CN115481711A - Optical computing device and system - Google Patents

Optical computing device and system Download PDF

Info

Publication number
CN115481711A
CN115481711A CN202110604196.5A CN202110604196A CN115481711A CN 115481711 A CN115481711 A CN 115481711A CN 202110604196 A CN202110604196 A CN 202110604196A CN 115481711 A CN115481711 A CN 115481711A
Authority
CN
China
Prior art keywords
optical signals
module
light
feedback
group
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
Application number
CN202110604196.5A
Other languages
Chinese (zh)
Inventor
李亚节
张翔
李静伟
董晓文
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.)
Huawei Technologies Co Ltd
Original Assignee
Huawei Technologies 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 Huawei Technologies Co Ltd filed Critical Huawei Technologies Co Ltd
Priority to CN202110604196.5A priority Critical patent/CN115481711A/en
Publication of CN115481711A publication Critical patent/CN115481711A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/06Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/067Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (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)
  • Optical Modulation, Optical Deflection, Nonlinear Optics, Optical Demodulation, Optical Logic Elements (AREA)

Abstract

A light computing device and system includes an input module, a feedback module, a reservoir module, and a detection module. The input module calculates a first group of optical signals based on a first weight to obtain a second group of optical signals, and the first group of optical signals carries first data to be calculated. The feedback module is configured to load the first calculation result on the feedback light to obtain a first group of feedback light signals, where the first calculation result is a calculation result obtained by processing second data through the optical calculation device, and the second data is data processed before the processing of the first data. The reservoir module is configured to calculate the first set of feedback light signals and the second set of light signals based on a second weight to obtain a third set of light signals. The detection module detects the light intensity of the third group of optical signals to obtain a second calculation result.

Description

Optical computing device and system
Technical Field
The present application relates to the field of artificial intelligence, and more particularly, to an optical computing device and system.
Background
In recent years, significant advances have been made in artificial neural networks, which will serve the real world more and more extensively. The network structure of neural networks is generally divided into a feedforward network and a recursive network. Feed-forward neural networks are mainly used for static (non-temporal) data processing, since individual input data, even if given sequentially, are processed independently. Recurrent neural networks are suitable for dynamic (temporal) data processing, since they can embed the temporal dependence of the input into their dynamic behavior. Many scenes in real life are temporal, such as prediction, adaptive filtering, computing device control or recognition, noise reduction, vision, and speech. With the advent of the big data era, dynamic signal processing will greatly reduce the overhead of storing and transmitting large data sets and the need for post-processing. A Reservoir Computation (RC) network in a recurrent neural network is suitable for various classification and prediction tasks.
The RC network is mainly characterized in that the input weight and the reservoir weight are fixed, only the output weight needs to be trained, the training time is greatly shortened, the training difficulty is reduced, and multiple tasks can be executed simultaneously. For complex tasks, the optical RC neural network needs to store the weight coefficients of a large matrix, and has high memory requirements, low computation speed, and high power consumption. In the optical neural network, the weight coefficient is an inherent part of the photon calculation device, information is transmitted at the optical speed, and the optical neural network is high in speed, large in bandwidth and low in power consumption.
In order to better realize the optical reservoir calculation, an optical calculation device with a simple calculation device structure, a large number of nodes, a wide-range adjustable topological structure and a small number of active devices is urgently needed.
Disclosure of Invention
The embodiment of the application provides an optical computing device and an optical computing system, which are used for realizing that the topological structure of the optical computing device can be adjusted in a large range under the conditions of simple structure and large manufacturing tolerance of the computing device.
In a first aspect, an embodiment of the present application provides an optical computing apparatus, which is applied to an optical RC network and used for implementing tasks such as classification and prediction. The method mainly comprises the following steps: the system comprises an input module, a feedback module, at least one reservoir module and a detection module. The input module may calculate a second set of optical signals based on the first weight after receiving the first set of optical signals, where the first set of optical signals is generated based on the first data to be calculated. The feedback module is configured to obtain a first group of feedback optical signals based on a first calculation result, where the first calculation result is a calculation result of second data, and the second data is data processed before the first data is processed. The reservoir module is used for receiving the second group of optical signals and the first group of feedback optical signals, and then calculating the first group of feedback optical signals and the second group of optical signals based on second weight to obtain a third group of optical signals; wherein the number of reservoir modules may be at least one. The detection module is used for detecting the light intensity of the third group of optical signals to obtain a second calculation result.
In the present application, the reservoir modules may be cascaded, i.e. the number of reservoir modules may be multiple. In an exemplary scenario, assuming that the optical computing device includes a reservoir module 1 and a reservoir module 2, the connection relationship may be as follows: the reservoir module 1 comprises at least one input port, at least one free propagation zone, at least one phase modulator and at least one output port. Wherein the at least one input port is configured to receive the second set of optical signals and the first set of feedback optical signals, and then the at least one phase modulator and the at least one free propagation region modulate and output the second set of optical signals and the first set of feedback optical signals as a set of intermediate optical signals; the reservoir module 2 comprises at least one input port, at least one free propagation zone, at least one phase modulator and at least one output port. Wherein the at least one input port is configured to receive the intermediate optical signal output by the reservoir module 1, and then the at least one phase modulator and the at least one free propagation region modulate the intermediate optical signal output by the reservoir module 1 and output the modulated intermediate optical signal as the third set of optical signals.
In this embodiment, the input module modulates the first group of optical signals based on the first weight, so that the first group of optical signals carrying data is subjected to weight adjustment at the input module to obtain a second group of optical signals. The second set of light signals and the first set of feedback light are modulated at the reservoir module based on the second weight to obtain a third set of light signals. Therefore, in the optical calculation process, the weight of the optical signal carrying data is divided into two sections for modulation, and the two weights are not affected with each other, so that the flexibility of weight adjustment of the optical signal is increased, and the large-range adjustment of the topological structure of the optical calculation device is realized. Meanwhile, the feedback module can modulate the feedback light according to the calculation result of the previous calculation process, so that the iterative modulation of the feedback light can be completed, and the convergence speed of the output weight in the light calculation device is accelerated.
In a possible implementation manner, the optical computing apparatus further includes an electrical processing module, and the electrical processing module feeds back the second calculation result to the feedback module, so that the optical computing apparatus modulates the feedback light at the next time according to the second calculation result, and thus, the iterative modulation of the feedback light is implemented.
In this embodiment, the light computing device may use different structures to implement the above functions, specifically including the following possible implementations:
in one possible implementation, the optical computing device utilizes a multi-mode interferometric structure to achieve the above described functionality. The reservoir module comprises at least one phase modulator and a first multimode interference structure, wherein the at least one phase modulator is connected with at least one input waveguide of the first multimode interference structure, namely one phase modulator is connected with one input waveguide, but not limited to that, each input waveguide is connected with one phase modulator; the at least one phase modulator is configured to adjust phases of the first set of feedback optical signals and the second set of optical signals based on the second weight to obtain a plurality of first intermediate optical signals; the first multimode interference structure is then configured to receive the plurality of first intermediate optical signals and output the plurality of first intermediate optical signals as the third set of optical signals based on slab waveguide transmission.
The input module comprises a second multimode interference structure and a plurality of first modulators, wherein the plurality of first modulators are connected with a plurality of output waveguides of the second multimode interference structure, namely, one first modulator is connected with one output waveguide of the second multimode interference structure, but not limited to that, each output waveguide is connected with one modulator; then the second multimode interference structure receives the first set of optical signals, wherein the first set of optical signals includes a plurality of optical signals of different wavelengths, and splits the first set of optical signals into a plurality of sets of second intermediate optical signals, wherein each set of second intermediate optical signals includes the plurality of optical signals of different wavelengths; the plurality of first modulators are configured to modulate the plurality of sets of second intermediate optical signals based on the first weight to obtain the second set of optical signals.
The feedback module includes a plurality of second modulators for loading the first calculation result on feedback light to obtain the first set of feedback light signals. Wherein the input port of the feedback module is connected to the output waveguide of the input module, and a modulator of the feedback module is connected to an input port of the feedback module.
Based on the above scheme, the second multimode interference structure comprises a 1 × N multimode interference structure or is an N × N multimode interference structure, and the first multimode interference structure is an N × N multimode interference structure.
In the technical scheme, the input module and the modulators in the reservoir module can adjust the modulation states thereof according to requirements to obtain different weight values, so that the optimization solution of various data sets is met. Meanwhile, in order to obtain more random weight values, the multimode interference structures serving as the input module and the reservoir module can split light unevenly, so that the tolerance of the optical computing device in the manufacturing process is larger.
In another possible implementation, the optical computing device utilizes an arrayed waveguide grating to achieve the above-described functionality. The input module comprises a plurality of third modulators and a first arrayed waveguide grating, wherein the plurality of third modulators are connected with a plurality of input waveguides of the first arrayed waveguide grating, namely one third modulator is connected with one input waveguide of the first arrayed waveguide grating; then the plurality of third modulators are configured to receive the first set of optical signals and modulate the first set of optical signals based on the first weight to obtain a plurality of third intermediate optical signals; and then the first arrayed waveguide grating receives the plurality of third intermediate optical signals and transmits and outputs the third intermediate optical signals as the second group of optical signals based on the slab waveguide.
The reservoir module comprises at least one phase modulator and a second arrayed waveguide grating, wherein the at least one phase modulator is located on the arrayed waveguide of the second arrayed waveguide grating, that is, one phase modulator is located on one arrayed waveguide of the second arrayed waveguide grating, but not limited to one phase modulator on each arrayed waveguide of the second arrayed waveguide grating; the second arrayed waveguide grating is used for receiving the first group of feedback optical signals and the second group of optical signals and outputting the first group of feedback optical signals and the second group of optical signals as a plurality of fourth intermediate optical signals based on slab waveguide transmission; the at least one phase modulator is configured to modulate a phase of the plurality of fourth intermediate optical signals based on the second claim to obtain a plurality of fifth intermediate optical signals; the second arrayed waveguide grating is also configured to receive the fifth intermediate optical signal and output the fifth intermediate optical signal as the third set of optical signals based on slab waveguide transmission.
The feedback module comprises a plurality of fourth modulators and a third arrayed waveguide grating, wherein the plurality of fourth modulators are connected with a plurality of input waveguides of the third arrayed waveguide grating, namely, one fourth modulator is connected with one input waveguide in the third arrayed waveguide grating, and each input waveguide of the third arrayed waveguide grating is connected with one fourth modulator. The plurality of fourth modulators receive the first set of modulated signals and output the first set of modulated signals as the first set of feedback optical signals based on slab waveguide transmission.
In the technical scheme, the input module and the modulators in the reservoir module can adjust the modulation states of the modulators according to requirements to obtain different weight values, so that the optimization solution of various data sets is met. Meanwhile, in order to obtain more random weight values, the array waveguide gratings serving as the input module and the reservoir module have large insertion loss unevenness to different wavelengths, and can accept large optical crosstalk among channels, so that the tolerance of the optical computing device in the manufacturing process is larger.
In a possible implementation manner, the detection module includes a semiconductor optical amplifier and a semiconductor optical detector, and the semiconductor optical amplifier and the semiconductor optical detector are configured to detect light intensity of the third group of optical signals and obtain the second calculation result.
Or, the detecting module includes a semiconductor light detector, and the semiconductor light detector is configured to detect the light intensity of the third group of optical signals and obtain the second calculation result.
In one possible implementation, the light computing device further includes an array of light sources, wherein a first portion of the array of light sources is configured to receive the first data and generate the first set of light signals based on the first data; a second portion of the array of light sources is for emitting feedback light.
It is to be understood that the light source arrays transmitting the first set of light signals and the feedback light may also be independent, i.e. one light source array is used to receive the first data and generate the first set of light signals based on the first data. And the other array of light sources is used to emit feedback light.
In a possible implementation, the optical computing device further comprises a chip, and the chip controls the optical computing device to implement the above scheme.
In a second aspect, the present application provides an optical signal processing method, which specifically includes: the light calculation device acquires the first group of optical signals and calculates the first group of optical signals based on first weight to obtain a second group of optical signals; meanwhile, the optical computing device also receives feedback light, and then loads a first computing result of second data pre-processed by the first data on the feedback light to obtain a first group of feedback optical signals; then the light calculating device calculates the first group of feedback light signals and the second group of light signals based on a second weight to obtain a third group of light signals; finally, the light computing device detects the light intensity of the third group of light signals to obtain a second computing result of the first data.
In one possible implementation, if each functional module of the optical computing device is formed by a multi-mode interference structure, the optical signal is processed as follows:
an input waveguide in an input module of the optical computing device receives the first set of optical signals and the feedback light, and a modulator in the input module of the optical computing device modulates the first set of optical signals based on the first weight to generate the second set of optical signals; then the second group of optical signals and the feedback light enter a feedback module of the optical computing device, wherein the feedback module modulates the feedback light according to the first computing result to obtain a first group of feedback optical signals; the feedback module inputs the first set of feedback optical signals and the second set of optical signals to a reservoir module of the optical computing device; the reservoir module modulates the first set of feedback optical signals and the second set of optical signals based on the second weight to obtain the third set of optical signals; and simultaneously, the reservoir module transmits the third group of optical signals to a detection module of the optical computing device, and then the detection module detects the third group of optical signals to obtain a second computing result of the first data.
In one possible implementation, if each functional module of the optical computing device is formed by an arrayed waveguide grating, the optical signal is processed as follows:
a portion of input waveguides in an input module of the optical computing device receiving a first set of optical signals, a modulator in the input module modulating the first set of optical signals based on the first weight to generate a second set of optical signals; meanwhile, the other part of input waveguides in the input module of the optical computing device receive the feedback light and input the feedback light into the feedback module of the optical computing device, and the feedback module modulates the feedback light according to the first computing result to obtain a first group of feedback light signals; then the feedback module outputs the first group of feedback optical signals, the input module outputs the second group of optical signals, and the first group of feedback optical signals and the second group of optical signals are input to a reservoir module of the optical computing device after being merged by a beam combiner; the reservoir module modulates the first set of feedback optical signals and the second set of optical signals based on the second weight to obtain the third set of optical signals; and simultaneously, the reservoir module transmits the third group of optical signals to a detection module of the optical computing device, and then the detection module detects the third group of optical signals to obtain a second computing result of the first data.
In a third aspect, the present application provides a light computing system comprising a processor and the light computing apparatus described in the first aspect above. The processor is configured to input the first data to the light computing device.
It will be appreciated that the processor may also be an electrical processing module in the light computing device.
The processor mentioned in any of the above may be a general Processing Unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or one or more integrated circuits for controlling the execution of programs of the above-mentioned data transmission methods.
Drawings
FIG. 1a is an exemplary architecture diagram of a light computing device in an embodiment of the present application;
FIG. 1b is another exemplary architecture diagram of a light computing device in an embodiment of the present application;
FIG. 1c is a schematic diagram of an optical signal processing flow based on the optical computing device shown in FIG. 1 a;
FIG. 2 is a schematic diagram of one embodiment of a light computing device in an embodiment of the present application;
FIG. 3 is a schematic diagram illustrating a modulation comparison between a first set of optical signals and a feedback optical signal based on an input module and a feedback module in the optical computing device shown in FIG. 2;
FIG. 4 is a schematic diagram of another embodiment of a light computing device in an embodiment of the present application;
FIG. 5 is a schematic diagram of another embodiment of a light computing device in an embodiment of the present application;
FIG. 6 is a schematic diagram of one embodiment of a light computing system in an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the embodiments of the present application are described below with reference to the accompanying drawings, and it is obvious that the described embodiments are only some embodiments of the present application, but not all embodiments. As can be known to those skilled in the art, with the advent of new application scenarios, the technical solution provided in the embodiments of the present application is also applicable to similar technical problems.
The terms "first," "second," and the like in the description and claims of this application and in the foregoing drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It will be appreciated that the data so used may be interchanged under appropriate circumstances such that the embodiments described herein may be implemented in other sequences than those illustrated or described herein. Moreover, the terms "comprises," "comprising," and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, computing device, article, or apparatus that comprises a list of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not expressly listed or inherent to such process, method, article, or apparatus. The naming or numbering of the steps appearing in the present application does not mean that the steps in the method flow have to be executed in the chronological/logical order indicated by the naming or numbering, and the named or numbered process steps may be executed in a modified order depending on the technical purpose to be achieved, as long as the same or similar technical effects are achieved. The division of the units presented in this application is a logical division, and in practical applications, there may be another division, for example, multiple units may be combined or integrated in another computing device, or some features may be omitted, or not executed, and in addition, the shown or discussed coupling or direct coupling or communication connection between each other may be through some interfaces, and the indirect coupling or communication connection between the units may be in an electrical or other similar form, which is not limited in this application. Furthermore, the units or sub-units described as the separate parts may or may not be physically separate, may or may not be physical units, or may be distributed in a plurality of circuit units, and some or all of the units may be selected according to actual needs to achieve the purpose of the present disclosure.
The present application provides a light computing device 100 as shown in fig. 1a, the light computing device 100 comprising an input module 101, a feedback module 102, a reservoir module 103, a detection module 104, and an electrical processing module 105.
The input module 101 includes but is not limited to: at least one input port, at least one output port, at least one modulator, and at least one free propagation region. At least one input port of the input module 101 is configured to receive a first set of optical signals, where the first set of optical signals is used to carry first data; then at least one modulator in the input module 101 is used to perform intensity modulation on the first set of optical signals, and at least one free propagation region in the input module 101 is used to transmit the first set of optical signals based on the slab waveguide, and after passing through the at least one modulator and the at least one free propagation region, the first set of optical signals received by the input module 101 is output as a second set of optical signals.
The feedback module 102 includes, but is not limited to: at least one input port, at least one output port, and at least one modulator. Wherein at least one input port of the feedback module 102 receives feedback light and receives a first calculation of second data; then, at least one modulator of the feedback module 102 modulates the feedback light according to the first calculation result to obtain a first group of feedback light signals; and finally, outputting the first group of feedback optical signals through an output port. In the light computing device 100 shown in fig. 1a, the feedback light received by the feedback module 102 is input via the input module 101.
The reservoir modules 103 include, but are not limited to: at least one input port, at least one free propagation region, at least one phase modulator, and at least one output port. Wherein the at least one input port is configured to receive the second set of optical signals and the first set of feedback optical signals, and then the at least one phase modulator and the at least one free propagation region modulate and output the second set of optical signals and the first set of feedback optical signals as a third set of optical signals.
The detection module 104 includes but is not limited to: at least one input port, at least one detector, and at least one output port. The at least one input port is used for receiving the third group of optical signals, and the at least one detector is used for detecting the light intensity of the third group of optical signals, performing nonlinear calculation to obtain a third calculation result, and converting the third calculation result from the optical signals into electric signals.
The electronic processing module 105 may be a CPU, microprocessor, ASIC, FPGA, or one or more integrated circuits for controlling the execution of programs for the various data transmission methods described above. The electrical processing module 105 is mainly used to provide the second calculation result detected by the detection module 104 to the feedback module 102, and calculate an output weight, output a classification or prediction result.
It is understood that the light computing device 100 shown in FIG. 1a further comprises an array of light sources for receiving the first data and then generating the first set of light signals based on the first data. And the array of light sources may also be used to emit feedback light.
It is to be appreciated that the present application also provides a light computing device 100 as shown in FIG. 1b, the light computing device 100 comprising an input module 101, a feedback module 102, a reservoir module 103, a detection module 104, and an electrical processing module 105.
The input module 101 includes but is not limited to: at least one input port, at least one output port, at least one modulator, and at least one free propagation region. At least one input port of the input module 101 is configured to receive a first set of optical signals, where the first set of optical signals is used to carry first data; then at least one modulator in the input module 101 is used to perform intensity modulation on the first set of optical signals, and at least one free propagation region in the input module 101 is used to transmit the first set of optical signals based on slab waveguide, and after passing through the at least one modulator and the at least one free propagation region, the first set of optical signals received by the input module 101 is output as a second set of optical signals.
The feedback module 102 includes, but is not limited to: at least one input port, at least one output port, and at least one modulator. Wherein at least one input port of the feedback module 102 receives feedback light and receives a first calculation of second data; then, at least one modulator of the feedback module 102 modulates the feedback light according to the first calculation result to obtain a first group of feedback light signals; and finally, outputting the first group of feedback optical signals through an output port. In the light computing device 100 shown in fig. 1b, the feedback light received by the feedback module 102 is not input via the input module 101.
The reservoir modules 103 include, but are not limited to: at least one input port, at least one free propagation region, at least one phase modulator, and at least one output port. Wherein the at least one input port is configured to receive the second set of optical signals and the first set of feedback optical signals, and then the at least one phase modulator and the at least one free propagation region modulate and output the second set of optical signals and the first set of feedback optical signals as a third set of optical signals. In this application, the reservoir modules 103 may be cascaded, i.e., the number of the reservoir modules 103 may be plural. For example, the reservoir module 103 includes a reservoir module 1 and a reservoir module 2, and the connection relationship may be as follows: the reservoir module 1 comprises at least one input port, at least one free propagation zone, at least one phase modulator and at least one output port. Wherein the at least one input port is configured to receive the second set of optical signals and the first set of feedback optical signals, and then the at least one phase modulator and the at least one free propagation region modulate and output the second set of optical signals and the first set of feedback optical signals as a set of intermediate optical signals; the reservoir module 2 comprises at least one input port, at least one free propagation zone, at least one phase modulator and at least one output port. Wherein the at least one input port is configured to receive the intermediate optical signal output by the reservoir module 1, and then the at least one phase modulator and the at least one free propagation region modulate the intermediate optical signal output by the reservoir module 1 and output the modulated intermediate optical signal as the third set of optical signals.
The detection module 104 includes, but is not limited to: at least one input port, at least one detector, and at least one output port. The at least one input port is used for receiving the third group of optical signals, and the at least one detector is used for detecting the light intensity of the third group of optical signals, performing nonlinear calculation to obtain a third calculation result, and converting the third calculation result from the optical signals into electric signals.
The electronic processing module 105 may be a CPU, microprocessor, ASIC, FPGA, or one or more integrated circuits for controlling the execution of programs for the various data transmission methods described above. The electrical processing module 105 is mainly used for providing the second calculation result detected by the detection module 104 to the feedback module 102, and calculating an output weight, outputting a classification or prediction result.
It is understood that the light computing device 100 shown in FIG. 1b further comprises an array of light sources for receiving the first data and then generating the first set of light signals based on the first data. And the array of light sources may also be used to emit feedback light. In addition, according to the functional blocks of the system shown in fig. 1a and 1b, an exemplary process flow of the optical computing apparatus 100 for an optical signal can be as shown in fig. 1 c:
the structure shown in FIG. 1a is taken as an example to explain: inputting first data in the current calculation process; generating a first set of optical signals based on the first data; the first set of optical signals is input to the input module 101, and the input module 101 also receives feedback light; the input module 101 modulates the first set of optical signals into a second set of optical signals based on a first weight and outputs the feedback light to the feedback module 102; the feedback module 102 receives a first calculation result fed back by the electrical processing module 105, and modulates the feedback light to generate a first group of feedback light signals based on the first calculation result, where the first calculation result is a calculation result obtained in the last calculation process; the first feedback optical signal and the second set of optical signals are combined and then input into the reservoir module 103; the reservoir module 103 modulates the second set of light signals and the first set of feedback light signals into a third set of light signals based on a second weight; inputting the third set of optical signals to the detection module 104; the detection module 104 detects the light intensity of the third set of optical signals to obtain a second calculation result; the electrical processing module 105 then feeds back the second calculation result to the feedback module 102, and the second calculation result is used in the next calculation process.
The above is a description of a system functional module of the optical computing apparatus 100, and according to different implementation processes, the optical computing apparatus 100 has the following possible implementation manners:
one possible implementation is shown in fig. 2, where the input module 101 and the reservoir module 103 in the light computing device 100 are formed by multimode interference (MMI).
The MMI as an input module 101 comprises at least one input waveguide (i.e. corresponding to at least one input port of the input module 101 depicted in fig. 1a or 1 b), a slab waveguide (i.e. corresponding to the free propagation region depicted in fig. 1a or 1 b) and at least one output waveguide (i.e. corresponding to at least one output port of the input module 101 depicted in fig. 1a or 1 b), to which at least one modulator of the input module 101 is connected (i.e. one modulator is connected to one output waveguide, but there are also cases where the output waveguide is not connected to a modulator); at least one input port of the feedback module 102 is connected to at least one output waveguide of the MMI (one input port of the feedback module 102 is connected to one output waveguide of the MMI, which may be in a one-to-one correspondence), at least one modulator of the feedback module 102 is connected to at least one input port of the feedback module 102 (i.e., one modulator is connected to one input port), respectively, at least one output port of the feedback module 102 is connected to at least one input waveguide of the MMI as a reservoir module 103 (i.e., corresponding to at least one input port of the reservoir module 103 depicted in fig. 1a or fig. 1 b); the at least one phase modulator of the reservoir module 103 is connected to at least one input waveguide of the MMI; the at least one input waveguide is connected to a slab waveguide (i.e. corresponding to the free propagation region depicted in fig. 1a or fig. 1 b); at least one output waveguide of the MMI (i.e. corresponding to at least one output port of the reservoir module 103 depicted in fig. 1a or fig. 1 b) is connected to at least one input port of the probe module 104; at least one detector of the detection module 104 is connected to the at least one input port (i.e. one detector is connected to an optical path corresponding to one input port); at least one output port of the detection module 104 is connected to the electrical processing module 105.
In this embodiment, the MMI of the input module 101 may be a 1 × N MMI or an N × N MMI, and the MMI of the reservoir module 103 may be an N × N MMI. Wherein the number of modulators comprised by the input module 101, the number of modulators comprised by the feedback module 102 and the number of phase modulators of the reservoir module 103 are not more than the number of output waveguides of the MMI. And the feedback module 102 includes the same number of modulators as the number of input ports corresponding to the detector 104. The modulator of the input module 101 and the modulator of the feedback module 102 may be MZI modulators or micro-ring modulators.
Under the structure shown in fig. 2, an application scenario is specifically described: suppose that the input module 101 selects 1 × 100 MMI and 100 micro-ring modulators, i.e. each output waveguide is connected to one micro-ring modulator; the feedback module 102 selects 100 micro-ring modulators; the reservoir module 103 adopts 100 × 100 MMIs and 100 phase modulators, i.e. each input waveguide is connected with one phase modulator; the detection module 104 selects a semiconductor optical amplifier and a detector, wherein an input port, the semiconductor optical amplifier, the detector and an output port of the detection module 104 are sequentially connected; the electrical processing module 105 is an FPGA board. In this application scenario, assuming that the first data is 4-dimensional data, the 4-dimensional data at the first time is loaded to signal lights of 4 different wavelengths respectively as a first set of optical signals. If the 1 x 100 MMI is optimally designed for 1550 nanometers (nm), then the 4 different wavelengths of the first set of optical signals should be chosen around 1550nm, and randomly around the modulation wavelength of the micro-ring modulator, respectively. For example, if the modulation wavelengths of the micro-ring modulator around 1550nm are 1534nm, 1542nm, 1550nm and 1558nm, respectively, the 1 st wavelength in the first set of optical signals may be randomly selected around 1534nm, the 2 nd wavelength in the first set of optical signals may be randomly selected around 1542nm, the 3 rd wavelength in the first set of optical signals may be randomly selected around 1550nm, and the 4 th wavelength in the first set of optical signals may be randomly selected around 1558 nm. This ensures that the 4 wavelengths have random intensities when the micro-ring modulator is unbiased, which has the effect of λ as shown in FIG. 3 1 、λ 2 、λ 3 、λ 4 As shown.
At a first instant, a first set of optical signals comprising 4 different wavelengths and feedback light comprising 1 wavelength are simultaneously input at the input waveguides of the 1 × 100 MMI. The above 5 different wavelengths of light are split into 100 paths at the output waveguides of the 1 x 100 MMI, respectively, and enter the micro-ring modulators of each path. On the same path, the light intensity of a first group of optical signals comprising 4 different wavelengths is random; the bias voltages of the 100 micro-ring modulators are random, so that the light intensity of the signal light with the same wavelength on the 100 paths is random, and the modulation of the first group of optical signals is completed to obtain the second group of optical signals. In this process, the wavelength of the feedback light is not in the modulation wavelength range of the micro-ring modulator of the input module 101, so the micro-ring modulator of the input module 101 does not modulate the feedback light, as shown in fig. 3.
At a first time, the input module 101 outputs 100 channels of the second set of optical signals, and then inputs the second set of optical signals and feedback light into the input port of the feedback module 102. The feedback module 102 further receives 100 electrical signals (also referred to as a second calculation result) provided by the electrical processing module 105, wherein the second calculation result is a calculation result calculated from data before the first time; the 100 electrical signals act on the micro-ring modulator of the feedback module 102, and then the micro-ring modulator modulates the feedback light to obtain a first set of feedback light signals (i.e., 100 modulated feedback lights). The micro-ring modulator of the feedback module 102 does not modulate the second set of optical signals, as shown in fig. 3.
In this configuration, the feedback module 102 inputs the first set of feedback light signals and the second set of light signals to the reservoir module 103 simultaneously. The 100 phase modulators in the reservoir module 103 modulate the optical signal on each path and the feedback optical signal according to the modulation state to output a plurality of intermediate optical signals, then input the plurality of intermediate optical signals into the slab waveguide (i.e., the free propagation region) of the MMI for transmission, and finally output a third set of optical signals (including 100 paths of optical signals, where the optical signals include the feedback optical signal and the data-carrying optical signal) from the output waveguide of the MMI.
The detection module 104 obtains the third group of optical signals, detects the light intensity of the third group of optical signals through the semiconductor optical amplifier and the detector, and performs nonlinear calculation to obtain the calculation result of the four-dimensional data at the first time. Meanwhile, the calculation result is converted from an optical signal to an electrical signal and output to the electrical processing module 105.
In this embodiment, assuming that the 4-dimensional data is to-be-classified data, the electrical processing module 105 calculates an output weight by using the real result of the training set and the electrical signal output by the detection module 104, and then the electrical processing module 105 calculates the classification result of the 4-dimensional data according to the output weight.
In the technical scheme, the input module and the modulators in the reservoir module can adjust the modulation states thereof according to requirements to obtain different weight values, so that the optimization solution of various data sets is met. Meanwhile, in order to obtain more random weight values, the MMI serving as the input module and the reservoir module can split light unevenly, so that the tolerance of the light computing device in the manufacturing process is larger.
One possible implementation is shown in fig. 4, where the input module 101 and the reservoir module 103 of the optical computing apparatus 100 are formed by Arrayed Waveguide Gratings (AWGs).
The AWG as input module 101 comprises at least one input waveguide (i.e., corresponding to at least one input port of the input module 101 depicted in fig. 1a or 1 b), an input slab waveguide region (i.e., corresponding to the free propagation region depicted in fig. 1a or 1 b), an arrayed waveguide, and an output slab waveguide region (i.e., corresponding to the free propagation region depicted in fig. 1a or 1 b), and at least one output waveguide (i.e., corresponding to at least one output port of the input module 101 depicted in fig. 1a or 1 b), to which at least one modulator of the input module 101 is connected (i.e., one modulator is connected to one input waveguide, but there are also cases where an input waveguide is not connected to a modulator); an AWG as a feedback module 102 comprises at least one input waveguide (i.e., corresponding to at least one input port of the feedback module 102 depicted in fig. 1a or fig. 1 b), an input slab waveguide region, an array waveguide, and an output slab waveguide region, and at least one output waveguide (i.e., corresponding to at least one output port of the feedback module 102 depicted in fig. 1a or fig. 1 b), with at least one modulator of the feedback module 102 being connected to the at least one input waveguide (i.e., one modulator being connected to one input waveguide); the input module 101 output waveguide and the output waveguide of the feedback module 102 are connected to the input waveguide of the reservoir module 103; the AWG as a reservoir module 103 comprises at least one input waveguide (i.e., corresponding to at least one input port of the reservoir module 103 depicted in fig. 1a or 1 b), an input slab waveguide region (i.e., corresponding to the free propagation region depicted in fig. 1a or 1 b), an arrayed waveguide and an output slab waveguide region (i.e., corresponding to the free propagation region depicted in fig. 1a or 1 b), and at least one output waveguide (i.e., corresponding to at least one output port of the reservoir module 103 depicted in fig. 1a or 1 b), at least one phase modulator of the reservoir module 103 being located on the arrayed waveguide, wherein one phase modulator is located on one arrayed waveguide, but there are also cases where an arrayed waveguide is not connected to a phase modulator, the output waveguide of the reservoir module 103 being connected to an input port of the detection module 104; at least one detector of the detection module 104 is connected to at least one input port of the detection module 104 (i.e. one detector is connected to an optical path corresponding to one input port); at least one output port of the detection module 104 is connected to the electrical processing module 105. In the configuration shown in fig. 4, the input waveguide of the input module 101 is directly connected to the input waveguide of the feedback module 102. That is, the input module 101 includes 2R input waveguides, wherein R input waveguides are used to receive the first set of optical signals and then input the first set of optical signals to the input slab waveguide region of the AWG, and the other R input waveguides receive feedback light and input the feedback light directly to the feedback module 102.
In this scenario, the AWG as the input module 101 may be an N × 1 AWG structure, and the AWG structure as the reservoir module 103 may be a 1 × N AWG structure. Wherein the input module 101 comprises no more modulators than the number of input waveguides of the AWG structure, the feedback module 102 comprises a number of modulators equal to the number of output waveguides of the reservoir module 103, and the number of phase modulators of the reservoir module 103 is no more than the number of array waveguides of the AWG structure. The modulator of the input module 101 and the modulator of the feedback module 102 may be MZI modulators, micro-ring modulators, or electro-absorption modulators.
Under the structure shown in fig. 4, an application scenario is specifically described: suppose that the input module 101 uses 64 × 1 AWG and 64 micro-ring modulators, i.e., each input waveguide is connected to one micro-ring modulator; the feedback module 102 selects 64 × 1 AWGs and 64 micro-ring modulators; the reservoir module 103 adopts 1 × 64 AWG and 64 phase modulators, i.e. each arrayed waveguide is connected with one phase modulator; the detection module 104 selects a semiconductor optical amplifier and a detector, wherein an input port, the semiconductor optical amplifier, the detector and an output port of the detection module 104 are sequentially connected; the electrical processing module 105 is an FPGA board. In the application scenario, the first data is one-dimensional data, and the first data at the first time is simultaneously loaded to 64 signal lights with different wavelengths as a first group of optical signals. The 64 input waveguides of the 64 × 1 AWG are each inputted with the first set of 64 optical signals of different wavelengths, i.e. the 64 optical signals at the first time instant are identical. It is understood that the 64 wavelengths satisfy the wavelength design of the 64 x 1 AWG structure. The 64 micro-ring modulators in the input module 101 perform random intensity modulation on the first group of optical signals with 64 different wavelengths respectively to output 64 paths of second group of optical signals.
At the first moment, feedback light with the same wavelength as the first group of optical signals is respectively transmitted to the other 64 input waveguides of the input module 101, and the wavelength of the first group of optical signals input to the 64 input waveguides of the feedback module 102, that is, to one input waveguide of the input module 101, is λ 1 、λ 2 、λ 3 、λ 4 Then, the wavelength of the feedback light input to one input waveguide of the feedback module 102 is λ 1 、λ 2 、λ 3 、λ 4 ). The 64 input waveguides of the feedback module 102 receive 64 feedback lights, and the feedback module 102 also receives 64 electrical signals (also called "feedback light signals") provided by the electrical processing module 105A second calculation result), wherein the second calculation result is a calculation result obtained by calculating data before the first moment; the 64 electrical signals act on the micro-ring modulator of the feedback module 102, and then the micro-ring modulator modulates the 64 feedback lights to obtain a first set of feedback optical signals (i.e. 64 modulated feedback optical signals).
In this configuration, the output waveguide of the input module 101 outputs only the second set of optical signals, and the output waveguide of the feedback module 102 outputs only the first set of feedback optical signals. When the second set of optical signals and the first set of feedback optical signals are input into the reservoir module 103, the second set of optical signals and the first set of feedback optical signals need to be combined by a beam combiner and then input into the input waveguide of the reservoir module 103. The second group of optical signals and the first group of feedback optical signals pass through an input slab waveguide region to obtain a group of intermediate optical signals, 64 phase modulators in the reservoir module 103 modulate the intermediate optical signals on each path according to the modulation state to output a plurality of intermediate optical signals, then the plurality of intermediate optical signals are input to the output slab waveguide region for transmission, and finally the output waveguide of the AWG outputs a third group of optical signals.
The detection module 104 obtains the third group of optical signals, detects the light intensity of the third group of optical signals through the semiconductor optical amplifier and the detector, and performs nonlinear calculation to obtain a calculation result of the four-dimensional data at the first time. Meanwhile, the calculation result is converted from an optical signal to an electrical signal and output to the electrical processing module 105.
In this embodiment, assuming that the one-dimensional data is to-be-predicted data, the electrical processing module 105 calculates an output weight by using the real result of the training set and the electrical signal output by the detection module 104, and then the electrical processing module 105 calculates a prediction result of the one-dimensional data according to the output weight.
In the technical scheme, the input module and the modulators in the reservoir module can adjust the modulation states thereof according to requirements to obtain different weight values, so that the optimization solution of various data sets is met. Meanwhile, in order to obtain more random weight values, the AWG serving as the input module and the reservoir module can have large insertion loss unevenness of different wavelengths, and can accept large optical crosstalk among channels, so that the tolerance of the optical computing device in the manufacturing process is larger.
One possible implementation is shown in fig. 5, where the input module 101 and the reservoir module 103 in the optical computing apparatus 100 are made of Arrayed Waveguide Gratings (AWGs).
The AWG as input module 101 comprises at least one input waveguide (i.e., corresponding to at least one input port of the input module 101 depicted in fig. 1a or 1 b), an input slab waveguide region (i.e., corresponding to the free propagation region depicted in fig. 1a or 1 b), an arrayed waveguide, and an output slab waveguide region (i.e., corresponding to the free propagation region depicted in fig. 1a or 1 b), and at least one output waveguide (i.e., corresponding to at least one output port of the input module 101 depicted in fig. 1a or 1 b), to which at least one modulator of the input module 101 is connected (i.e., one modulator is connected to one input waveguide, but there are also cases where an input waveguide is not connected to a modulator); the AWG as the feedback module 102 includes at least one input waveguide (i.e., corresponding to at least one input port of the feedback module 102 depicted in fig. 1a or fig. 1 b), an input slab waveguide region, an arrayed waveguide, and an output slab waveguide region, and at least one output waveguide (i.e., corresponding to at least one output port of the feedback module 102 depicted in fig. 1a or fig. 1 b), with which at least one modulator of the feedback module 102 is connected (i.e., one modulator is connected to one input waveguide); the output waveguides of the input module 101 and the feedback module 102 are connected to the input waveguides of the reservoir module 103; the AWG as a reservoir module 103 comprises at least one input waveguide (i.e., corresponding to at least one input port of the reservoir module 103 depicted in fig. 1a or 1 b), an input slab waveguide region (i.e., corresponding to the free propagation region depicted in fig. 1a or 1 b), an arrayed waveguide and an output slab waveguide region (i.e., corresponding to the free propagation region depicted in fig. 1a or 1 b), and at least one output waveguide (i.e., corresponding to at least one output port of the reservoir module 103 depicted in fig. 1a or 1 b), at least one phase modulator of the reservoir module 103 being located on the arrayed waveguide, wherein one phase modulator is located on one arrayed waveguide, but there are also cases where an arrayed waveguide is not connected to a phase modulator, the output waveguide of the reservoir module 103 being connected to an input port of the detection module 104; at least one detector of the detection module 104 is connected to at least one input port of the detection module 104 (i.e. one detector is connected to an optical path corresponding to one input port); at least one output port of the detection module 104 is connected to the electrical processing module 105. In the configuration shown in fig. 5, the input module 101 and the feedback module 102 are independent from each other. That is, the input module 101 includes R input waveguides for receiving the first set of optical signals and then inputting the first set of optical signals to the input slab waveguide region of the AWG; the feedback module 102 includes R input waveguides for receiving feedback light.
In this scenario, the AWG as the input module 101 may be an N × 1 AWG structure, and the AWG as the reservoir module 103 may be a 1 × N AWG structure. Wherein the input module 101 comprises no more modulators than the number of input waveguides of the AWG structure, the feedback module 102 comprises a number of modulators equal to the number of output waveguides of the reservoir module 103, and the number of phase modulators of the reservoir module 103 is no more than the number of array waveguides of the AWG structure. The modulator of the input module 101 and the modulator of the feedback module 102 may be MZI modulators, micro-ring modulators, or electro-absorption modulators.
Under the structure shown in fig. 5, an application scenario is specifically described: suppose that the input module 101 uses 64 × 1 AWGs and 64 micro-ring modulators, i.e., each input waveguide is connected to one micro-ring modulator; the feedback module 102 selects 64 × 1 AWGs and 64 micro-ring modulators; the reservoir module 103 adopts 1 × 64 AWG and 64 phase modulators, i.e. each arrayed waveguide is connected with one phase modulator; the detection module 104 selects a semiconductor optical amplifier and a detector, wherein an input port, the semiconductor optical amplifier, the detector and an output port of the detection module 104 are sequentially connected; the electrical processing module 105 is an FPGA board. In the application scenario, the first data is one-dimensional data, and the first data at the first time is simultaneously loaded to 64 signal lights with different wavelengths as a first group of optical signals. The 64 input waveguides of the 64 × 1 AWG are respectively input with the first group of optical signals with 64 different wavelengths, that is, the 64 optical signals at the first time are identical. It is understood that the 64 wavelengths satisfy the wavelength design of the 64 x 1 AWG structure. The 64 micro-ring modulators in the input module 101 respectively perform random intensity modulation on the first group of optical signals with 64 different wavelengths, and output 64 paths of second group of optical signals.
At the first moment, the 64 input waveguides of the feedback module 102 respectively input the feedback light with the same wavelength as the first group of optical signals (that is, the wavelength of the first group of optical signals input on one input waveguide of the input module 101 is λ 1 、λ 2 、λ 3 、λ 4 Then, the wavelength of the feedback light input to one input waveguide of the feedback module 102 is λ 1 、λ 2 、λ 3 、λ 4 ). The 64 input waveguides of the feedback module 102 receive 64 feedback lights, and the feedback module 102 also receives 64 electrical signals (also referred to as a second calculation result) provided by the electrical processing module 105, where the second calculation result is a calculation result calculated from data before the first time; the 64 electrical signals act on the micro-ring modulator of the feedback module 102, and then the micro-ring modulator modulates the 64 feedback lights to obtain a first set of feedback optical signals (i.e. 64 modulated feedback optical signals).
In this configuration, the input module 101 outputs only the second set of optical signals and the feedback module 102 outputs only the first set of feedback optical signals. When the second set of optical signals and the first set of feedback optical signals are input to the reservoir module 103, a beam combiner is required to combine the second set of optical signals and the first set of feedback optical signals, and then input the combined signals to the input waveguide of the reservoir module 103. The second group of optical signals and the first group of feedback optical signals pass through an input slab waveguide region to obtain a group of intermediate optical signals, 64 phase modulators in the reservoir module 103 modulate the intermediate optical signals on each path according to the modulation state to output a plurality of intermediate optical signals, then the plurality of intermediate optical signals are input to the output slab waveguide region for transmission, and finally the output waveguide of the AWG outputs a third group of optical signals.
The detection module 104 obtains the third group of optical signals, detects the light intensity of the third group of optical signals through the semiconductor optical amplifier and the detector, and performs nonlinear calculation to obtain the calculation result of the four-dimensional data at the first time. Meanwhile, the calculation result is converted from an optical signal to an electrical signal and output to the electrical processing module 105.
In this embodiment, assuming that the one-dimensional data is to-be-predicted data, the electrical processing module 105 calculates an output weight by using the real result of the training set and the electrical signal output by the detection module 104, and then the electrical processing module 105 calculates a predicted result of the one-dimensional data according to the output weight.
In the technical scheme, the input module and the modulators in the reservoir module can adjust the modulation states of the modulators according to requirements to obtain different weight values, so that the optimization solution of various data sets is met. Meanwhile, in order to obtain more random weight values, the AWG serving as the input module and the reservoir module can have large insertion loss unevenness of different wavelengths, and can accept large optical crosstalk among channels, so that the tolerance of the optical computing device in the manufacturing process is larger.
Based on the scheme, no matter the MMI structure or the AWG structure is adopted, the number of the active devices is in a linear relation with the number of the output waveguides of the reservoir module, and therefore the number of the active devices is small.
The present embodiment also provides a light computing system 600, which includes a processor 601 and the light computing device 100 shown in fig. 1a or fig. 1b to fig. 5. Wherein the processor 601 is configured to input the first data to the light computing device 100. Wherein the processor 601 may be an electrical processing module in the optical computing device 100, i.e., the electrical processing module may also be used to input the first data to the optical computing device. In this embodiment, the processor may be a CPU, a microprocessor, an ASIC, or one or more integrated circuits for controlling the execution of the programs of the above-described aspects of the data transmission method.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the optical computing apparatus, the apparatus and the unit described above may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed optical computing devices, apparatus and methods may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical functional division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another computing apparatus, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit may be implemented in the form of hardware, or may also be implemented in the form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, which are essential or part of the technical solutions contributing to the prior art, or all or part of the technical solutions, may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

Claims (13)

1. A light computing device, comprising:
the device comprises an input module, a calculation module and a calculation module, wherein the input module is used for calculating a first group of input optical signals based on first weight to obtain a second group of optical signals, and the first group of optical signals is used for carrying first data to be calculated;
a feedback module, configured to obtain a first group of feedback optical signals based on a first calculation result, where the first calculation result is a calculation result of second data, and the second data is data processed before the first data is processed;
the reservoir module is used for calculating the first group of feedback optical signals and the second group of optical signals based on second weight to obtain a third group of optical signals;
and the detection module is used for detecting the light intensity of the third group of optical signals to obtain a second calculation result.
2. The light computing device of claim 1, further comprising:
and the electric processing module is used for sending the second calculation result to the feedback module.
3. The light computing device of claim 1 or 2, wherein the reservoir module comprises:
at least one phase modulator and a first multimode interference structure, the at least one phase modulator connected to at least one input waveguide of the first multimode interference structure;
the at least one phase modulator is configured to adjust phases of the first group of feedback optical signals and the second group of optical signals based on the second weight to obtain a plurality of first intermediate optical signals;
the first multimode interference structure is configured to receive the plurality of first intermediate optical signals and output the plurality of first intermediate optical signals as the third set of optical signals based on slab waveguide transmission.
4. The light computing device of claim 3, wherein the input module comprises:
a second multi-mode interference structure and a plurality of first modulators connected to a plurality of output waveguides of the second multi-mode interference structure;
the second multimode interference structure is configured to receive the first set of optical signals, the first set of optical signals including a plurality of optical signals of different wavelengths, and split the first set of optical signals into a plurality of sets of second intermediate optical signals, each set of second intermediate optical signals including the plurality of optical signals of different wavelengths;
the plurality of first modulators are configured to modulate the plurality of sets of second intermediate optical signals based on the first weights to obtain the second set of optical signals.
5. The light computing device of claim 4, wherein the second multimode interference structure comprises a 1 x N multimode interference structure or an N x N multimode interference structure.
6. The light computing device of any of claims 3 to 5, wherein the feedback module comprises:
a plurality of second modulators for loading the first calculation results on feedback light to obtain the first set of feedback light signals.
7. The light computing device of claim 1 or 2, wherein the input module comprises:
the modulator comprises a plurality of third modulators and a first arrayed waveguide grating, wherein the third modulators are connected with a plurality of input waveguides of the first arrayed waveguide grating;
the plurality of third modulators are configured to receive the first set of optical signals and modulate the first set of optical signals based on the first weight to obtain a plurality of third intermediate optical signals;
and the first arrayed waveguide grating receives the plurality of third intermediate optical signals and transmits and outputs the third intermediate optical signals as the second group of optical signals based on the slab waveguide.
8. The light computing device of claim 7, wherein the reservoir module comprises:
at least one phase modulator and a second arrayed waveguide grating, the at least one phase modulator being located on an arrayed waveguide of the second arrayed waveguide grating;
the second arrayed waveguide grating is used for receiving the first group of feedback optical signals and the second group of optical signals and outputting the first group of feedback optical signals and the second group of optical signals as a plurality of fourth intermediate optical signals based on slab waveguide transmission;
the at least one phase modulator is configured to modulate the phases of the plurality of fourth intermediate optical signals based on the second weight to obtain a plurality of fifth intermediate optical signals;
the second arrayed waveguide grating is further configured to transmit and output the fifth intermediate optical signals as the third group of optical signals based on a slab waveguide.
9. The light computing device of claim 7 or 8, wherein the feedback module comprises:
a plurality of fourth modulators connected to a plurality of input waveguides of a third arrayed waveguide grating;
the plurality of fourth modulators are used for loading the first calculation result on feedback light to obtain a first group of modulation signals;
and the third arrayed waveguide grating is used for receiving the first group of modulation signals and outputting the first group of modulation signals as the first group of feedback optical signals based on slab waveguide transmission.
10. The light computing device of any of claims 1-9, wherein the detection module comprises: a semiconductor optical amplifier and a semiconductor optical detector;
the semiconductor optical amplifier and the semiconductor optical detector are used for detecting the light intensity of the third group of optical signals and acquiring the second calculation result;
alternatively, the first and second electrodes may be,
the detection module comprises a semiconductor light detector;
and the semiconductor optical detector is used for detecting the light intensity of the third group of optical signals and acquiring the second calculation result.
11. A light computing device according to any one of claims 1-9, further comprising:
an array of light sources, a first light source device in the array of light sources to receive the first data and to generate the first set of light signals based on the first data; a second light source device in the array of light sources is for emitting the feedback light.
12. The light computing device of any of claims 1-11, wherein the computing device comprises a chip.
13. A light computing system, characterized by a processor and a light computing device according to any of claims 1-12, the processor being adapted to input the first data to the light computing device.
CN202110604196.5A 2021-05-31 2021-05-31 Optical computing device and system Pending CN115481711A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110604196.5A CN115481711A (en) 2021-05-31 2021-05-31 Optical computing device and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110604196.5A CN115481711A (en) 2021-05-31 2021-05-31 Optical computing device and system

Publications (1)

Publication Number Publication Date
CN115481711A true CN115481711A (en) 2022-12-16

Family

ID=84418996

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110604196.5A Pending CN115481711A (en) 2021-05-31 2021-05-31 Optical computing device and system

Country Status (1)

Country Link
CN (1) CN115481711A (en)

Similar Documents

Publication Publication Date Title
Shi et al. Deep neural network through an InP SOA-based photonic integrated cross-connect
CN112232503B (en) Computing device, computing method, and computing system
CN115222035B (en) Photon neural network convolution acceleration chip
CN115169542B (en) Two-dimensional photon convolution acceleration system and device for convolution neural network
US20240063936A1 (en) Systems and methods for utilizing photonic degrees of freedom in a photonic processor
Jiang et al. Photonic convolution neural network based on interleaved time-wavelength modulation
WO2022016894A1 (en) Photonic neural network
CN112101540B (en) Optical neural network chip and calculation method thereof
CN111753977A (en) Optical neural network convolution layer chip, convolution calculation method and electronic equipment
CN112308224A (en) Optical neural network device, chip and optical implementation method for neural network calculation
CN113805641B (en) Photonic neural network
De Lima et al. Scalable wideband principal component analysis via microwave photonics
CN115481711A (en) Optical computing device and system
CN113325917A (en) Light computing device, system and computing method
CN114565091A (en) Optical neural network device, chip and optical implementation method for neural network calculation
CN114325932B (en) On-chip integrated all-optical neural network optical computing chip
CN112783260A (en) Optical computing device, optical computing method and computing system
Prucnal et al. Neuromorphic photonics: current status and challenges
CN117436494B (en) Neural network operation chip and operation method
CN112001487B (en) Photonic neural network
CN112232487B (en) Optical neural network chip and calculation method thereof
Zhang et al. Time-stretch optical neural network with time-division multiplexing
JP7273342B2 (en) Optical information processing device
Shi et al. Inp photonic circuit for deep neural networks
JP7110822B2 (en) optical signal processor

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