 239000010703 silicon Substances 0.000 title claims abstract description 39
- XUIMIQQOPSSXEZ-UHFFFAOYSA-N Silicon Chemical compound [Si] XUIMIQQOPSSXEZ-UHFFFAOYSA-N 0.000 title claims abstract description 37
- 238000012549 training Methods 0.000 title claims abstract description 29
- 238000011065 in-situ storage Methods 0.000 title claims abstract description 20
- 230000008878 coupling Effects 0.000 claims abstract description 20
- 238000010168 coupling process Methods 0.000 claims abstract description 20
- 238000005859 coupling reaction Methods 0.000 claims abstract description 20
- 230000008054 signal transmission Effects 0.000 claims abstract description 12
- 230000005540 biological transmission Effects 0.000 claims abstract description 10
- 230000004044 response Effects 0.000 claims abstract description 10
- 238000005452 bending Methods 0.000 claims abstract description 5
- 239000000758 substrate Substances 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 5
- 230000006870 function Effects 0.000 abstract description 38
- 230000004913 activation Effects 0.000 abstract description 23
- 238000004364 calculation method Methods 0.000 abstract description 8
- 238000010586 diagram Methods 0.000 description 15
- 210000002569 neuron Anatomy 0.000 description 12
- 238000004422 calculation algorithm Methods 0.000 description 9
- 238000000034 method Methods 0.000 description 9
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000010354 integration Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 3
- 238000000411 transmission spectrum Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000009022 nonlinear effect Effects 0.000 description 2
- 239000012782 phase change material Substances 0.000 description 2
- 239000002210 silicon-based material Substances 0.000 description 2
- 238000012421 spiking Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- LEVVHYCKPQWKOP-UHFFFAOYSA-N [Si].[Ge] Chemical compound [Si].[Ge] LEVVHYCKPQWKOP-UHFFFAOYSA-N 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 239000002253 acid Substances 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 150000003376 silicon Chemical class 0.000 description 1
Classifications
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B6/00—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings
- G02B6/24—Coupling light guides
- G02B6/26—Optical coupling means
- G02B6/28—Optical coupling means having data bus means, i.e. plural waveguides interconnected and providing an inherently bidirectional system by mixing and splitting signals
- G02B6/293—Optical coupling means having data bus means, i.e. plural waveguides interconnected and providing an inherently bidirectional system by mixing and splitting signals with wavelength selective means
- G02B6/29331—Optical coupling means having data bus means, i.e. plural waveguides interconnected and providing an inherently bidirectional system by mixing and splitting signals with wavelength selective means operating by evanescent wave coupling
- G02B6/29335—Evanescent coupling to a resonator cavity, i.e. between a waveguide mode and a resonant mode of the cavity
- G02B6/29338—Loop resonators
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B6/00—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings
- G02B6/10—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/0499—Feedforward networks
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B6/00—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings
- G02B6/10—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type
- G02B6/12—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type of the integrated circuit kind
- G02B2006/12083—Constructional arrangements
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B6/00—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings
- G02B6/10—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type
- G02B6/12—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type of the integrated circuit kind
- G02B2006/12083—Constructional arrangements
- G02B2006/12088—Monomode
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B6/00—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings
- G02B6/10—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type
- G02B6/12—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type of the integrated circuit kind
- G02B2006/12083—Constructional arrangements
- G02B2006/1209—Multimode
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B6/00—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings
- G02B6/10—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type
- G02B6/12—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type of the integrated circuit kind
- G02B2006/12083—Constructional arrangements
- G02B2006/12097—Ridge, rib or the like
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B6/00—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings
- G02B6/10—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type
- G02B6/12—Light guides; Structural details of arrangements comprising light guides and other optical elements, e.g. couplings of the optical waveguide type of the integrated circuit kind
- G02B2006/12133—Functions
- G02B2006/12147—Coupler
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Optics & Photonics (AREA)
- Optical Integrated Circuits (AREA)
Abstract
The invention relates to a silicon-based light-operated micro-ring resonator for in-situ training of an optical neural network, and belongs to the field of light calculation. The micro-ring resonator can realize different responses of forward/reverse transmission so as to meet the calculation of nonlinear activation functions of an optical neural network reasoning stage and loss function gradients of a training stage. The micro-ring resonator consists of 2n signal transmission waveguides, n waveguide coupling areas, a ring resonant cavity and n optical power areas, wherein n is more than or equal to 1. The signal transmission waveguide is used for realizing the input and output of forward and reverse optical signals; the waveguide coupling region is used for coupling optical signals with specific wavelengths into the ring resonant cavity; the ring resonant cavity is used for resonating an optical signal with a specific wavelength and changing the output power and the phase of the optical signal; the optical power region is used for generating optical power, and changing the bending degree of the regional waveguide, so that the resonance condition of the ring resonant cavity is changed, and different responses of forward/reverse transmission signals are realized.
Description
Technical Field
The invention belongs to the field of optical computation, and relates to a silicon-based light-operated micro-ring resonator for in-situ training of an optical neural network.
Background
With the increase of the complexity of processing tasks, the amount of computation in the artificial neural network is also increasing. However, the integrated circuit is limited by the "electronic bottleneck", and the operation speed of a single processor (core) is very limited. While computational pressure can be relieved in a multi-processor (core) fashion, coordination among different processors (cores), signal delays, power consumption scaling, and processor size increases are problematic. Therefore, there is a need for efficient and rapid computation using new techniques. Photons are the most potential computational carriers by virtue of their ultra-low latency and highly parallel capabilities. With the continuous development of integration technology, various signal processing and operation can be realized in a microchip, and photon calculation is generated. The construction of the neural network by using an optical method is an important research branch, and is the direction with the most application value.
Artificial neural networks can be divided into a variety of structures due to their task-oriented differences. But most commonly also most widely used neural networks typically include matrix-vector multiplication linear matrix computing units, convolution computing units, nonlinear activation function units, and the like. The nonlinear activation function unit plays a very important role in realizing corresponding functions of the neural network. The method introduces nonlinear factors to neurons, can approximate any nonlinear function infinitely, and enables the artificial neural network to have perfect computing and signal processing capabilities. The architecture of the optical neural network is generally the same as that of the artificial neural network, i.e. different units of the artificial neural network are realized by using photonic devices. There are several schemes for implementing nonlinear activation function units by using silicon-based integrated optical devices, such as a racetrack micro-ring resonator scheme (Aashu Jha,Chaoran Huang,Hsuan-Tung Peng,et al.Photonic spiking neural networks and graphene-on-silicon spiking neurons.Journal of Lightwave Technology,2022.), based on mixed integrated graphene-silicon materials, a racetrack micro-ring resonator scheme (Bo Wu,Hengkang Li,Weiyu Tong,et al.Low-threshold all-optical nonlinear activation function based on a Ge/Si hybrid structure in a microring resonator.Optical Materials Express,2022.), based on germanium-silicon mixed materials, a silicon-based micro-ring resonator loading phase change material scheme (Ziling Fu,Zhi Wang,Peter Bienstman,et al.Programmable low-power consumption all-optical nonlinear activation functions using a micro-ring resonator with phase-change materials.Optics Express,2022.), and the like. Therefore, the silicon-based micro-ring resonator becomes an ideal device for realizing the nonlinear activation function unit due to the small size and high nonlinear effect.
For the neural network for supervised learning, training of the neural network in advance is required in order to realize the corresponding functions. The most commonly used training method in the optical neural network is based on an error back propagation algorithm, and the weight is updated by calculating the gradient of the loss function to each weight in the network, so that the network output approximates to the target output. At present, most optical neural networks adopt offline training calculation (Jiahui Wang,Sean P.Rodrigues,Ercan M.Dede,et al."Microring-based programmable coherent optical neural networks,"Opt.Express,2023.), to extract the output of the network, corresponding gradients are calculated offline by a computer, and then the calculation result is returned to the optical neural network to update weights. Such a method undoubtedly gives up the advantages of optical fast operation, brings great delay, and as the network scale increases, the calculation amount of the computer also increases sharply, which is very unfriendly to tasks requiring frequent training. Researchers have demonstrated that the gradient of the loss function to each weight in the network can be calculated using the original optical neural network back-transmitting optical signal method, known as in-situ back-propagation algorithm (Ziyu Gu,Zicheng Huang,Yesheng Gao,et al."Training optronic convolutional neural networks on an optical system through backpropagation algorithms,"Opt.Express,2022.)., but it has been found through derivation that if it is desired to rely entirely on the original network to perform gradient calculation, a nonlinear activation function unit in the network is required to respond differently to forward and back-propagated signals. If the response on forward propagation is g (·) the response on reverse propagation should be its derivative g' (·). 2021, university of electronics Guo Xian believe in a religion teaches an optical neural network in-situ training scheme (Xianxin Guo,Thomas D.Barrett,Zhiming M.Wang,et al."Backpropagation through nonlinear units for the all-optical training of neural networks,"Photon.Res.9,2021.). based on saturated absorption but this nonlinear effect is low in silicon materials and therefore difficult to implement on silicon substrates and large scale integration.
According to analysis, the scheme capable of achieving in-situ training of the optical neural network is few, and few devices for achieving the function on silicon are few, so that development of the scheme capable of meeting in-situ training of the optical neural network and based on the silicon devices has important practical value and wide application prospect.
Disclosure of Invention
In view of the above, the invention aims to provide a silicon-based light-operated micro-ring resonator for in-situ training of an optical neural network, which fills the blank of in-situ training by using a silicon-based device by adopting an error back propagation algorithm, has the advantages of simple structure, easy realization, lower cost and realization of large-scale integration.
In order to achieve the above purpose, the present invention provides the following technical solutions:
the silicon-based light-operated micro-ring resonator for in-situ training of the optical neural network consists of 2n signal transmission waveguides, n waveguide coupling areas, a ring resonant cavity and n optical power areas, wherein n is more than or equal to 1;
the signal transmission waveguide is used for realizing the input and output of forward and reverse optical signals;
the waveguide coupling region is used for coupling optical signals with specific wavelengths into the ring resonant cavity;
The ring resonant cavity is used for resonating an optical signal with a specific wavelength and changing the output power and the phase of the optical signal;
The optical power region is used for generating optical power, changing the bending degree of the regional waveguide, changing the resonance condition of the ring resonant cavity and realizing different responses of forward transmission signals and reverse transmission signals.
Further, the signal transmission waveguide is a silicon-based single-mode waveguide or a multi-mode waveguide, and the cross section of the signal transmission waveguide is ridge-shaped or strip-shaped.
Furthermore, the waveguide coupling area is formed by coupling a silicon-based waveguide and a silicon-based ring resonant cavity, and the silicon-based waveguide and the silicon-based ring resonant cavity adopt a lateral or vertical coupling mode.
Furthermore, the annular resonant cavity is a circular, runway-shaped or special-shaped resonant cavity, the waveguides forming the resonant cavity are silicon-based single-mode waveguides or multi-mode waveguides, and the cross section of the annular resonant cavity is strip-shaped.
Further, the optical power region generates optical gradient force between the annular resonant cavity and the substrate structure or between the annular resonant cavity and the adjacent waveguide structure so as to change the bending degree of the resonant cavity or the waveguide and change the resonant condition of the resonant cavity.
The invention has the beneficial effects that: the nonlinear activation function and the derivative thereof of the optical neural network are realized by using a single silicon-based integrated device, the technical difficulty that the nonlinear activation function unit needs to provide different responses in an reasoning stage and a training stage in the in-situ training of the optical neural network is solved, and the device has the advantages of simple structure, easy realization, easy large-scale integration and strong availability.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the instrumentalities and combinations particularly pointed out in the specification.
Drawings
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in the following preferred detail with reference to the accompanying drawings, in which:
FIG. 1 is a schematic diagram of the present invention;
FIG. 2 is a schematic diagram of a neural network back propagation algorithm;
FIG. 3 is a schematic diagram of a structure of a silicon-based light-controlled micro-ring resonator according to the present invention, (a) is a system block diagram of the silicon-based light-controlled micro-ring resonator, and (b) is a 3-D structure schematic diagram of the silicon-based light-controlled micro-ring resonator; (c) a schematic side structure;
FIG. 4 is a graph of the amount of waveguide deformation versus optical gradient force for forward optical signal wavelengths slightly less than the initial resonance wavelength;
FIG. 5 shows the output characteristics of the forward and reverse optical signals at different wavelength differences, and (a) and (c) show the transmission spectra of the device at different forward optical signal powers; (b) The output characteristic of the forward optical signal is the difference between the wavelength of the forward optical signal and the initial resonance wavelength; (d) The output characteristic of the reverse optical signal is the output characteristic of the reverse optical signal under the different difference values of the wavelength of the reverse optical signal and the initial resonance wavelength;
FIG. 6 shows the fitting result, (a) shows the activation function g; (b) is the approximate derivative H (λb) of the activation function;
FIG. 7 is a diagram of a fully connected neural network architecture;
Fig. 8 is a convolutional neural network architecture diagram.
Reference numerals: a 1-signal transmission waveguide, a 2-waveguide coupling region, a 3-ring resonator, a 4-optical power region.
Detailed Description
Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the illustrations provided in the following embodiments merely illustrate the basic idea of the present invention by way of illustration, and the following embodiments and features in the embodiments may be combined with each other without conflict.
Wherein the drawings are for illustrative purposes only and are shown in schematic, non-physical, and not intended to limit the invention; for the purpose of better illustrating embodiments of the invention, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the size of the actual product; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numbers in the drawings of embodiments of the invention correspond to the same or similar components; in the description of the present invention, it should be understood that, if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc., that indicate an azimuth or a positional relationship based on the azimuth or the positional relationship shown in the drawings, it is only for convenience of describing the present invention and simplifying the description, but not for indicating or suggesting that the referred device or element must have a specific azimuth, be constructed and operated in a specific azimuth, so that the terms describing the positional relationship in the drawings are merely for exemplary illustration and should not be construed as limiting the present invention, and that the specific meaning of the above terms may be understood by those of ordinary skill in the art according to the specific circumstances.
Fig. 1 is a schematic diagram of the present invention. Fig. 2 shows a schematic diagram of an optical neural network back propagation algorithm, in which the jth neuron of the forward propagation k layer can be represented by the following two equations in a common neural network.
Wherein z j k is the state of the jth neuron of the k layer after passing through the linear operation unit, w ji k is the weight between the jth neuron of the k layer and the ith neuron of the k-1 layer, alpha i k-1 is the output of the ith neuron of the k-1 layer, and g (·) is a nonlinear activation function. And when training by using the back propagation algorithm, the gradient of the loss function F is as follows:
Wherein the method comprises the steps of Is the error function of the j-th neuron of the k layers, and can be obtained according to the chain rule
Where g' (. Cndot.) is the bias of the activation function g (-) to z j k. Thus, to achieve in-situ training, the nonlinear activation function unit needs to provide different responses in the forward and backward propagation processes, i.e., the forward propagation response is g (·) and the backward propagation response is g' (·).
FIG. 3 shows a schematic structural diagram of an embodiment of a silicon-based optically controlled micro-ring resonator for in situ training of an optical neural network according to the present invention. The silicon-based light-operated micro-ring resonator comprises a signal transmission waveguide 1, a waveguide coupling area 2, a ring resonant cavity 3 and a light power area 4.
Fig. 3 (a) is a block diagram of a silicon-based optically controlled micro-ring resonator system. Fig. 3 (b) and 3 (c) show a 3-D structure schematic diagram of this embodiment and a side structural diagram thereof. In the embodiment of the invention, the signal transmission waveguide 1 is a single-mode waveguide with a strip-shaped cross section, the waveguide coupling area 2 is the lateral coupling of a straight waveguide and a circular resonant cavity, the annular resonant cavity 3 is a circular resonant cavity, the optical power area 4 adopts a substrate structure and an arc waveguide to generate optical gradient force, one part of the substrate is corroded by HF acid, and one part of the annular resonant cavity 3 is suspended. The forward and reverse transmitted optical signals are of different wavelengths, assuming lambda f and lambda b, respectively.
The power formula in embodiments of the invention may be expressed as
Where n eff is the effective index, d is the waveguide-to-substrate spacing, and U is the energy in the ring cavity, which can be expressed as
Where P in is the optical power in the straight waveguide, c is the speed of light in vacuum, lambda r is the resonant wavelength of the ring cavity, tau i -1 is the intrinsic decay rate of the optical field amplitude in the ring cavity, and tau e -1 is the external decay rate of the optical field amplitude. Substituting formula (6) into formula (5) gives a device of optical power of
Where x is the amount of waveguide deformation. The deformation generates an elastic force which is opposite to the direction of the light, and the size of the deformation increases linearly with the increase of the deformation amount. When the force is equal to the optical force, the deformation amount reaches the maximum, the cantilever reaches a balance, and the system is stable.
In the embodiment of the invention, the silicon-based straight waveguide has the width of 500nm, the height of 220nm, the radius of the ring resonator is 20 mu m, the coupling interval is 250nm, the interval between the cantilever and the substrate in the ring resonator is 200nm, and the initial resonant wavelength of the ring resonator is lambda r0 = 1552.298nm through calculation. Assuming that λ f = 1552.297nm, the relationship between the optical gradient force and the amount of waveguide deformation at different input optical signal powers can be obtained as shown in fig. 4. As can be seen, as the amount of waveguide deformation increases, the resonant wavelength of the ring resonator exhibits a red shift, resulting in a decrease in optical gradient force. In addition, the transmission spectrum of the ring cavity may also change when a different lambda f is selected. Fig. 5 (a) depicts a schematic diagram of the selection of different lambda f, while the corresponding transmission spectrum is shown in fig. 5 (b). Wherein lambda f is 1552.297nm, 1552.338nm, 1552.368nm, respectively. To construct an activation function suitable for classification tasks, λ f = 1552.297nm is finally selected. Likewise, selecting a different lambda b will also result in a different transmission performance. FIG. 5 (c) depicts a corresponding schematic diagram, with lambda b chosen to be 1552.268nm, 1552.298nm, 1552.318nm, resulting in the transmission curve of FIG. 5 (d). In order for the transmission curve of the counter-propagating signal to meet the derivative of the forward curve, λ b = 1552.298nm is finally chosen.
The curve can be fitted to the above formula (8) and formula (9), wherein the derivative of formula (8) is formula (10), whenIn this case, the brackets in equation (10) can be regarded as 1, and equation (10) is equivalent to equation (9), that is, the backward optical signal propagation function is the derivative of the forward optical signal propagation function.
Where P f,out is the forward optical signal output power, P f,in is the forward optical signal input power, P b,out is the reverse optical signal output power, P b,in is the reverse optical signal input power, α 0 is the resonant optical depth, and α is the fitting factor. When α 0 =5.6 and α=0.85, the fitting effect is the best, and the result is shown in fig. 6, and fig. 6 (a) is an activation function g; fig. 6 (b) is an approximate derivative H (λb) of the activation function.
In order to verify that the nonlinear activation function and the approximate derivative thereof realized by the method can be applied to the neural network, the nonlinear activation function and the approximate derivative thereof are substituted into the fully-connected neural network for classification capability test, and the fully-connected architecture is shown in fig. 7, wherein the fully-connected architecture comprises an input layer and a hidden layer, the input layer comprises 784-dimensional input vectors, the hidden layer comprises two layers, each layer comprises 128 neurons, and the output layer comprises 10 neurons. The accuracy of the nonlinear activation function and the ReLU function which are proposed by the test are consistent, and can reach 98%.
In addition to the fully-connected neural network, the embodiment of the invention substitutes the function and the approximate derivative into the convolutional neural network for classification capability test, and the convolutional neural network architecture is shown in fig. 8. Besides the input and output layers, a convolution layer, a pooling layer and a full connection layer are also included. The convolution layers have two layers, 32 and 64 channels respectively, and each layer convolves the input signal with a convolution kernel of 5x5, with a step size of 1, and no padding. After the convolutional network, classification is performed by a fully-connected network of individual 128 neuron hidden layers and 10 neuron output layers. The classification test is performed on the MNIST data set, the Kuzushiji-MNIST (KMNIST) data set and the Extended-MNIST (EMNIST) data set by using the network, the forward/backward signal propagation function, the forward propagation function and the theoretical derivative thereof realized by the embodiment of the invention are brought into the network, respectively represent a nonlinear activation function and an in-situ backward propagation algorithm, and are compared with a common activation function and a training algorithm, and the results are shown in table 1. The embodiment of the invention has high precision in classification tasks, and can be widely applied to all-optical neural networks for in-situ training.
TABLE 1
Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the claims of the present invention.
Claims (5)
1.A light-operated micro-ring resonator of silicon based for optical neural network normal position training, its characterized in that: the micro-ring resonator consists of 2n signal transmission waveguides, n waveguide coupling areas, a ring resonant cavity and n optical power areas, wherein n is more than or equal to 1;
the signal transmission waveguide is used for realizing the input and output of forward and reverse optical signals;
the waveguide coupling region is used for coupling optical signals with specific wavelengths into the ring resonant cavity;
The ring resonant cavity is used for resonating an optical signal with a specific wavelength and changing the output power and the phase of the optical signal;
The optical power region is used for generating optical power, changing the bending degree of the regional waveguide, changing the resonance condition of the ring resonant cavity and realizing different responses of forward transmission signals and reverse transmission signals.
2. The silicon-based optically controlled microring resonator for in-situ training of an optical neural network of claim 1, wherein: the signal transmission waveguide is a silicon-based single-mode waveguide or a multi-mode waveguide, and the cross section of the signal transmission waveguide is ridge-shaped or strip-shaped.
3. The silicon-based optically controlled microring resonator for in-situ training of an optical neural network of claim 1, wherein: the waveguide coupling area is formed by coupling a silicon-based waveguide and a silicon-based annular resonant cavity, and the silicon-based waveguide and the silicon-based annular resonant cavity adopt a lateral or vertical coupling mode.
4. The silicon-based optically controlled microring resonator for in-situ training of an optical neural network of claim 1, wherein: the ring resonant cavity is a circular, runway-shaped or special-shaped resonant cavity, the waveguides forming the resonant cavity are silicon-based single-mode waveguides or multi-mode waveguides, and the cross section of the ring resonant cavity is strip-shaped.
5. The silicon-based optically controlled microring resonator for in-situ training of an optical neural network of claim 1, wherein: the optical power region generates optical gradient force between the annular resonant cavity and the substrate structure or between the annular resonant cavity and the adjacent waveguide structure so as to change the bending degree of the resonant cavity or the waveguide and change the resonant condition of the resonant cavity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410143193.XA CN118192010B (en) | 2024-02-01 | 2024-02-01 | Silicon-based light-operated micro-ring resonator for in-situ training of optical neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410143193.XA CN118192010B (en) | 2024-02-01 | 2024-02-01 | Silicon-based light-operated micro-ring resonator for in-situ training of optical neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118192010A true CN118192010A (en) | 2024-06-14 |
| CN118192010B CN118192010B (en) | 2024-08-13 |
Family
ID=91410516
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410143193.XA Active CN118192010B (en) | 2024-02-01 | 2024-02-01 | Silicon-based light-operated micro-ring resonator for in-situ training of optical neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118192010B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1494662A (en) * | 2000-09-22 | 2004-05-05 | ��������ͨ���о�Ժ | Methods of altering resonance of waveguide micro-resonators |
| US20150168803A1 (en) * | 2013-12-13 | 2015-06-18 | Telefonaktiebolaget L M Ericsson (Publ) | Parallel and WDM Silicon Photonics Integration in Information and Communications Technology Systems |
| CN105629387A (en) * | 2016-03-14 | 2016-06-01 | 华中科技大学 | Micro-ring waveguide device |
| CN110780506A (en) * | 2019-07-17 | 2020-02-11 | 中国地质大学（武汉） | All-optical tunable filter based on silicon-based micro-ring |
| CN111721336A (en) * | 2020-03-09 | 2020-09-29 | 浙江工业大学 | Self-interference micro-ring resonant cavity sensing classification identification method based on supervised learning |
| CN115202127A (en) * | 2022-07-20 | 2022-10-18 | 浙江大学绍兴研究院 | Method for realizing all-optical nonlinear activation function based on silicon-based micro-ring resonator |
| CN116049610A (en) * | 2023-01-16 | 2023-05-02 | 之江实验室 | Response system and method of optical nonlinear function |
| CN116243427A (en) * | 2023-02-22 | 2023-06-09 | 浙江大学 | Multichannel amplitude equalizer based on micro-ring resonator array |
-
2024
- 2024-02-01 CN CN202410143193.XA patent/CN118192010B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1494662A (en) * | 2000-09-22 | 2004-05-05 | ��������ͨ���о�Ժ | Methods of altering resonance of waveguide micro-resonators |
| US20150168803A1 (en) * | 2013-12-13 | 2015-06-18 | Telefonaktiebolaget L M Ericsson (Publ) | Parallel and WDM Silicon Photonics Integration in Information and Communications Technology Systems |
| CN105629387A (en) * | 2016-03-14 | 2016-06-01 | 华中科技大学 | Micro-ring waveguide device |
| CN110780506A (en) * | 2019-07-17 | 2020-02-11 | 中国地质大学（武汉） | All-optical tunable filter based on silicon-based micro-ring |
| CN111721336A (en) * | 2020-03-09 | 2020-09-29 | 浙江工业大学 | Self-interference micro-ring resonant cavity sensing classification identification method based on supervised learning |
| CN115202127A (en) * | 2022-07-20 | 2022-10-18 | 浙江大学绍兴研究院 | Method for realizing all-optical nonlinear activation function based on silicon-based micro-ring resonator |
| CN116049610A (en) * | 2023-01-16 | 2023-05-02 | 之江实验室 | Response system and method of optical nonlinear function |
| CN116243427A (en) * | 2023-02-22 | 2023-06-09 | 浙江大学 | Multichannel amplitude equalizer based on micro-ring resonator array |
Non-Patent Citations (2)
| Title |
|---|
| 桂林;左健存;吴中林;王胜利;: "一种包含非线性光学效应的硅基微环稳态模型", 光学学报, no. 04, 1 December 2017 (2017-12-01) * |
| 费瑶;何玉铭;杨富华;李兆峰;: "端面反射和超模损耗对波导环形谐振腔输出谱线的影响", 中国激光, no. 05, 25 December 2017 (2017-12-25) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118192010B (en) | 2024-08-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109639359B (en) | Photonic neural network convolution layer chip based on micro-ring resonator | |
| US11334107B2 (en) | Apparatus and methods for optical neural network | |
| Bai et al. | Photonic multiplexing techniques for neuromorphic computing | |
| Filipovich et al. | Silicon photonic architecture for training deep neural networks with direct feedback alignment | |
| Wu et al. | Low-threshold all-optical nonlinear activation function based on a Ge/Si hybrid structure in a microring resonator | |
| CN114358271B (en) | Time-wavelength interweaving photon neural network convolution acceleration chip | |
| Brückerhoff-Plückelmann et al. | Chalcogenide phase-change devices for neuromorphic photonic computing | |
| Kojima et al. | Deep neural networks for inverse design of nanophotonic devices | |
| Xu et al. | Silicon-based optoelectronics for general-purpose matrix computation: a review | |
| Wang et al. | Metasurface on integrated photonic platform: from mode converters to machine learning | |
| CN113961035A (en) | Deep learning-oriented high-efficiency reconfigurable all-optical neural network computing chip architecture | |
| Tanomura et al. | Scalable and robust photonic integrated unitary converter based on multiplane light conversion | |
| Liao et al. | Integrated photonic neural networks: Opportunities and challenges | |
| US20220300796A1 (en) | Photonic neural network | |
| Filipovich et al. | Monolithic silicon photonic architecture for training deep neural networks with direct feedback alignment | |
| CN115905792A (en) | Mach-Zehnder interferometer network for optical real number matrix calculation | |
| CN118192010B (en) | Silicon-based light-operated micro-ring resonator for in-situ training of optical neural network | |
| Ou et al. | Hypermultiplexed Integrated Tensor Optical Processor | |
| Fu et al. | Integrated diffractive optical neural network with space-time interleaving | |
| CN117032392A (en) | Silicon-based polarization multiplexing integrated optical computing chip | |
| Peserico et al. | Design and testing of silicon photonic 4F system for convolutional neural networks | |
| Chen et al. | Hypermultiplexed integrated-photonics-based tensor optical processor | |
| Shen et al. | Large-scale neuromorphic systems enabled by integrated photonics | |
| Soroush et al. | Predicting broadband resonator-waveguide coupling for microresonator frequency combs through fully connected and recurrent neural networks and attention mechanism | |
| CN113918118A (en) | Thin-film lithium niobate micro-ring filter array and optical vector-matrix multiplier |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |