CN117454951A - Optical neural network device - Google Patents
Optical neural network device Download PDFInfo
- Publication number
- CN117454951A CN117454951A CN202311426215.5A CN202311426215A CN117454951A CN 117454951 A CN117454951 A CN 117454951A CN 202311426215 A CN202311426215 A CN 202311426215A CN 117454951 A CN117454951 A CN 117454951A
- Authority
- CN
- China
- Prior art keywords
- mirror
- optical
- neural network
- fourier transform
- light emitting
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000003287 optical effect Effects 0.000 title claims abstract description 166
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 50
- 238000001514 detection method Methods 0.000 claims abstract description 11
- 239000004973 liquid crystal related substance Substances 0.000 claims description 26
- XUIMIQQOPSSXEZ-UHFFFAOYSA-N Silicon Chemical compound [Si] XUIMIQQOPSSXEZ-UHFFFAOYSA-N 0.000 claims description 11
- 229910052710 silicon Inorganic materials 0.000 claims description 8
- 239000010703 silicon Substances 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 6
- 230000033228 biological regulation Effects 0.000 claims description 5
- 230000010354 integration Effects 0.000 abstract description 8
- 230000005540 biological transmission Effects 0.000 abstract description 6
- 238000004364 calculation method Methods 0.000 abstract description 6
- 238000005516 engineering process Methods 0.000 abstract description 5
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000013461 design Methods 0.000 abstract description 3
- 230000001105 regulatory effect Effects 0.000 description 11
- 230000001276 controlling effect Effects 0.000 description 9
- 230000006872 improvement Effects 0.000 description 9
- 238000003384 imaging method Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000002086 nanomaterial Substances 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000000034 method Methods 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 239000000523 sample Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000002834 transmittance Methods 0.000 description 1
Classifications
-
- 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
- G06N3/0675—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means using electro-optical, acousto-optical or opto-electronic means
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Neurology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an optical neural network device, which comprises an integrated chip and an optical path component, wherein the integrated chip comprises a light emitting part and a detecting part, the light emitting part and the detecting part are integrated on the same integrated chip, an optical signal emitted by the light emitting part sequentially passes through a Fourier transform part, a first reflecting mirror, a second reflecting mirror and an inverse Fourier transform part and then reaches the detecting part, and a mask part is arranged between the Fourier transform part and an optical path of the inverse Fourier transform part. The optical neural network device has the advantages that the volume is greatly reduced through the design of the optical path component, the light emitting part and the detection part can be integrated on the same integrated chip, the integration level is improved, the data transmission speed is faster, the loss is less, and the influence of stray light on the calculation result is avoided, so that the optical neural network has better application prospect, and the progress of the optical neural network technology is promoted.
Description
Technical Field
The invention relates to the field of optical computation of primitive optical elements, in particular to an optical neural network device.
Background
The optical neural network built by the optical element can partially replace an electric signal processing unit to carry out convolution processing, the key characteristics of a target are extracted, compared with the processing of an electric signal, the optical neural network has higher operation speed and higher energy efficiency ratio, but the application of the existing optical neural network in the scene has some problems, wherein the optical neural network mainly has the advantages that compared with an electronic element, the volume of the optical element is overlarge, the limitation of the focal length of an optical path is combined, a longer space distance is required to be occupied between the sending and receiving of the optical signal, and the propagation error of the optical signal and the loss of the electric signal exist in the longer distance, so that the optical neural network built by the existing optical element cannot well meet the use requirement, and the development of the optical neural network is restricted.
Disclosure of Invention
In order to solve at least one of the problems of the prior art that the optical neural network is oversized and has propagation errors of optical signals and loss of electrical signals, the invention aims to provide an optical neural network device with smaller volume and higher integration level 。
To achieve the above object, an embodiment of the present invention provides an optical neural network device, including:
the integrated chip comprises a light emitting part and a detecting part, wherein the light emitting part is used for emitting light signals, the detecting part is used for receiving the light signals, and the light emitting part and the detecting part are integrated on the same integrated chip;
an optical path component including a fourier transform section, a first mirror, a second mirror, an inverse fourier transform section, and a mask section;
the light signal emitted by the light emitting part sequentially passes through the Fourier transform part, the first reflecting mirror, the second reflecting mirror and the inverse Fourier transform part and then reaches the detection part, and the mask part is arranged between the light paths from the Fourier transform part to the inverse Fourier transform part.
As a further improvement of the present invention, the fourier transform section, and/or the inverse fourier transform section, and/or the mask section is provided as a superlens, a supersurface device, a diffractive optical element, or a curved lens.
As a further improvement of the present invention, a plane perpendicular to the centers of the light emitting part and the detecting part is a symmetrical plane, the first reflecting mirror and the second reflecting mirror are symmetrically disposed on both sides of the symmetrical plane, and an optical path from the light emitting part to the first reflecting mirror and an optical path from the detecting part to the second reflecting mirror are symmetrically disposed on both sides of the symmetrical plane.
As a further improvement of the present invention, the first mirror and the second mirror are each provided as a planar mirror, the first mirror is inclined in a direction 45 ° away from the lower side of the integrated chip, the second mirror is inclined in a direction 45 ° toward the lower side of the integrated chip side, and the mask portion is provided at an intermediate position of the first mirror and the second mirror.
As a further improvement of the present invention, the fourier transform unit and the inverse fourier transform unit are provided as the same fourier lens, the first mirror and the second mirror are both provided as concave mirrors, the specifications of the first mirror and the second mirror are the same, and the optical signal emitted from the light emitting unit passes through the fourier lens obliquely downward, passes through the first mirror and the second mirror, and then passes through the fourier lens obliquely upward.
As a further improvement of the present invention, the distances from the fourier lens to the light paths of the light emitting section and the detecting section are each equal to the focal length of the fourier lens, the distances from the fourier lens to the light paths of the first reflecting mirror and the second reflecting mirror are each equal to the sum of the focal length of the fourier lens and the focal length of the concave reflecting mirror, and the distances from the light paths of the first reflecting mirror and the second reflecting mirror are each equal to twice the focal length of the concave reflecting mirror.
As a further improvement of the present invention, the optical path assembly further includes a third plane mirror and a fourth plane mirror, and the optical signal passes through the fourier lens obliquely downward after passing through the third plane mirror, the first mirror, the second mirror, and the fourth plane mirror, and then passes through the fourier lens obliquely upward;
the mask part is arranged at the position of the reflected light signal of the third plane mirror or the fourth plane mirror;
the third plane mirror and the fourth plane mirror are arranged as the same plane mirror.
As a further improvement of the present invention, the distances of the fourier lens to the light emitting section, the detecting section, the third plane mirror, and the fourth plane mirror are equal to the focal length of the fourier lens, the distances of the first mirror to the light path of the third plane mirror, and the distances of the second mirror to the light path of the fourth plane mirror are equal to the focal length of the concave mirror, and the distances of the light paths of the first mirror and the second mirror are equal to twice the focal length of the concave mirror.
As a further improvement of the present invention, the mask portion is provided on an optical path between the fourier lens and the first mirror or an optical path between the fourier lens and the second mirror, and a distance from the mask portion to the optical path of the fourier lens is equal to a focal length of the fourier lens.
As a further improvement of the present invention, the integrated chip includes a processor, the processor outputs a light emitting signal to the light emitting part and receives an optical signal of the detecting part, the mask part is a liquid crystal on silicon part, the optical neural network device includes a regulating and controlling component, the processor outputs a voltage signal to the regulating and controlling component according to the received optical signal, the regulating and controlling component applies a controllable voltage to the liquid crystal on silicon part according to the voltage signal, and the controllable voltage is used for controlling the spatial distribution of the liquid crystal in the liquid crystal on silicon part so as to change the modulation result of the optical signal by the mask part.
Compared with the prior art, the invention has the following beneficial effects: according to the optical neural network device, the light emitting part and the detection part are integrated on the same integrated chip through the design of the light path component, so that on one hand, a traditional light path is folded, the space volume required to be occupied by the optical neural network device is greatly reduced relative to the traditional linear optical neural network, on the other hand, the data transmission speed on the chip is faster, the loss is less and the stability is better on the same integrated chip, on the other hand, the light signal is emitted and the light signal is received on the same straight line, the influence of stray light on the light calculation result is avoided, and therefore the optical neural network has a better application prospect, and advances of the optical neural network technology are promoted.
Drawings
Fig. 1 is a schematic structural view of an optical neural network device according to a first embodiment of the present invention;
FIG. 2 is a schematic diagram of a mask portion of an embodiment of the present invention under the control of a control element to change the spatial distribution of liquid crystal;
fig. 3 is a schematic structural diagram of an optical neural network device according to an embodiment of the second embodiment of the present invention;
fig. 4 is a schematic structural view of an optical neural network device according to another embodiment of the second embodiment of the present invention;
fig. 5 is a schematic structural diagram of an optical neural network device according to a third embodiment of the present invention;
s1, a symmetry plane; 10. an integrated chip; 11. a light emitting section; 12. a detection unit; 20. an optical path component; 21. a Fourier transform unit; 22. an inverse fourier transform unit; 23. a first mirror; 24. a second mirror; 25. a mask portion; 30. an optical path component; 31. a fourier lens; 32. a mask portion; 33. a first mirror; 34. a second mirror; 40. an optical path component; 41. a fourier lens; 42. a third plane mirror; 43. a first mirror; 44. a second mirror; 45. a fourth planar mirror; 46. and a mask portion.
Detailed Description
The present invention will be described in detail below with reference to specific embodiments shown in the drawings. These embodiments are not intended to limit the invention and structural, methodological, or functional modifications of these embodiments that may be made by one of ordinary skill in the art are included within the scope of the invention.
It will be appreciated that terms such as "upper," "above," "lower," "below," and the like, as used herein, refer to spatially relative positions and are used for ease of description to describe one element or feature's relationship to another element or feature as illustrated in the figures. The term spatially relative position may be intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures.
An embodiment of the invention provides an optical neural network device with smaller volume and higher integration level.
The optical neural network device comprises an integrated chip 10 and an optical path component, wherein the integrated chip 10 comprises a light emitting part 11 and a detecting part 12, the light emitting part 11 is used for emitting light signals, the detecting part 12 is used for receiving the light signals, the optical path component can modulate the light signals according to processing tasks, extract characteristic parameters of input images, realize parallel analog calculation and change the propagation direction of the light signals. In application, the optical neural network device can be used as an optical convolution layer of a hybrid neural network to carry out convolution operation on an input image so as to finish tasks such as image recognition, image classification and the like. The light path assembly is described in more detail below in three embodiments.
The light emitting section 11 and the detecting section 12 of the present embodiment are integrated on the same integrated chip 10, as shown in fig. 1, 3, 4, 5. Correspondingly, the optical path module includes a fourier transform unit, a first mirror, a second mirror, an inverse fourier transform unit, and a mask unit, and the optical signal emitted from the light emitting unit 11 sequentially passes through the fourier transform unit, the first mirror, the second mirror, and the inverse fourier transform unit, and then reaches the detecting unit 12, and the mask unit is disposed between the optical paths from the fourier transform unit to the inverse fourier transform unit. The different transmittances of the light at different positions on the mask part can realize the regulation and control of the amplitude of the complex amplitude signal in the optical frequency domain. The optical signal is converted into an optical frequency domain signal by the fourier transform unit, modulated by the mask unit, and then enters the inverse fourier transform unit to be transformed into a spatial domain optical signal, so that the convolution calculation is performed on the input image corresponding to the optical signal based on the convolution kernel using the mask unit.
Compared with the existing 4f optical path system, the position relationship between the light emitting part 11 and the detecting part 12 is changed, on the other hand, the light emitting part 11 and the detecting part 12 can emit light signals and receive light signals on the same integrated chip 10 through the arrangement of the first reflecting mirror and the second reflecting mirror, which is equivalent to folding an optical path, so that the whole volume is reduced, the light emitting part 11 and the detecting part 12 are manufactured on the same integrated chip 10, the assembly process difficulty is reduced, and the product cost is reduced.
The fourier transform section, and/or the inverse fourier transform section, and/or the mask section in this embodiment are provided as a superlens, a supersurface device, a diffractive optical element (Diffractive Optical Elements, DOE), or a curved lens, which can achieve miniaturization and planarization of the optical element while retaining its optical function, and can regulate and control the amplitude and phase of an input optical signal simultaneously by specially designing the geometry and arrangement of nanoscale microstructure elements in the superlens.
The fourier transform unit and the inverse fourier transform unit can focus light in a three-dimensional space and perform fourier transform on an incident light signal.
For the mask part, the geometric dimensions of the nano-structures at different positions on the superlens, such as parameters of height, period, arrangement mode, radius, geometric shape and the like, can be flexibly regulated and controlled so as to realize the simultaneous regulation and control of the amplitude and the phase of the optical frequency domain complex amplitude signal in the Fourier plane. The nanostructure size and spatial distribution of the mask portion can be further determined by calculating the required convolution kernel, i.e., the complex amplitude distribution in the frequency domain via fourier transform.
Further, the integrated chip 10 includes a processor, the processor outputs a light emitting signal to the light emitting portion 11 and receives an optical signal of the detecting portion 12, the mask portion is a liquid crystal on silicon device, the optical neural network device includes a regulating component, the processor outputs a voltage signal to the regulating component according to the received optical signal, the regulating component applies a controllable voltage to the liquid crystal on silicon device according to the voltage signal, and the controllable voltage is used for controlling a spatial distribution of liquid crystal in the liquid crystal on silicon device so as to change a modulation result of the optical signal by the mask portion.
In this embodiment, as shown in fig. 2, the mask portion is manufactured by adopting a liquid crystal on silicon technology (LCoS, liquid Crystal on Silicon), so that the spatial distribution of the amplitude and the phase of the optical signal can be modulated simultaneously, and the spatial distribution of the liquid crystal in the liquid crystal layer is controlled by addressing and controlling the voltage applied to the mask plate, so that different modulating effects of the spatial distribution of the amplitude and the phase of the optical signal are realized, that is, the programmable complex amplitude modulating capability is realized, and the variable convolution kernel in the whole convolution calculation optical neural network is corresponding.
More specifically, the adjusting and controlling component applies voltage to the liquid crystal layer of the mask part on each pixel to enable the liquid crystal molecules in the liquid crystal layer at the corresponding position to rotate spatially, after the liquid crystal molecules rotate, the polarization state of the incident light can be changed to realize the amplitude adjustment and control of the emergent light by utilizing the birefringence effect of the liquid crystal molecules, and when the liquid crystal molecules rotate spatially, the equivalent refractive index of the liquid crystal layer can change to further realize the phase adjustment of the incident light.
Therefore, the phase and amplitude of the incident light can be regulated and controlled simultaneously by regulating the voltages applied to the two ends of the liquid crystal layer of the mask part, and different regulating and controlling effects can be realized by changing the magnitude of the added voltage, namely the programmable regulating and controlling capability of the incident light complex amplitude is realized, so that the continuous iteration of the convolution kernel can be realized through programming the convolution kernel to construct the mask part meeting the requirement of a specific task.
Different implementations of the optical neural network device are further described below in three embodiments. The difference is that the mirror adopted in the embodiment 1 is a plane mirror to realize imaging of the 4f system, the mirror adopted in the embodiment 2 is a concave mirror, the fourier transform part and the inverse fourier transform part are set to be the same lens, and finally the 8f imaging system is realized, and the embodiment 3 adds a new plane mirror on the basis of the embodiment 2 to realize the 8f imaging system with further folding space.
In addition, in order to clearly express the position and direction described in the present embodiment, in the following embodiment, the definition refers to the direction in fig. 1, the integrated chip 10 is located on the left side of the optical path component, the opposite direction is the right side, the light emitting portion 11 is located above the detecting portion 12, the opposite direction is the lower side, and the up-down-left-right direction is defined by only one name, and is not limited to "up-down" in the physical sense; for example, when the integrated chip 10 is laid on a horizontal plane, the light emitting section 11 and the detecting section 12 may be located on the same plane in the up-down direction in the matter.
And defines a plane perpendicular to the centers of the light emitting part 11 and the detecting part 12 as a symmetrical plane S1, the first reflecting mirror and the second reflecting mirror are symmetrically arranged at both sides of the symmetrical plane S1, and the light path from the light emitting part 11 to the first reflecting mirror and the light path from the detecting part 12 to the second reflecting mirror are symmetrically arranged at both sides of the symmetrical plane S1.
Example 1
In this embodiment, as shown in fig. 1, the first mirror 23 and the second mirror 24 are each provided as a planar mirror, the first mirror 23 is inclined in a direction 45 ° away from the side of the integrated chip 10, the second mirror 24 is inclined in a direction 45 ° away from the side of the integrated chip 10, and the mask portion 25 is provided at an intermediate position between the first mirror 23 and the second mirror 24.
The light beam from the light emitting unit 11 to the fourier transform unit 21 is limited to the right horizontal direction, is reflected downward after reaching the first mirror 23, is reflected leftward after reaching the second mirror 24 through the mask unit 25, and reaches the detection unit 12 on the same integrated chip 10 as the light emitting unit 11, wherein the focal lengths of the fourier transform unit 21 and the inverse fourier transform unit 22 are the same as f, and the distances of the light emitting unit 11 to the fourier transform unit 21 and the inverse fourier transform unit 22 to the detection unit 12 are f, the distances of the light path from the fourier transform unit 21 to the first mirror 23 and the light path from the inverse fourier transform unit 22 to the second mirror 24 are f/2, and the distances and the positions of the mask unit 25 to the first mirror 23 and the light path from the mask unit 25 to the second mirror 24 are f/2, respectively, as shown in fig. 1.
The optical neural network of the present embodiment is an improved 4f imaging system under a folded optical path, and the description is continued with reference to fig. 1:
first, the light emitting unit 11 is generally constituted by a parallel light source, for example, a laser light source combined with a waveguide array, a laser light source combined with a digital microscopic device, or a laser light source combined with a liquid crystal spatial light modulator, generates an image signal, and emits an optical signal of parallel light to the fourier transform unit 21. Wherein the laser light source outputs a high-frequency signal in combination with the waveguide array, and accordingly, the detection section 12 uses the photodiode array as a detector to realize high-frequency detection of the signal, thereby realizing convolution operation of the input image signal and different convolution kernels at a higher frequency, and reducing the time required for iterating the convolution kernels. If the light emitting unit 11 emits a low frequency signal, the detecting unit 12 may use a CMOS detector.
Next, the optical signal is transmitted through the fourier transform unit 21, and then the optical signal in the spatial domain emitted from the light emitting unit 11 is transformed into the frequency domain by fourier transform.
The optical signal will then pass through the first mirror 23, and the first mirror 23 changes the propagation direction of the optical signal, so that the fourier transformed optical signal frequency domain distribution image is imaged in the perpendicular direction (fourier plane) to the original optical signal propagation direction, achieving an object plane (the plane where the light emitting part 11 is located) and an off-axis of the fourier plane.
The optical signal then passes through the mask portion 25 and the second mirror 24, and then passes through the inverse fourier transform portion 22 to perform an inverse fourier transform operation, so that the optical signal frequency domain distribution is converted into a spatial domain, and is imaged at the image plane (the plane where the probe portion 12 is located). The fourier transform unit 21 and the inverse fourier transform unit 22 are designed in the same manner to ensure equal focal lengths, and the phases of the super surfaces of the fourier transform unit 21 and the inverse fourier transform unit 22 are identical. The distances from the center positions of the first reflecting mirror 23 and the second reflecting mirror 24 to the object plane are equal, and the reflecting angles are 45 degrees, so that the object plane and the image plane can be located on the same plane, and meanwhile, the Fourier plane for frequency domain optical signal imaging after the optical signals are subjected to Fourier transformation is located at the middle position of the two reflecting mirrors. The folded optical path introduced into the super surface not only effectively reduces the propagation path of the optical signal in the optical system and is beneficial to the miniaturization and integration of the whole system, but also can realize the output image signal and the input image signal on the same integrated chip, and the transmission distance is reduced by half compared with the traditional 4f imaging system.
In addition, since the light emitting part 11 and the detecting part 12 are not on the same straight line, the incident light signal area and the light collecting light signal area are easily separated by adding the light shielding plate, so that the effect of better eliminating stray light is achieved, and the crosstalk of stray light signals is avoided.
Example 2
In the present embodiment, the fourier transform unit and the inverse fourier transform unit are provided as the same fourier lens 31, and the first mirror 33 and the second mirror 34 are each provided as a concave mirror, and the first mirror 33 and the second mirror 34 have the same specifications. The present embodiment can further reduce the overall volume, and can further reduce the length in the lateral direction with respect to embodiment 1, and can further reduce the use of one inverse fourier transform unit, that is, use of one fourier lens 31 to simultaneously realize the functions of both the fourier transform unit and the inverse fourier transform unit, that is, the operation of performing inverse fourier transform on the input image to the frequency domain and the conversion of the frequency domain signal to the spatial domain is performed by one fourier lens 31, thereby improving the imaging accuracy and further ensuring that the input image and the output image are located on the same plane. In fig. 3 or 4, the optical signal emitted from the light emitting unit 11 passes through the fourier lens 31 obliquely downward, passes through the first mirror 33 and the second mirror 34, and then passes through the fourier lens 31 obliquely upward.
The concave mirror may be a spherical mirror or a parabolic mirror, and the first mirror 33 and the second mirror 34 have the same focal length, and this embodiment includes two embodiments, one of which is shown in fig. 3, where the focal lengths of the first mirror 33, the second mirror 34, and the fourier lens 31 are all the same, and are all f1. Another embodiment is shown in fig. 4, where the focal length of the fourier lens 31 is f1, and the focal lengths of the first mirror 33 and the second mirror 34 are the same and f2, and f1 is not equal to f2.
The distances of the fourier lens 31 to the optical paths of the light emitting section 11 and the detecting section 12 are equal to the focal length of the fourier lens 31, the distances of the fourier lens 31 to the optical paths of the first mirror 33 and the second mirror 34 are equal to the sum of the focal length of the fourier lens 31 and the focal length of the concave mirror, and the distances of the optical paths of the first mirror 33 and the second mirror 34 are equal to twice the focal length of the concave mirror. That is, the optical path assembly 30 and the direction of the optical path are in a vertically symmetrical form, and an optical system of 8f1 shown in fig. 3 or an optical system of 4f1+4f2 shown in fig. 4 is formed, thereby realizing reduction of components and reduction of production cost.
In other embodiments, the focal lengths of the first mirror 33 and the second mirror 34 may be different, so that an effect of zooming in or out may be achieved.
Since the optical path is inclined from the light emitting section 11 to the first reflecting mirror 33 and from the detecting section 12 to the second reflecting mirror 34 not in the horizontal direction, the horizontal component of the distance in the left-right direction is smaller, that is, the length direction is smaller, and the optical path is further folded and the integration is higher in order to achieve the same optical path transmission distance.
The mask portion 32 is provided on the optical path between the fourier lens 31 and the first mirror 33 or on the optical path between the fourier lens 31 and the second mirror 34, and the distance of the mask portion 32 to the optical path of the fourier lens 31 is equal to the focal length of the fourier lens 31. That is, in fig. 3 and 4, the mask portion 32 may be moved up to the optical path between the fourier lens 31 and the second reflecting mirror 34 as another embodiment.
Other optical signal variations and optical principles are described in example 1.
Example 3
As shown in fig. 5, the optical path assembly 40 of the present embodiment further includes a third plane mirror 42 and a fourth plane mirror 45, and the optical signal passes through the fourier lens 41 obliquely downward, and then passes through the third plane mirror 42, the first mirror 43, the second mirror 44, and the fourth plane mirror 45, and then passes through the fourier lens 41 obliquely upward.
The mask 46 is provided at a position of the third plane mirror 42 or the fourth plane mirror 45 where the light signal is reflected, that is, the mask 46 is integrated with the third plane mirror 42 or the fourth plane mirror 45, and modulates the light signal while reflecting the light.
In the embodiment of fig. 5, the third planar mirror 42 and the fourth planar mirror 45 are provided as identical planar mirrors.
In other embodiments, the third plane mirror 42 and the fourth plane mirror 45 may be provided as two mirrors on the same plane.
Further, as shown in fig. 5, the focal length of the fourier lens 41 is f1, the focal lengths of the first mirror 43 and the second mirror 44 are the same and f2, f1 is not equal to f2, the distances of the optical paths of the fourier lens 41 to the light emitting section 11, the detecting section 12, the third plane mirror 42 and the fourth plane mirror 45 are equal to the focal length f1 of the fourier lens 41, the distances of the optical paths of the first mirror 43 to the third plane mirror 42, and the distances of the optical paths of the second mirror 44 to the fourth plane mirror 45 are equal to the focal length f2 of the concave mirror, and the distances of the optical paths of the first mirror 43 and the second mirror 44 are equal to twice the focal length f2 of the concave mirror.
In addition to the embodiment 2, since the optical path is folded between the first mirror 43, the second mirror 44, the third plane mirror 42 and the fourth plane mirror 45 a plurality of times during the period from the emission of the optical path from the fourier lens 41 to the re-emission of the fourier lens 41, the optical path is further folded, for example, the component in the left-right direction of the distance f1+f2 from the fourier lens 41 to the rightmost end is added to the distance in the left-right direction of the concave mirror, and is further reduced to the component in the left-right direction of f1, so that the horizontal component of the distance in the left-right direction can be smaller, that is, the length direction can be smaller, and the optical path can be further folded and the integration degree is higher, compared with the embodiment 2, in order to realize the same optical path transmission distance.
Other optical signal variations and optical principles are described in example 1.
Compared with the common technology, the embodiment has the following beneficial effects:
(1) According to the optical neural network device, the light emitting part and the detection part are integrated on the same integrated chip through the design of the light path component, so that on one hand, a traditional light path is folded, the space volume required to be occupied by the optical neural network device is greatly reduced relative to the traditional linear optical neural network, on the other hand, the data transmission speed on the chip is faster, the loss is less and the stability is better on the same integrated chip, on the other hand, the light signal is emitted and the light signal is received on the same straight line, the influence of stray light on the light calculation result is avoided, and therefore the optical neural network has a better application prospect.
(2) By further describing the three embodiments, the length of the optical path in the left-right direction can be further compressed to be smaller, so that higher integration level is realized, and the use of components can be reduced while the volume is further reduced, for example, the fourier transform part and the inverse fourier transform part are arranged to be the same fourier lens, so that on one hand, the cost is reduced, the integration level is improved, on the other hand, the imaging precision is improved, the image signal and the input image signal can be further ensured to be output on the same integrated chip, the benefit that the light emitting part and the detection part in the point (1) are integrated on the same integrated chip is further promoted, and the advancement of the optical neural network technology is further promoted.
It should be understood that although the present disclosure describes embodiments, not every embodiment is provided with a separate embodiment, and that this description is for clarity only, and that the skilled artisan should recognize that the embodiments may be combined as appropriate to form other embodiments that will be understood by those skilled in the art.
The above list of detailed descriptions is only specific to practical embodiments of the present invention, and they are not intended to limit the scope of the present invention, and all equivalent embodiments or modifications that do not depart from the spirit of the present invention should be included in the scope of the present invention.
Claims (10)
1. An optical neural network device, comprising:
the integrated chip comprises a light emitting part and a detecting part, wherein the light emitting part is used for emitting light signals, the detecting part is used for receiving the light signals, and the light emitting part and the detecting part are integrated on the same integrated chip;
an optical path component including a fourier transform section, a first mirror, a second mirror, an inverse fourier transform section, and a mask section;
the light signal emitted by the light emitting part sequentially passes through the Fourier transform part, the first reflecting mirror, the second reflecting mirror and the inverse Fourier transform part and then reaches the detection part, and the mask part is arranged between the light paths from the Fourier transform part to the inverse Fourier transform part.
2. The optical neural network device according to claim 1, wherein the fourier transform section, and/or the inverse fourier transform section, and/or the mask section is provided as a superlens, a supersurface device, a diffractive optical element, or a curved lens.
3. The optical neural network device according to claim 1, wherein a plane perpendicular to centers of the light emitting portion and the detecting portion is a symmetrical plane, the first mirror and the second mirror are symmetrically disposed on both sides of the symmetrical plane, and an optical path from the light emitting portion to the first mirror and an optical path from the detecting portion to the second mirror are symmetrically disposed on both sides of the symmetrical plane.
4. The optical neural network device according to claim 3, wherein the first mirror and the second mirror are each provided as a planar mirror, the first mirror is inclined in a direction 45 ° away from a side of the integrated chip, the second mirror is inclined in a direction 45 ° toward a side close to the integrated chip, and the mask portion is provided at an intermediate position between the first mirror and the second mirror.
5. The optical neural network device according to claim 3, wherein the fourier transform unit and the inverse fourier transform unit are provided as a same fourier lens, the first mirror and the second mirror are each provided as a concave mirror, the first mirror and the second mirror are identical in specification, and the optical signal emitted from the light emitting unit passes obliquely downward from the fourier lens, passes through the first mirror and the second mirror, and then passes obliquely upward into the fourier lens.
6. The optical neural network device of claim 5, wherein the distances of the optical paths of the fourier lens to the light emitting section and the detecting section are each equal to a focal length of the fourier lens, the distances of the optical paths of the fourier lens to the first mirror and the second mirror are each equal to a sum of the focal length of the fourier lens and the focal length of the concave mirror, and the distances of the optical paths of the first mirror and the second mirror are each equal to twice the focal length of the concave mirror.
7. The optical neural network device of claim 5, wherein the optical path assembly further includes a third planar mirror and a fourth planar mirror, the optical signal passing obliquely downward from the fourier lens, and then passing obliquely upward through the third planar mirror, the first mirror, the second mirror, and the fourth planar mirror before passing into the fourier lens;
the mask part is arranged at the position of the reflected light signal of the third plane mirror or the fourth plane mirror;
the third plane mirror and the fourth plane mirror are arranged as the same plane mirror.
8. The optical neural network device of claim 7, wherein the distance of the fourier lens to the optical paths of the light emitting section, the detecting section, the third plane mirror, and the fourth plane mirror is equal to the focal length of the fourier lens, the distance of the first mirror to the optical path of the third plane mirror, and the distance of the second mirror to the optical path of the fourth plane mirror is equal to the focal length of the concave mirror, and the distance of the optical paths of the first mirror and the second mirror is equal to twice the focal length of the concave mirror.
9. The optical neural network device according to claim 5, wherein the mask portion is provided on an optical path between the fourier lens and the first mirror or an optical path between the fourier lens and the second mirror, and a distance of the mask portion to the optical path of the fourier lens is equal to a focal length of the fourier lens.
10. The optical neural network device of claim 1, wherein the integrated chip includes a processor that outputs a light-emitting signal to the light-emitting portion and receives the light signal from the detecting portion, the mask portion is a liquid crystal on silicon member, the optical neural network device includes a regulation and control component that outputs a voltage signal to the regulation and control component according to the received light signal, and the regulation and control component applies a controllable voltage to the liquid crystal on silicon member according to the voltage signal, the controllable voltage being used to control a spatial distribution of liquid crystal in the liquid crystal on silicon member to change a modulation result of the light signal by the mask portion.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311426215.5A CN117454951A (en) | 2023-10-31 | 2023-10-31 | Optical neural network device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311426215.5A CN117454951A (en) | 2023-10-31 | 2023-10-31 | Optical neural network device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117454951A true CN117454951A (en) | 2024-01-26 |
Family
ID=89592504
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311426215.5A Pending CN117454951A (en) | 2023-10-31 | 2023-10-31 | Optical neural network device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117454951A (en) |
-
2023
- 2023-10-31 CN CN202311426215.5A patent/CN117454951A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111512189B (en) | Resonant waveguide grating and application thereof | |
| AU2020324338A1 (en) | Ultra-wide field-of-view flat optics | |
| EP2912369B1 (en) | Illumination device | |
| CN100385275C (en) | Active optical phase conjugating method and apparatus | |
| JP6363619B2 (en) | Optical phased array | |
| CN112088328B (en) | 2D pupil expander using holographic Bragg gratings | |
| US20200348526A1 (en) | Device for Forming at Least One Tilted Focused Beam in the Near Zone, From Incident Electromagnetic Waves | |
| US10871657B2 (en) | Floating image display | |
| US20170261651A1 (en) | Array-based camera lens system | |
| US11474395B2 (en) | Birefringent polymer based surface relief grating | |
| JP2001228420A (en) | Device for performing dynamic control of luminous flux direction within wide visual field | |
| CN115298598A (en) | Thin waveguide wavelength selective projector | |
| US20220086372A1 (en) | Multi-Modal Computational Imaging via Metasurfaces | |
| Boo et al. | Metasurface wavefront control for high-performance user-natural augmented reality waveguide glasses | |
| US20230081415A1 (en) | Ultra-compact multilayer metasurface imaging system integrating lenses and spaces | |
| Yang et al. | A review of liquid crystal spatial light modulators: devices and applications | |
| Peng et al. | Metalens in Improving Imaging Quality: Advancements, Challenges, and Prospects for Future Display | |
| CN117454951A (en) | Optical neural network device | |
| US11815853B2 (en) | PVH in-band chromatic correction using metasurface | |
| KR20230020336A (en) | Meta-lens, imaging optics, electronic device including the same | |
| CN116670545A (en) | Folded optical path including a supersurface | |
| Zeng et al. | Generating dipole trap arrays based on dielectric metasurface at 808 nm | |
| CN117764131A (en) | Optical neural network device | |
| US20240168284A1 (en) | Holographic vr display | |
| Wang et al. | Innovative OPA-based optical chip for enhanced digital holography |
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 |