CN113805641A - Photon neural network - Google Patents

Photon neural network Download PDF

Info

Publication number
CN113805641A
CN113805641A CN202111116962.XA CN202111116962A CN113805641A CN 113805641 A CN113805641 A CN 113805641A CN 202111116962 A CN202111116962 A CN 202111116962A CN 113805641 A CN113805641 A CN 113805641A
Authority
CN
China
Prior art keywords
modulator
array
convolution kernel
neural network
light
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
Application number
CN202111116962.XA
Other languages
Chinese (zh)
Other versions
CN113805641B (en
Inventor
刘胜平
田野
李强
赵洋
王玮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
United Microelectronics Center Co Ltd
Original Assignee
United Microelectronics Center Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by United Microelectronics Center Co Ltd filed Critical United Microelectronics Center Co Ltd
Priority to CN202111116962.XA priority Critical patent/CN113805641B/en
Publication of CN113805641A publication Critical patent/CN113805641A/en
Application granted granted Critical
Publication of CN113805641B publication Critical patent/CN113805641B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06E—OPTICAL COMPUTING DEVICES; COMPUTING DEVICES USING OTHER RADIATIONS WITH SIMILAR PROPERTIES
    • G06E1/00—Devices for processing exclusively digital data
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

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

Abstract

The invention provides a photon neural network, which comprises a laser array, a modulator array, a first beam combiner, a modulator, a beam splitter, a delay line, a second beam combiner and a detector, wherein the laser array is arranged on the first beam combiner; the laser array emits a plurality of laser beams with different wavelengths; the modulator array modulates the laser output by the laser array, and loads the weight of the convolution kernel to the light emitted by the laser array; the first beam combiner combines the modulated multiple laser beams to obtain a first light beam; the modulator modulates the first light beam and loads information to be processed on the first light beam; the beam splitter splits a first light beam loaded with information to be processed to obtain a plurality of light beams with different wavelengths; the delay line enables the split beams to generate equal difference delay; the second beam combiner combines a plurality of beams after the generation of the equal difference delay to obtain a second beam; the detector receives the second light beam to realize the dislocation addition of the signals. The number of control circuits is reduced while calculation errors due to performance differences are eliminated.

Description

Photon neural network
Technical Field
The invention relates to the field of integrated photons, in particular to a photonic neural network.
Background
In recent years, artificial intelligence technology has deepened into the aspects of our social life. Especially, in the last decade, the operation scale of the artificial intelligence technology is getting larger and higher, and the dependence degree on hardware is also getting higher and higher. According to the OpenAI statistics of the organization of artificial intelligence research, the floating point calculation amount of the deep artificial neural network in 2012-2020 is increased at an amazing rate, which is doubled about every 3-4 months, and the increase rate of the Moore's law of the integrated circuit is far exceeded. Meanwhile, moore's law has gradually approached the physical limits of semiconductor technology and the limitations of microelectronic fabrication processes, and is facing the problem of failure. More than 90% of calculated amount in the artificial intelligence algorithm is matrix operation, and the photonic network is very suitable for matrix calculation. Therefore, developing the photon neural network becomes an effective solution for carrying out large-scale artificial intelligence operation. However, the photonic neural network on chip that has been proposed so far is mainly constructed based on the mach-zehnder interferometer (MZI), and when the scale of the convolution kernel (n × n) becomes large, that is, the matrix operation scale becomes large, the number of MZIs increases exponentially, which greatly increases the area of the photonic neural network. Meanwhile, each MZI in the photonic neural network needs two electrical ports to control the phase change of the MZI, and a larger number of electrical ports are needed for the photonic neural network on a large scale. In addition, each input end and each output end of the photonic neural network need to be connected with a modulator and a detector to realize the input of the signal to be processed and the output of the processed signal respectively. As the scale of the convolution kernel of the photonic neural network increases, the number of modulators and detectors also increases linearly. Signals between different ports of the photonic neural network are input/output through different modulators/detectors. Therefore, performance differences between the modulator/detector may cause different inputs/outputs to occur after the same signal passes through. Especially for high precision operations, the impact of this difference will be greater. For example, when the performance difference between modulators/detectors connected to different channels of a photonic neural network is greater than 1%, the operational accuracy of the neural network will be less than 6 bits. This greatly limits the computational accuracy of the photonic neural network.
Disclosure of Invention
The invention provides a photonic neural network, comprising: the system comprises a laser array, a modulator array, a first beam combiner, a modulator, a beam splitter, a delay line, a second beam combiner and a detector;
the laser array is used for emitting a plurality of laser beams with different wavelengths;
the modulator array is used for modulating the laser output by the laser array 1 and loading the weight of the convolution kernel to the light with different wavelengths emitted by the laser array;
the first beam combiner is used for combining the multiple lasers loaded with the weights to obtain a first light beam;
the modulator is used for modulating the first light beam and loading information to be processed on the first light beam;
the beam splitter is used for splitting the first light beam loaded with the information to be processed to obtain a plurality of light beams with different wavelengths;
the delay line is used for enabling the split beams to generate equal difference delay;
the second beam combiner is used for combining the beams after the equal difference time delay to obtain a second beam;
the detector is used for receiving the second light beam and realizing the dislocation addition of signals.
Optionally, the method further comprises: a phase shifter;
the phase shifter is arranged between the delay line and the second beam combiner;
the phase shifter is used for changing the group refractive index n of the waveguidegA change in the delay amount in the delay line, i.e., a change in the arithmetic delay, is caused.
Optionally, the method further comprises: optical wire bonding;
the optical wire bonding is used for connecting the modulator array and the first beam combiner;
and the multiple beams of light modulated by the modulator array enter the first beam combiner through the optical lead bonding.
OptionallyThe delay tolerance delta tau of the equal-difference delay and the bandwidth BW of the modulatormIn inverse proportion;
the length of the delay line is proportional to the speed of light c and the delay tolerance, and is proportional to the group refractive index n of the waveguidegIn inverse proportion.
Optionally, the number of weights in the channel number convolution kernel of the laser array 1; for an n × n convolution kernel, the number of channels is n2(ii) a For a convolution kernel of n × m, the number of channels is nm, where n and m are integers arbitrarily greater than or equal to 1.
Optionally, the first beam combiner, the beam splitter, or the second beam combiner is formed by: array waveguide grating, multimode interference coupler or broadband directional coupler.
Optionally, for an n × n convolution kernel, the operating bandwidth BW of the modulator arrayeOperating bandwidth BW with said modulator 4mSatisfies the following formula:
BWm≥N×n2×BWewherein, N represents the operation times of a convolution kernel in an artificial intelligence architecture, and N represents the size of the convolution kernel;
for an n x m convolution kernel, the operating bandwidth BW of the modulator arrayeOperating bandwidth BW with said modulatormSatisfies the following formula:
BWm≥N×n×m×BWewherein, N represents the operation times of the convolution kernel in the artificial intelligence architecture, and N × m represents the size of the convolution kernel.
Optionally, typical values for the central wavelength separation δ λ of the laser array and the modulator array are between 0.4nm and 4.8nm, satisfying the following formula:
δλ<ΔΛ/n2where n represents the size of the convolution kernel and Λ represents the wavelength operating range of the modulator.
Optionally, the typical value of the wavelength interval of the laser array is between 0.2nm and 4.8nm, the typical value of the wavelength interval of the first beam combiner is between 0.2nm and 4.8nm, the typical value of the wavelength interval of the beam splitter is between 0.2nm and 4.8nm, and the typical value of the wavelength interval of the second beam combiner is between 0.2nm and 4.8 nm.
Optionally, the laser array and the modulator array employ iii-v semiconductor materials; the first beam combiner, the modulator, the beam splitter, the delay line, the second beam combiner and the detector are made of silicon-based or carbon-based substrate materials.
The photonic neural network provided in the embodiment of the application comprises a laser array, a modulator array, a first beam combiner, a modulator, a beam splitter, a delay line, a second beam combiner and a detector; the laser array is used for emitting a plurality of laser beams with different wavelengths; the modulator array is used for modulating the laser output by the laser array and loading the weight of the convolution kernel to the light with different wavelengths emitted by the laser array; the first beam combiner is used for combining the multiple lasers loaded with the weights to obtain a first light beam; the modulator is used for modulating the first light beam and loading information to be processed on the first light beam; the beam splitter is used for splitting the first light beam loaded with the information to be processed to obtain a plurality of light beams with different wavelengths; the delay line is used for generating equal difference delay for the split beams; the second beam combiner is used for combining a plurality of beams of light after the generation of the equal difference time delay to obtain a second light beam; the detector is used for receiving the second light beam and realizing the dislocation addition of signals.
According to one or more technical schemes provided in the embodiment of the application, information to be processed can be loaded on a plurality of beams of light beams through one modulator, signals on the plurality of beams of light beams are received through one detector, and dislocation addition is completed. The method avoids a plurality of control circuits required by a plurality of modulators and a plurality of detectors, reduces the number of the control circuits, and simultaneously eliminates the calculation error caused by the performance difference among a plurality of modulators and the performance difference among a plurality of detectors.
Drawings
Further details, features and advantages of the disclosure are disclosed in the following description of exemplary embodiments, taken in conjunction with the accompanying drawings, in which:
FIG. 1 shows a schematic diagram of a photonic neural network according to an exemplary embodiment of the present invention;
FIG. 2 shows a schematic diagram of a convolution kernel according to an exemplary embodiment of the present invention;
fig. 3 shows a schematic diagram of a misalignment addition calculation according to an exemplary embodiment of the present invention.
Detailed Description
Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While certain embodiments of the present disclosure are shown in the drawings, it is to be understood that the present disclosure may be embodied in various forms and should not be construed as limited to the embodiments set forth herein, but rather are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the disclosure are for illustration purposes only and are not intended to limit the scope of the disclosure.
It should be understood that the various steps recited in the method embodiments of the present disclosure may be performed in a different order, and/or performed in parallel. Moreover, method embodiments may include additional steps and/or omit performing the illustrated steps. The scope of the present disclosure is not limited in this respect.
The term "include" and variations thereof as used herein are open-ended, i.e., "including but not limited to". The term "based on" is "based, at least in part, on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Relevant definitions for other terms will be given in the following description. It should be noted that the terms "first", "second", and the like in the present disclosure are only used for distinguishing different devices, modules or units, and are not used for limiting the order or interdependence relationship of the functions performed by the devices, modules or units.
It is noted that references to "a", "an", and "the" modifications in this disclosure are intended to be illustrative rather than limiting, and that those skilled in the art will recognize that "one or more" may be used unless the context clearly dictates otherwise.
The names of messages or information exchanged between devices in the embodiments of the present disclosure are for illustrative purposes only, and are not intended to limit the scope of the messages or information.
An embodiment of the present invention provides a photonic neural network, as shown in fig. 1, including: the system comprises a laser array 1, a modulator array 2, a first beam combiner 3, a modulator 4, a beam splitter 5, a delay line 6, a second beam combiner 7 and a detector 8.
The laser array 1 is used for emitting a plurality of laser lights different in wavelength, and the laser array portion is used as a light source of a photonic neural network, in which the number of laser lights is equal to the number of weights in the convolution kernel.
The modulator array 2 is used for modulating the laser output by the laser array 1, and loading the weight of the convolution kernel to the light with different wavelengths emitted by the laser array 1 to realize a matrix network in the photonic neural network. The laser array 1 and the modulator array 2 are adopted to reduce the overall volume of the device. Illustratively, the modulator array loads the laser light from the laser array with weights from the convolution kernel, each beam of light modulated by the modulator array 2 including a weight from the convolution kernel. The weights in the convolution kernels may be loaded regularly according to the wavelength of the light beam emitted by the laser array 1, for example, the light beam may be loaded with the corresponding convolution kernel weights according to the order of the wavelength from large to small or from small to large, and the embodiment does not limit the specific method for loading the weights.
To ensure that each beam of light emitted in laser array 1 is modulated, the number of channels in laser array 1 is equal to the number of channels in modulator array 2.
The modulator array 2 may be any modulator array capable of modulating light, including but not limited to an electro-absorption modulator, an electro-optical modulator, a thermo-optical modulator, an acousto-optical modulator, and the like, and the present embodiment does not limit the kind of the modulator array 2.
The first beam combiner 3 is configured to combine the multiple laser beams loaded with the weights to obtain a first light beam. After passing through the first beam combiner 3, the multiple beams of laser with convolution kernel weight information are combined into a first beam, which can be modulated by a modulator, and information to be processed is loaded on the beam. Modulating each beam of light by multiple modulators is avoided because of errors introduced by differences in modulator performance. The first beam combiner 3 is used for combining a plurality of beams, illustratively, the first beam combiner includes, but is not limited to, an arrayed waveguide grating, a multimode interference coupler, a cascaded broadband directional coupler, and the like, and the present embodiment does not limit the specific type of the first beam combiner 3. In order to ensure that each beam is combined, the number of channels of the first beam combiner 3 is the same as the number of channels of the laser array 1.
The modulator 4 is used for modulating the first light beam and loading the information to be processed on the first light beam. The information to be processed may be information such as images or sounds, and the type of the information to be processed is not limited herein. The information to be processed is loaded on the first light beam by the modulator 4, and each light beam synthesizing the first light beam is loaded with the information to be processed due to the principle of independent propagation of light, and multiplication of the convolution kernel weight and the information to be processed is completed. Illustratively, the modulator 4 includes, but is not limited to, a mach-zehnder modulator, where the modulator has a large bandwidth and can cover all wavelengths in the laser array, and the specific type of the modulator 4 is not limited herein.
In some alternative embodiments, the information to be processed is modulated onto the first beam, each of the beams that make up the first beam is modulated with all of the information to be processed. The information loaded on each beam at this time is the product of the convolution kernel weights loaded by the beam at modulator array 2 multiplied by the total information to be processed. In the convolutional neural network, only the multiplication result of the weight loaded on each beam of light and the information to be processed at a specific position needs to be added, and the position of the information to be processed corresponding to each beam of light is different, so that the information loaded on each beam of light needs to be added in a staggered manner.
The beam splitter 5 is configured to split the first light beam loaded with the information to be processed to obtain multiple light beams with different wavelengths. The first beam after passing through the modulator 4 comprises each beam emitted from the laser array 1, wherein each beam performs a multiplication of the convolution kernel weight carried by the beam with the information to be processed. The first light beam is split by the beam splitter 5, and in order to ensure that the split light and the light passing through the first beam combiner 3 correspond to each other one by one, the number of channels of the beam splitter 5 is equal to that of the channels of the first beam combiner 3. Illustratively, the beam splitter 5 may be an arrayed waveguide grating, a multimode interference coupler, or a cascaded broadband directional coupler, etc., and the specific type of the beam splitter 5 is not limited herein.
The delay line 6 is used for generating equal-difference delay for the plurality of split beams of light, wherein the generation of the equal-difference delay means that the delay amount of each beam of light is different, but the delay amount is in an equal-difference array, and the delay amounts are used for staggering the information loaded on each beam of light, so that the information detected by the detector at the same moment is just the information in the same convolution kernel. To ensure that each beam of light is delayed, the number of delay lines is equal to the number of channels in the laser array 1. Illustratively, the structure of the delay line may be a single-mode waveguide or a multi-mode waveguide, and the width and height of the waveguide may be designed according to the requirements of the photonic neural network, and the structure, width and height of the delay line are not limited herein. The waveguide structure of the delay line may include a strip waveguide, a ridge waveguide, a slab waveguide, a step waveguide, or a graded waveguide, which is not limited herein.
The second beam combiner 7 is used for combining the beams after the generation of the equal difference time delay to obtain a second beam; an equal-difference time delay is generated among the plurality of light beams synthesizing the second light beam. Illustratively, the second beam combiner 7 includes, but is not limited to, an arrayed waveguide grating, a multimode interference coupler, a cascaded broadband directional coupler, or the like, and the kind of the beam combiner is not limited herein.
The detector 8 is used for receiving the second light beam and realizing the dislocation addition of signals. Most of the delayed lights with different wavelengths in the second light beam irradiate the detector 8, and the detector completes addition operation on the light signals received at each moment, and because the lights with different wavelengths generate equal difference time delay, the detector can complete the addition operation of the information in the convolution kernel, and the function of the whole convolution neural network is realized. The detector may be a photodetector, for example, and the type of detector is not limited herein.
Exemplarily, as shown in fig. 2 and 3, the offset adding process of the exemplary embodiment of the present invention is as follows: fig. 2 shows a part of information to be processed, the thickened boxes in fig. 2 represent convolution kernels in a convolution neural network, numbers in the convolution kernels only represent the positions of the part of information in the convolution kernels, and do not represent the information included in the part of information. FIG. 3 shows the process of photonic neural network dislocation summation, λ1-λ9The nine light beams with different wavelengths are sequentially shown, the detector 8 only receives information on a number of columns at the same time and adds the received information, a vertical frame in the graph 3 corresponds to a convolution kernel in the graph 2, and the detector 8 receives the information in the vertical frame in the graph 3 at a certain time and adds the received information, so that the addition operation of the content in the convolution kernel is completed, and the function of the convolution neural network is realized. The numerals in fig. 3 only indicate the correspondence between the position and the information in the convolution kernel in fig. 2, and do not indicate the information included in the part.
By the photon neural network provided by the embodiment, the extraction of the information characteristics to be processed can be completed through one detector and one modulator, and the function of the convolutional neural network is realized. The problems of large area and low integration level of the photonic neural network caused by a plurality of control loops required by a plurality of modulators are solved. Having only one detector and modulator at the same time also prevents systematic errors introduced due to performance differences between the individual modulators and/or detectors. The photonic neural network with high integration level and high calculation precision can be realized.
In some optional embodiments, in order to ensure the accuracy of the equal-difference delay generated between the light beams and prevent information in a convolution kernel which is not irradiated onto the detector at the same time, the convolution neural network may further include a phase shifter 9; the phase shifter is arranged between the delay line 6 and the second beam combiner 7; the phase shifter 9 is used to change the group refractive index n of the waveguidegCausing a change in the amount of delay in the delay line. The delay amount among each beam can be controlled to be strictly equal through the change of the phase shifter to the delay amount in the delay line, and accurate equal-difference delay is generated among a plurality of beams. The phase shifter 9 may include thermo-optic shiftingA phase shifter, an electro-optic phase shifter, or an acousto-optic phase shifter, etc., and the specific type of the phase shifter is not limited herein.
In some alternative embodiments, the weights of the convolution kernels may also be represented by the light intensity of the light beam. Illustratively, modulator array 2 may load convolution kernel weight information by varying the intensity of the light beam emitted by laser array 1. For example, the first light intensity represents that the weight of the convolution kernel is 1, when the weight represented by the light beam is 0.5, the light intensity of the light beam is adjusted to be half of the first light intensity, and when the weight represented by the light beam is 3, the light intensity of the light beam can be adjusted to be 3 times of the first light intensity. This description is merely illustrative of one possible embodiment and does not limit the specific modulation scheme of the modulator array 2. Illustratively, the convolution kernel weight information may also be represented by changing the phase, polarization state, etc. of the light.
In some optional embodiments, to simplify the installation process of the optical system, an optical wire bonding 10; the optical wire bonding 10 is used for connecting the modulator array 2 and the first beam combiner 3; the multiple beams of light modulated by the modulator array 2 enter the first beam combiner 3 through the optical wire bonding 10. Optical wire bonding is an effective way to achieve monolithic integration between devices on substrates of different materials, connecting devices on two substrates of different materials together by 3D printing of waveguide structures. Compared with other hybrid integration technologies, the scheme is lower in cost and higher in yield. In the optional embodiment, the hybrid integration of devices among different substrates can be effectively realized, the manufacturing cost is reduced, and the yield is improved.
In some alternative embodiments, the delay tolerance Δ τ of the equipotent delay is related to the bandwidth BW of the modulatormIn inverse proportion, i.e. Δ τ is 1/BWmOtherwise, the signals are misplaced when added, and the identification accuracy is reduced. To ensure that the delay line 6 can provide an accurate delay amount, the length Δ L of the delay line 6 can be proportional to the speed of light c and the delay tolerance Δ τ, and the group refractive index n of the material used for the delay line 6gIn inverse proportion, i.e. Δ L ═ c Δ τ/ngIt is ensured that the light can generate accurate time delay after passing through the corresponding time delay line 6.
In some alternative embodiments, the number of channels of laser array 1 is equal to the number of weights in the convolution kernel; for an n × n convolution kernel, the number of channels is n2(ii) a For a convolution kernel of n × m, the number of channels is nm, where n and m are integers arbitrarily greater than or equal to 1. For example, for a square convolution kernel, the scale of the convolution kernel is 3 and the number of channels in the laser array is 9. For a non-square convolution kernel with unequal length and width, n is 3 and m is 4, the number of channels in the laser array is 12. Each beam emitted by the laser represents a weight in a convolution kernel, which in a convolutional neural network is mostly a matrix of squares.
In some alternative embodiments, for an n × n convolution kernel, the following equation is satisfied between the operating bandwidth BWe of modulator array 2 and the operating bandwidth BWm of modulator 4:
BWm≥N×n2xBWe, where N represents the number of operations of a convolution kernel in the artificial intelligence calculation process, and N represents the size of the convolution kernel.
For an n × m convolution kernel, the operating bandwidth BW of the modulator array 2eOperating bandwidth BW with modulator 4mSatisfies the following formula:
BWm≥N×n×m×BWewherein, N represents the operation times of the convolution kernel in the artificial intelligence architecture, and N × m represents the size of the convolution kernel.
Because the artificial intelligence model is that a convolution kernel and all information in the picture are gradually convoluted, the loading frequency of the information to be processed on the chip is higher than the change frequency of the weight in the convolution kernel. If the loading frequency of the information to be processed is less than the weight change frequency, the accuracy is reduced or the bandwidth in operation is wasted, that is, the operation bandwidth is less than the actual bandwidth of the device.
In some alternative embodiments, the typical value of the center wavelength separation δ λ between the laser array 1 and the modulator array 2 is between 0.4nm and 4.8nm, and the specific size satisfies the following formula:
δλ<ΔΛ/n2where n denotes the size of the convolution kernel and Λ denotes the wavelength operating range of the modulator 4.
If the wavelength interval is too large, the working bandwidth of the modulator 4 is required to be wide enough, otherwise, the information to be processed is loaded to different wavelengths to be different, and the network accuracy is reduced.
In some alternative embodiments, typical values for the wavelength interval of the laser array 1 are between 0.2nm-4.8nm, typical values for the wavelength interval of the first beam combiner 3 are between 0.2nm-4.8nm, typical values for the wavelength interval of the beam splitter 5 are between 0.2nm-4.8nm, and typical values for the wavelength interval of the second beam combiner 7 are between 0.2nm-4.8 nm.
The typical values are related to the current manufacturing capability of the device, and the values are determined based on the current process manufacturing capability. If the wavelength interval is too large, the working bandwidth of the modulator 4 is required to be wide enough, otherwise, the information to be processed is loaded to different wavelengths, and different results occur, so that the network accuracy is reduced.
In order to ensure the accuracy of the operation result of the photonic neural network, the output optical power between the channels of the laser array 1 and the characteristics of the modulator array 2 need to be tested and calibrated in advance. The operating bandwidth of the detector 8 is equal to the operating bandwidth of the modulator 4 and the wavelength operating range of the detector 8 should be greater than or equal to the wavelength operating range of the modulator 4. The center wavelength of the laser array 1 coincides with the center wavelength of the modulator array 2. The central wavelengths of the first beam combiner 3, the beam splitter 5 and the second beam combiner 7 are consistent with the central wavelength of the laser array.
In some alternative embodiments, the laser array 1 and the electro-absorption modulator array 2 may be fabricated using iii-v semiconductor materials; the III-V group semiconductor material is gallium arsenic or indium phosphorus or the III-V group semiconductor material matched with one of the gallium arsenic and the indium phosphorus in lattice. The modulator 4, delay line 6, detector 8 and phase shifter 9 may be fabricated using silicon-based or carbon-based substrate materials. This example merely shows an alternative embodiment, and does not limit the fabrication and materials of the optoelectronic device.
In some alternative embodiments, the optical devices of two different substrates may be connected together in a hybrid integration manner to form a photonic neural network. This embodiment only represents an optional implementation manner, and does not limit the integration manner of the photonic neural network.
In summary, the photonic neural network in the embodiment of the present invention only needs one set of modulator/detector to respectively realize input and output of signals, so that the calculation accuracy of the photonic neural network is improved. Meanwhile, the matrix operation part in the photonic neural network is realized by a scheme of combining a modulator and a delay line, and the number of the electrical ports is not more than 1/n of the matrix structure formed by MZI; the overall area of the network is not more than 1/n of the photonic neural network formed by the MZI.
Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations fall within the scope defined by the appended claims.

Claims (10)

1. A photonic neural network, comprising:
the device comprises a laser array (1), a modulator array (2), a first beam combiner (3), a modulator (4), a beam splitter (5), a delay line (6), a second beam combiner (7) and a detector (8);
the laser array (1) is used for emitting a plurality of laser beams with different wavelengths;
the modulator array (2) is used for modulating the laser output by the laser array (1) and loading the weight of the convolution kernel to the laser with different wavelengths emitted by the laser array (1);
the first beam combiner (3) is used for combining the multiple lasers loaded with the weights to obtain a first light beam;
the modulator (4) is used for modulating the first light beam and loading information to be processed on the first light beam;
the beam splitter (5) is used for splitting the first light beam loaded with the information to be processed to obtain a plurality of light beams with different wavelengths;
the delay line (6) is used for generating equal-difference delay for the split multiple beams of light;
the second beam combiner (7) is used for combining a plurality of beams after the generation of the equal difference time delay to obtain a second beam;
the detector (8) is used for receiving the second light beam and realizing the dislocation addition of signals.
2. The photonic neural network of claim 1, further comprising: a phase shifter (9);
the phase shifter (9) is arranged between the delay line (6) and the second beam combiner (7);
the phase shifter (9) is used for changing the group refractive index n of the waveguidegCausing a change in the amount of delay in the delay line.
3. The photonic neural network of claim 1, further comprising: an optical wire bond (10);
the optical wire bonding (10) is used for connecting the modulator array (2) and the first beam combiner (3);
the multiple beams of light modulated by the modulator array (2) enter the first beam combiner (3) through the optical wire bonding (10).
4. The photonic neural network of claim 1,
the delay tolerance delta tau of the equal-difference delay and the bandwidth BW of the modulatormIn inverse proportion;
the length of the delay line (6) is proportional to the speed of light c and the delay tolerance delta tau and to the group refractive index n of the waveguidegIn inverse proportion.
5. The photonic neural network according to claim 1, wherein the number of channels of the laser array (1) is equal to the number of weights in the convolution kernel; for an n × n convolution kernel, the number of channels is n2(ii) a For a convolution kernel of n × m, the number of channels is nm, where n and m are integers arbitrarily greater than or equal to 1.
6. The photonic neural network according to claim 1, wherein the first combiner (3), the beam splitter (5) or the second combiner (7) is constituted by:
arrayed waveguide grating, multimode interference coupler or broadband directional coupler.
7. The photonic neural network according to claim 1, wherein the operational bandwidth BW of the modulator array (2) is such that for an n x n convolution kerneleAn operating bandwidth BW with the modulator (4)mSatisfies the following formula:
BWm≥N×n2×BWewherein, N represents the operation times of the convolution kernel in the artificial intelligence architecture, and N multiplied by N represents the size of the convolution kernel;
for a convolution kernel of n x m, the operating bandwidth BW of the modulator array (2)eAn operating bandwidth BW with the modulator (4)mSatisfies the following formula:
BWm≥N×n×m×BWewherein, N represents the operation times of the convolution kernel in the artificial intelligence architecture, and N × m represents the size of the convolution kernel.
8. The photonic neural network according to claim 1, wherein typical values of the central wavelength separation δ λ of the laser array (1) and the modulator array (2) are between 0.4nm-4.8nm, satisfying the following formula:
δλ<ΔΛ/n2where n × n represents the size of the convolution kernel and Λ represents the wavelength operating range of the modulator (4).
9. The photonic neural network according to claim 1, characterized in that a typical value of the wavelength interval of the laser array (1) is between 0.2nm-4.8 nm; typical values of the wavelength interval of the first combiner (3) are between 0.2nm and 4.8 nm; typical values of the wavelength interval of the beam splitter (5) are between 0.2nm and 4.8 nm; typical values for the wavelength interval of the second combiner (7) are between 0.2nm and 4.8 nm.
10. The photonic neural network of any one of claims 1 to 9, wherein the laser array (1) and the modulator array (2) employ a iii-v semiconductor material; the first beam combiner (3), the modulator (4), the beam splitter (5), the delay line (6), the second beam combiner (7) and the detector (8) are made of silicon-based or carbon-based substrate materials.
CN202111116962.XA 2021-09-23 2021-09-23 Photonic neural network Active CN113805641B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111116962.XA CN113805641B (en) 2021-09-23 2021-09-23 Photonic neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111116962.XA CN113805641B (en) 2021-09-23 2021-09-23 Photonic neural network

Publications (2)

Publication Number Publication Date
CN113805641A true CN113805641A (en) 2021-12-17
CN113805641B CN113805641B (en) 2023-08-01

Family

ID=78896480

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111116962.XA Active CN113805641B (en) 2021-09-23 2021-09-23 Photonic neural network

Country Status (1)

Country Link
CN (1) CN113805641B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114819132A (en) * 2022-06-27 2022-07-29 之江实验室 Photon two-dimensional convolution acceleration method and system based on time-wavelength interleaving
CN115222035A (en) * 2022-09-20 2022-10-21 之江实验室 Photon neural network convolution acceleration chip
CN115508958A (en) * 2022-10-08 2022-12-23 深圳中科天鹰科技有限公司 Photon chip based on optical neural network
CN115508958B (en) * 2022-10-08 2024-05-24 深圳中科天鹰科技有限公司 Photonic chip based on optical neural network

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080154815A1 (en) * 2006-10-16 2008-06-26 Lucent Technologies Inc. Optical processor for an artificial neural network
US20190019100A1 (en) * 2017-07-11 2019-01-17 Charles ROQUES-CARMES Optical Ising Machines and Optical Convolutional Neural Networks
CN111860822A (en) * 2020-07-20 2020-10-30 联合微电子中心有限责任公司 All-optical nonlinear activation function implementation method and device of optical neural network
CN112001487A (en) * 2020-07-20 2020-11-27 联合微电子中心有限责任公司 Photon neural network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080154815A1 (en) * 2006-10-16 2008-06-26 Lucent Technologies Inc. Optical processor for an artificial neural network
US20190019100A1 (en) * 2017-07-11 2019-01-17 Charles ROQUES-CARMES Optical Ising Machines and Optical Convolutional Neural Networks
CN111860822A (en) * 2020-07-20 2020-10-30 联合微电子中心有限责任公司 All-optical nonlinear activation function implementation method and device of optical neural network
CN112001487A (en) * 2020-07-20 2020-11-27 联合微电子中心有限责任公司 Photon neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
申玉霞;李飞;: "基于优化神经网络的光纤激光器的最优设计", 激光杂志, no. 02 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114819132A (en) * 2022-06-27 2022-07-29 之江实验室 Photon two-dimensional convolution acceleration method and system based on time-wavelength interleaving
CN115222035A (en) * 2022-09-20 2022-10-21 之江实验室 Photon neural network convolution acceleration chip
CN115222035B (en) * 2022-09-20 2022-12-30 之江实验室 Photon neural network convolution acceleration chip
CN115508958A (en) * 2022-10-08 2022-12-23 深圳中科天鹰科技有限公司 Photon chip based on optical neural network
CN115508958B (en) * 2022-10-08 2024-05-24 深圳中科天鹰科技有限公司 Photonic chip based on optical neural network

Also Published As

Publication number Publication date
CN113805641B (en) 2023-08-01

Similar Documents

Publication Publication Date Title
US11269179B2 (en) Photonic filter bank system and method of use
US10656336B1 (en) Method for phase-based photonic computing
US11656485B2 (en) Photonic bandgap phase modulator, optical filter bank, photonic computing system, and methods of use
Zah et al. Multiwavelength DFB laser arrays with integrated combiner and optical amplifier for WDM optical networks
CN113805641B (en) Photonic neural network
JP4436451B2 (en) Optical signal amplification 3-terminal device
WO2021007302A1 (en) Systems and methods for parallel photonic computing
KR20230129447A (en) Balanced photonic architecture for matrix computation
WO2022016894A1 (en) Photonic neural network
CN112101540B (en) Optical neural network chip and calculation method thereof
Kebort et al. Monolithic four-channel (QUAD) integrated widely tunable transmitter in indium phosphide
US11143822B1 (en) Optical filter system and method of operation
JPWO2006011262A1 (en) Optical signal amplification 3-terminal device
CN113630207A (en) Photon neural network chip
US20230142781A1 (en) Photonic Ising Compute Engine with An Optical Phased Array
US11500410B1 (en) System and method for parallel photonic computation
US20210409144A1 (en) Optical filter system and method of operation
CN112232487B (en) Optical neural network chip and calculation method thereof
Shi et al. Emulation and modelling of semiconductor optical amplifier-based all-optical photonic integrated deep neural network with arbitrary depth
CN112001487B (en) Photonic neural network
WO2023091675A1 (en) Optical neural network with gain from parity time optical couplers
WO2023038637A1 (en) Optical filter system and method of operation
CN115508958A (en) Photon chip based on optical neural network
CN115513774A (en) Multi-wavelength distributed feedback laser
Rodrigues Transreceptor Óptico Integrado Para PONs Híbridas

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