[go: up one dir, main page]

CN116266771A - Convolution operation device and convolution operation method based on multi-mode interference - Google Patents

Convolution operation device and convolution operation method based on multi-mode interference Download PDF

Info

Publication number
CN116266771A
CN116266771A CN202111558032.XA CN202111558032A CN116266771A CN 116266771 A CN116266771 A CN 116266771A CN 202111558032 A CN202111558032 A CN 202111558032A CN 116266771 A CN116266771 A CN 116266771A
Authority
CN
China
Prior art keywords
laser
multimode interference
beams
laser light
convolution operation
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
CN202111558032.XA
Other languages
Chinese (zh)
Inventor
李明
孟祥彦
石暖暖
张国杰
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.)
Institute of Semiconductors of CAS
Original Assignee
Institute of Semiconductors of CAS
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 Institute of Semiconductors of CAS filed Critical Institute of Semiconductors of CAS
Priority to CN202111558032.XA priority Critical patent/CN116266771A/en
Publication of CN116266771A publication Critical patent/CN116266771A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B10/00Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
    • H04B10/50Transmitters
    • H04B10/501Structural aspects
    • H04B10/503Laser transmitters
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • G06F17/15Correlation function computation including computation of convolution operations
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B10/00Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
    • H04B10/07Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems
    • H04B10/075Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal
    • H04B10/079Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal using measurements of the data signal
    • H04B10/0795Performance monitoring; Measurement of transmission parameters
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B10/00Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
    • H04B10/25Arrangements specific to fibre transmission
    • H04B10/2581Multimode transmission
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B10/00Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
    • H04B10/50Transmitters
    • H04B10/516Details of coding or modulation
    • H04B10/532Polarisation modulation
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B10/00Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
    • H04B10/60Receivers

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Electromagnetism (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Optics & Photonics (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Optical Communication System (AREA)

Abstract

The present disclosure provides a convolution operation device based on multimode interference, comprising: the laser generation module is used for generating first laser; the data input module is used for modulating the data to be processed onto the first laser; the delay module is used for delaying and splitting the modulated first laser to obtain N beams of second laser; the multimode interference coupler is used for simultaneously carrying out multimode interference on N beams of second lasers according to a preset convolution matrix and outputting M groups of optical signals; the detector array comprises M photoelectric detectors, and the M photoelectric detectors are used for respectively receiving M groups of optical signals and detecting the intensity of the M groups of optical signals to obtain convolution operation results; wherein M and N are positive integers. The present disclosure also provides a convolution operation method based on multimode interference.

Description

基于多模干涉的卷积运算装置和卷积运算方法Convolution operation device and convolution operation method based on multi-mode interference

技术领域technical field

本公开涉及大数据技术和微波光子学领域,具体涉及一种基于多模干涉的卷积运算装置及其卷积运算方法。The present disclosure relates to the fields of big data technology and microwave photonics, in particular to a multimode interference-based convolution operation device and a convolution operation method thereof.

背景技术Background technique

卷积运算作为一种可以广泛应用于信号处理、物理学、图像处理、人工智能等领域的大数据运算方法,需要依托于高性能计算机提高数据处理的效率。As a big data computing method that can be widely used in signal processing, physics, image processing, artificial intelligence and other fields, convolution operation needs to rely on high-performance computers to improve the efficiency of data processing.

随着5G技术、自动驾驶技术以及高速网络系统的发展,需要通过卷积运算进行处理的数据量呈指数型的增长。因此,传统的基于冯诺依曼架构的计算机正日益面临着延时、能耗和处理速率的挑战。With the development of 5G technology, autonomous driving technology, and high-speed network systems, the amount of data that needs to be processed through convolution operations is increasing exponentially. Therefore, traditional von Neumann-based computers are increasingly facing challenges in latency, energy consumption, and processing speed.

发明内容Contents of the invention

有鉴于此,为了克服上述问题的至少一个方面,本公开提供一种基于多模干涉的卷积运算装置,包括:激光发生模块,用于产生第一激光;数据输入模块,用于将待处理数据调制到所述第一激光上;延时模块,用于对调制后的第一激光进行延时和分束,得到N束第二激光;至少一个多模干涉耦合器,用于根据预设卷积矩阵同时对所述N束第二激光进行多模干涉,输出M组光信号;以及探测器阵列包括M个光电探测器,所述M个光电探测器用于分别接收所述M组光信号,并探测所述M组光信号的强度,得到卷积运算结果;其中,M和N为正整数。In view of this, in order to overcome at least one aspect of the above-mentioned problems, the present disclosure provides a multi-mode interference-based convolution operation device, including: a laser generation module, used to generate the first laser light; a data input module, used to process The data is modulated onto the first laser light; the delay module is used to delay and split the modulated first laser light to obtain N beams of second laser light; at least one multimode interference coupler is used to The convolution matrix simultaneously performs multi-mode interference on the N beams of second laser light, and outputs M groups of optical signals; and the detector array includes M photodetectors, and the M photodetectors are used to respectively receive the M groups of optical signals , and detecting the intensities of the M groups of optical signals to obtain a convolution operation result; wherein, M and N are positive integers.

根据本公开实施例,所述多模干涉耦合器包括N个输入端口和M个输出端口;所述多模干涉耦合器,用于通过所述N个输入端口分别接收所述N束第二激光,并分别根据N组预设分光比将所述N束第二激光分束至所述M个输出端口,得到M组光信号;其中,所述N组预设分光比与所述N束第二激光一一对应。According to an embodiment of the present disclosure, the multimode interference coupler includes N input ports and M output ports; the multimode interference coupler is configured to respectively receive the N beams of second laser light through the N input ports , and split the N second laser beams to the M output ports respectively according to N sets of preset splitting ratios to obtain M sets of optical signals; wherein, the N sets of preset splitting ratios are the same as the N beams of the first The two lasers correspond one to one.

根据本公开实施例,所述预设卷积矩阵包括M个卷积核;所述多模干涉耦合器,用于分别根据所述M个卷积核调整所述N束第二激光的强度,得到M组光信号;其中,所述M个卷积核与所述M组光信号一一对应。According to an embodiment of the present disclosure, the preset convolution matrix includes M convolution kernels; the multimode interference coupler is configured to adjust the intensity of the N beams of second laser light according to the M convolution kernels respectively, M groups of optical signals are obtained; wherein, the M convolution kernels are in one-to-one correspondence with the M groups of optical signals.

根据本公开实施例,所述多模干涉耦合器包括N个输入波导、M个输出波导和多模干涉区;所述N个输入波导和M个输出波导分别以预设间距安装在所述多模干涉区两侧。According to an embodiment of the present disclosure, the multimode interference coupler includes N input waveguides, M output waveguides, and a multimode interference region; the N input waveguides and M output waveguides are respectively installed on the multimode Both sides of the mode interference area.

根据本公开实施例,所述多模干涉区包括多个多模干涉单元和多个调控件,所述多个调控件分别用于调整所述多个多模干涉单元中一个多模干涉单元的折射率。According to an embodiment of the present disclosure, the multimode interference area includes a plurality of multimode interference units and a plurality of adjustment parts, and the plurality of adjustment parts are respectively used to adjust a multimode interference unit in the plurality of multimode interference units. refractive index.

根据本公开实施例,所述多模干涉耦合器还包括N个偏振控制器;所述N个偏振控制器,用于分别控制所述N束第二激光的偏振态。According to an embodiment of the present disclosure, the multimode interference coupler further includes N polarization controllers; the N polarization controllers are configured to respectively control the polarization states of the N beams of second laser light.

根据本公开实施例,在所述至少一个多模干涉耦合器包括多个多模干涉耦合器的情况下,所述多个多模干涉耦合器并联。According to an embodiment of the present disclosure, when the at least one multimode interference coupler includes a plurality of multimode interference couplers, the plurality of multimode interference couplers are connected in parallel.

根据本公开实施例,所述数据输入模块包括:任意波形产生器,用于将待处理数据转换为待处理信号;电光调制器,用于接收待所述处理信号和所述第一激光,将所述待处理信号调制到所述第一激光上。According to an embodiment of the present disclosure, the data input module includes: an arbitrary waveform generator, used to convert the data to be processed into a signal to be processed; an electro-optical modulator, used to receive the signal to be processed and the first laser, and convert The signal to be processed is modulated onto the first laser.

根据本公开实施例,所述延时模块包括分束器和至少N-1个光延时线,所述待处理数据包括连续输入的N位数据,每位数据持续的时间为一个输入周期;所述分束器用于将调制后的第一激光分为N束,得到N个激光分量;所述至少N-1个光延时线,用于对所述N个激光分量分别进行延时,得到N束第二激光;其中,所述N束第二激光的N-1束第二激光被依次延时一个输入周期,以使所述N位数据同时输入所述多模干涉耦合器。According to an embodiment of the present disclosure, the delay module includes a beam splitter and at least N-1 optical delay lines, the data to be processed includes continuously input N-bit data, and the duration of each bit of data is one input cycle; The beam splitter is used to divide the modulated first laser into N beams to obtain N laser components; the at least N-1 optical delay lines are used to delay the N laser components respectively, Obtaining N beams of second laser light; wherein, N-1 beams of second laser beams of the N beams of second laser light are sequentially delayed by one input period, so that the N-bit data is simultaneously input into the multimode interference coupler.

本公开还提供一种多模干涉的卷积运算方法,基于上述任意一项所述的多模干涉的卷积运算装置,所述方法包括:通过激光发生模块产生第一激光;通过数据输入模块将待处理数据调制到所述第一激光上;通过延时模块对调制后的第一激光进行延时和分束,得到N束第二激光;通过至少一个多模干涉耦合器根据预设卷积矩阵同时对所述N束第二激光进行多模干涉,输出M组光信号;通过探测器阵列包括M个光电探测器,所述M个光电探测器分别接收所述M组光信号,并探测所述N组光信号的强度,得到卷积运算结果;其中,M和N为正整数。The present disclosure also provides a multi-mode interference convolution operation method, based on the multi-mode interference convolution operation device described in any one of the above, the method includes: generating the first laser through the laser generation module; through the data input module Modulate the data to be processed onto the first laser; delay and split the modulated first laser through a delay module to obtain N beams of second laser; pass at least one multimode interference coupler according to the preset volume The product matrix performs multi-mode interference on the N beams of second laser light at the same time, and outputs M groups of optical signals; the detector array includes M photodetectors, and the M photodetectors respectively receive the M groups of optical signals, and Detecting the intensities of the N groups of optical signals to obtain a convolution operation result; wherein, M and N are positive integers.

与现有技术相比,本公开具有以下有益效果:Compared with the prior art, the present disclosure has the following beneficial effects:

1、利用光具有的大带宽、高速率和低延时等优点,对光的传输时间和光的波长进行设计,通过光传输进行高效的大数据处理。1. Utilize the advantages of large bandwidth, high speed and low delay of light, design the transmission time and wavelength of light, and carry out efficient big data processing through optical transmission.

2、使用不同波长的多个激光分量实现多个卷积核的并行卷积运算,不仅提高了数据的输入速率,还使得数据处理过程更加高速。2. Using multiple laser components with different wavelengths to realize the parallel convolution operation of multiple convolution kernels not only improves the data input rate, but also makes the data processing process faster.

3、通过对多模干涉耦合器中结构参数的特殊设计、折射率的区域性精细调控、输入光的偏振控制等技术实现卷积核的实时调整,使卷积运算系统具有完全可重构性和了扩展性。3. Through the special design of the structural parameters in the multimode interference coupler, the regional fine adjustment of the refractive index, and the polarization control of the input light, the real-time adjustment of the convolution kernel is realized, so that the convolution operation system is fully reconfigurable and scalability.

附图说明Description of drawings

为了更完整地理解本公开及其优势,现在将参考结合附图的以下描述,其中:For a more complete understanding of the present disclosure and its advantages, reference should now be made to the following description taken in conjunction with the accompanying drawings, in which:

图1示意性示出了本公开实施例的基于多模干涉的卷积运算装置的示意图;FIG. 1 schematically shows a schematic diagram of a multimode interference-based convolution operation device according to an embodiment of the present disclosure;

图2示意性示出了根据本公开实施例的激光延时的示意图;Fig. 2 schematically shows a schematic diagram of laser time delay according to an embodiment of the present disclosure;

图3示意性示出了本公开实施例的多模干涉耦合器的结构示意图;FIG. 3 schematically shows a schematic structural diagram of a multimode interference coupler according to an embodiment of the present disclosure;

图4示意性示出了本公开另一实施例的多模干涉耦合器的结构示意图;FIG. 4 schematically shows a schematic structural diagram of a multimode interference coupler according to another embodiment of the present disclosure;

图5示意性示出了本公开另一实施例的多模干涉耦合器的结构示意图;FIG. 5 schematically shows a schematic structural diagram of a multimode interference coupler according to another embodiment of the present disclosure;

图6示意性示出了根据本公开另一实施例的基于多模干涉的卷积运算装置的示意图;以及FIG. 6 schematically shows a schematic diagram of a convolution operation device based on multi-mode interference according to another embodiment of the present disclosure; and

图7示意性示出了本公开实施例的多模干涉的卷积运算方法的流程图。Fig. 7 schematically shows a flow chart of a multimode interference convolution operation method according to an embodiment of the present disclosure.

附图标记说明Explanation of reference signs

1 激光发生模块1 laser generator module

2 数据输入模块2 Data input module

3 延时模块3 delay module

4 多模干涉耦合器4 Multimode interference coupler

41 输入波导41 Input waveguide

42 输出波导42 output waveguide

43 多模干涉区43 Multimode interference zone

44 偏振控制器44 Polarization controller

5 探测器阵列5 detector array

51 光电探测器51 photodetector

6 光耦合器6 Optocoupler

具体实施方式Detailed ways

为了更清楚地说明本公开实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,这些描述只是示例性的,而并非要限制本公开的范围。对于本领域技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。在下面的详细描述中,为便于解释,阐述了许多具体的细节以提供对本公开实施例的全面理解。除非另外定义,本公开使用的技术术语或者科学术语应当为本公开所属领域内具有一般技能的人士所理解的通常意义。In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, these descriptions are only exemplary. It is not intended to limit the scope of the present disclosure. For those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure. Unless otherwise defined, the technical terms or scientific terms used in the present disclosure shall have the usual meanings understood by those skilled in the art to which the present disclosure belongs.

在此使用的术语仅仅是为了描述具体实施例,而并非意在限制本公开。在此使用的术语“包括”、“包含”等表明了所述特征、步骤、操作和/或部件的存在,但是并不排除存在或添加一个或多个其他特征、步骤、操作或部件。The terminology used herein is for the purpose of describing particular embodiments only, and is not intended to be limiting of the present disclosure. The terms "comprising", "comprising", etc. used herein indicate the presence of stated features, steps, operations and/or components, but do not exclude the presence or addition of one or more other features, steps, operations or components.

图1示意性示出了根据本公开实施例的基于多模干涉的卷积运算装置的示意图。如图1所示,本公开提供了一种基于多模干涉的卷积运算装置100包括:激光发生模块1、数据输入模块2、延时模块3、至少一个多模干涉耦合器4和探测器阵列5。Fig. 1 schematically shows a schematic diagram of a convolution operation device based on multi-mode interference according to an embodiment of the present disclosure. As shown in FIG. 1 , the present disclosure provides a multimode interference-based convolution operation device 100 including: a laser generation module 1, a data input module 2, a delay module 3, at least one multimode interference coupler 4 and a detector array5.

激光发生模块1、数据输入模块2、延时模块3、至少一个多模干涉耦合器4和探测器阵列5依次通过光纤跳线连接。The laser generating module 1 , the data input module 2 , the delay module 3 , at least one multimode interference coupler 4 and the detector array 5 are sequentially connected through optical fiber jumpers.

基于多模干涉的卷积运算装置100还包括光放大器,光放大器可位于光学链路中任意位置,用于放大光学链路中激光的功率。The multi-mode interference-based convolution operation device 100 also includes an optical amplifier, which can be located anywhere in the optical link, and is used to amplify the power of the laser in the optical link.

激光发生模块1,用于产生第一激光。The laser generating module 1 is used to generate the first laser.

为实现对多个卷积核的卷积运算,第一激光可以包括N个激光分量,N为正整数。In order to implement convolution operations on multiple convolution kernels, the first laser may include N laser components, where N is a positive integer.

示例性地,激光发生模块1包括N个激光器和合束器。该N个激光器分别生成N个具有不同波长的激光,合束器将N个具有不同波长的激光合束为第一激光。合束器包括但不限于光耦合器、波分复用器、密集波分复用器和阵列波导光栅。Exemplarily, the laser generating module 1 includes N lasers and a beam combiner. The N lasers respectively generate N lasers with different wavelengths, and the beam combiner combines the N lasers with different wavelengths into a first laser. Beam combiners include, but are not limited to, optical couplers, wavelength division multiplexers, dense wavelength division multiplexers, and arrayed waveguide gratings.

示例性地,激光发生模块1可以是多波长激光器。多波长激光器产生一束激光,该激光包括具有不同波长的N个激光分量。Exemplarily, the laser generating module 1 may be a multi-wavelength laser. A multi-wavelength laser generates a laser beam including N laser components with different wavelengths.

数据输入模块2,用于将待处理数据调制到第一激光上。The data input module 2 is used for modulating the data to be processed onto the first laser.

数据输入模块2包括任意波形产生器和电光调制器。任意波形产生器和电光调制器通过电缆连接。The data input module 2 includes an arbitrary waveform generator and an electro-optic modulator. The arbitrary waveform generator and the electro-optic modulator are connected by a cable.

数据输入模块2还可以包括电放大器,电放大器可位于任意波形产生器和电光调制器之间,电放大器通过电缆两端分别连接任意波形产生器和电光调制器,用于接收并放大待处理信号的功率,将放大后的待处理信号发送给电光调制器。The data input module 2 can also include an electrical amplifier, which can be located between the arbitrary waveform generator and the electro-optical modulator, and the electrical amplifier is connected to the arbitrary waveform generator and the electro-optical modulator through the two ends of the cable, for receiving and amplifying the signal to be processed The power of the amplified signal to be processed is sent to the electro-optical modulator.

任意波形产生器,用于将待处理数据转换为待处理信号。待处理数据包括但不限于图像、音频、视频和文本。任意波形产生器将待处理的图像、音频和文本等转换为待处理的电信号。Arbitrary waveform generator for converting data to be processed into signal to be processed. Data to be processed includes, but is not limited to, images, audio, video, and text. Arbitrary waveform generators convert images, audio and text to be processed into electrical signals to be processed.

示例性地,可以通过可编程脉冲发生器(PPG)将待处理数据转换为待处理信号,也可以通过现场可编程门阵列(FPGA)、中央处理器(CPU)、图形处理器(GPU)、专用集成电路(ASIC)等逻辑运算的单元与数模转换器的组合将待处理数据转换为待处理信号。Exemplarily, the data to be processed can be converted into a signal to be processed by a programmable pulse generator (PPG), or a field programmable gate array (FPGA), a central processing unit (CPU), a graphics processing unit (GPU), The combination of a logical operation unit such as an application-specific integrated circuit (ASIC) and a digital-to-analog converter converts the data to be processed into a signal to be processed.

电光调制器,用于接收待处理信号和第一激光,将待处理信号调制到第一激光上。The electro-optic modulator is used to receive the signal to be processed and the first laser, and modulate the signal to be processed onto the first laser.

待处理的电信号作为调制信号,通过电光调制器将待处理信号调制到第一激光上。可理解地,通过电光调制器将待处理数据加载到第一激光的强度上。The electrical signal to be processed is used as a modulation signal, and the signal to be processed is modulated onto the first laser by the electro-optic modulator. Understandably, the data to be processed is loaded on the intensity of the first laser light through the electro-optic modulator.

需要说明的是,调制后的第一激光的强度发生改变。调制后的第一激光包括的N个激光分量的强度都发生了改变,但每个激光分量的强度是发生同等比例的变化。调制后的激光分量之间的相对强度关系与调制前激光分量之间的相对强度关系一致。It should be noted that the intensity of the modulated first laser light changes. The intensity of the N laser components included in the modulated first laser light all changes, but the intensity of each laser component changes in the same proportion. The relative intensity relationship between the modulated laser components is consistent with the relative intensity relationship between the laser components before modulation.

延时模块3,用于对调制后的第一激光进行延时和分束,得到N束第二激光。The delay module 3 is configured to delay and split the modulated first laser light to obtain N beams of second laser light.

由于在进行卷积运算时,需要将N位待卷积数据同时输入至多模干涉耦合器4,而待处理数据是串行的调制到第一激光上,因此需要对调制后的第一激光的N个激光分量进行不同程度的延时,使得N位待卷积数据可以同时输入至多模干涉耦合器4。Since when performing the convolution operation, it is necessary to simultaneously input N bits of data to be convolved into the multimode interference coupler 4, and the data to be processed is serially modulated onto the first laser, so it is necessary to The N laser components are delayed to different degrees, so that the N bits of data to be convolved can be input to the multimode interference coupler 4 at the same time.

延时模块3包括分束器和至少N-1个光延时线。待处理数据包括连续输入的N位数据,每位数据持续的时间为一个输入周期。The delay module 3 includes a beam splitter and at least N-1 optical delay lines. The data to be processed includes continuously input N-bit data, and the duration of each bit of data is one input cycle.

分束器用于将调制后的第一激光分为N束,得到N个激光分量。分束器根据波长将调制后的第一激光分束为N束,每个波长分为一束,每一束形成一个激光分量。分束器包括但不限于波分解复用器、密集波分解复用器、阵列波导光栅和波形整形器。The beam splitter is used to divide the modulated first laser beam into N beams to obtain N laser components. The beam splitter splits the modulated first laser beam into N beams according to the wavelength, each wavelength is divided into one beam, and each beam forms a laser component. Beam splitters include, but are not limited to, wave division multiplexers, dense wave division multiplexers, arrayed waveguide gratings, and wave shapers.

至少N-1个光延时线分别对N个激光分量中的N-1个激光分量进行延时,得到N束第二激光。其中,N束第二激光中的N-1束第二激光被依次延时一个输入周期,以使N位数据同时分别输入多模干涉耦合器4。At least N-1 optical delay lines respectively delay N-1 laser components in the N laser components to obtain N beams of second laser light. Among the N second laser beams, N−1 second laser beams are sequentially delayed for one input period, so that N bits of data are input to the multimode interference coupler 4 at the same time.

图2示意性示出了根据本公开实施例的激光延时的示意图。Fig. 2 schematically shows a schematic diagram of laser time delay according to an embodiment of the present disclosure.

如图2(A)所示,激光发生模块1产生第一激光,第一激光包括4束激光Laser1~Laser4。待处理数据包括数据X1~X4,待处理数据的输入频率为f。在经过数据输入模块2后,激光Laser1~Laser4均携带有待处理数据X1~X4。由于待处理数据的输入频率为f,因此每个待处理数据之间的输入时刻相差一个输入周期T=1/f。As shown in FIG. 2(A), the laser generating module 1 generates a first laser, and the first laser includes four laser beams Laser 1 -Laser 4 . The data to be processed includes data X 1 to X 4 , and the input frequency of the data to be processed is f. After passing through the data input module 2, the lasers Laser 1 -Laser 4 all carry data X 1 -X 4 to be processed. Since the input frequency of the data to be processed is f, the input time difference between each data to be processed is an input period T=1/f.

若4束激光同时输入多模干涉耦合器4,多模干涉耦合器4在第一个数据输入周期只能同时接收到待处理数据X1,在第二个数据输入周期只能同时接收到待处理数据X2If 4 beams of laser light are input to the multimode interference coupler 4 at the same time, the multimode interference coupler 4 can only receive the data X 1 to be processed at the same time in the first data input cycle, and can only receive the data to be processed at the same time in the second data input cycle Data X 2 is processed.

如图2(B)所示,通过延时模块3对4束激光分别进行延时,使得多模干涉耦合器4在某一时刻可以同时接收到待处理数据X1~X4。延时操作包括将激光Laser1延时3个输入周期、将激光Laser2延时2个输入周期、将激光Laser3延时1个输入周期,对激光Laser4不延时。As shown in FIG. 2(B), the delay module 3 delays the four laser beams separately, so that the multimode interference coupler 4 can simultaneously receive the data X 1 -X 4 to be processed at a certain moment. The delay operation includes delaying the laser Laser 1 for 3 input cycles, delaying the laser Laser 2 for 2 input cycles, delaying the laser Laser 3 for 1 input cycle, and not delaying the laser 4 .

至少一个多模干涉耦合器4,用于根据预设卷积矩阵同时对N束第二激光进行多模干涉,输出M组光信号。At least one multimode interference coupler 4 is used to simultaneously perform multimode interference on N beams of second laser light according to a preset convolution matrix, and output M groups of optical signals.

多模干涉耦合器4包括N个输入端口和M个输出端口。多模干涉耦合器4通过N个输入端口分别接收N束第二激光,并分别根据N组预设分光比将N束第二激光分束至M个输出端口,通过M个输出端口输出M组光信号。The multimode interference coupler 4 includes N input ports and M output ports. The multi-mode interference coupler 4 respectively receives N beams of second laser light through N input ports, and splits N beams of second laser light into M output ports according to N groups of preset splitting ratios, and outputs M groups through M output ports. light signal.

N组预设分光比与N束第二激光一一对应。每束第二激光根据对应的分光比分束至M个输出端,每个输出端口都输出由N束第二激光分束而来的激光。The N sets of preset splitting ratios correspond to the N beams of second laser light one by one. Each second laser beam is split to M output terminals according to the corresponding beam splitting ratio, and each output port outputs laser beams split by N second laser beams.

例如,多模干涉耦合器4包括4个输入端口和5个输出端口,多模干涉耦合器4通过4个输入端口分别接收4束激光Laser’1~Laser’4。根据4组预设分光比S11~S15、S21~S25、S31~S35、S41~S45对将4束激光Laser’1~Laser’4进行分束。For example, the multimode interference coupler 4 includes 4 input ports and 5 output ports, and the multimode interference coupler 4 respectively receives 4 laser beams Laser' 1 -Laser' 4 through the 4 input ports. The four laser beams Laser' 1 -Laser' 4 are beam-split according to 4 sets of preset splitting ratios S 11 -S 15 , S 21 -S 25 , S 31 -S 35 , and S 41 -S 45 .

根据预设分光比S11~S15将激光Laser’1分束至5个输出端口,根据预设分光比S21~525将激光Laser’2分束至5个输出端口,根据预设分光比S31~S35将激光Laser’3分束至5个输出端口,根据预设分光比S41~S45将激光Laser’4分束至5个输出端口。5个输出端口输出5组光信号,第一个输出端口输出的光信号包括激光Laser’1~Laser’4分别根据预设分光比S11、S21、S31和S41分束至第一个输出端口的激光,第二个输出端口输出的光信号包括激光Laser’1~Laser’4分别根据预设分光比S12、S22、S32和S42分束至第二个输出端口的激光,第三个输出端口输出的光信号包括激光Laser’1~Laser’4分别根据预设分光比S13、S23、S33和S43分束至第三个输出端口的激光,第四个输出端口输出的光信号包括激光Laser’1~Laser’4分别根据预设分光比S14、S24、S34和S44分束至第四个输出端口的激光,以及第五个输出端口输出的光信号包括激光Laser’1~Laser’4分别根据预设分光比S15、S25、S35和S45分束至第五个输出端口的激光。Split the laser Laser' 1 to 5 output ports according to the preset split ratio S 11 ~ S 15 , split the laser Laser' 2 to 5 output ports according to the preset split ratio S 21 ~ 5 25 , and split the laser beam to 5 output ports according to the preset split ratio The ratios S 31 to S 35 split the laser Laser' 3 to 5 output ports, and split the laser Laser' 4 to 5 output ports according to the preset splitting ratios S 41 to S 45 . The 5 output ports output 5 groups of optical signals. The optical signals output by the first output port include lasers Laser' 1 ~ Laser' 4 respectively according to the preset splitting ratios S 11 , S 21 , S 31 and S 41 to the first The laser output from the first output port, the optical signal output from the second output port includes the lasers Laser' 1 ~ Laser' 4 split to the second output port according to the preset splitting ratios S 12 , S 22 , S 32 and S 42 respectively Laser, the optical signal output by the third output port includes lasers Laser' 1 ~ Laser' 4 split into the third output port according to the preset splitting ratios S 13 , S 23 , S 33 and S 43 respectively, and the fourth The optical signals output by the first output ports include the laser beams split by Laser' 1 ~ Laser' 4 to the fourth output port according to the preset splitting ratios S 14 , S 24 , S 34 and S 44 respectively, and the fifth output port The output optical signals include the laser beams split by the lasers Laser' 1 to Laser' 4 to the fifth output port according to the preset splitting ratios S 15 , S 25 , S 35 and S 45 respectively.

N束激光根据预设分光比进行分束实质是激光上负载的待处理数据根据预设卷积矩阵进行卷积运算。预设卷积矩阵包括M个卷积核。多模干涉耦合器分别根据M个卷积核的其中一个卷积核调整N束第二激光的强度,得到M组光信号。其中,M个卷积核与M组光信号一一对应。The splitting of N beams of lasers according to the preset splitting ratio is essentially that the data to be processed loaded on the laser is convoluted according to the preset convolution matrix. The preset convolution matrix includes M convolution kernels. The multimode interference coupler adjusts the intensity of the N beams of second laser light according to one of the M convolution kernels respectively to obtain M groups of optical signals. Wherein, the M convolution kernels are in one-to-one correspondence with the M groups of optical signals.

示例性的,预设卷积矩阵为

Figure BDA0003418647360000081
卷积核分别为{(K11 ...K1N),...,(KM1 ... KMN)},待处理数据包括X1~XN,输出的卷积运算结果为Y1~YM。Exemplarily, the preset convolution matrix is
Figure BDA0003418647360000081
The convolution kernels are {(K 11 ... K 1N ), ..., (K M1 ... K MN )}, the data to be processed includes X 1 ~ X N , and the output convolution operation result is Y 1 ~ Y M .

卷积运算过程可表示为式(1):The convolution operation process can be expressed as formula (1):

Figure BDA0003418647360000082
Figure BDA0003418647360000082

卷积运算结果Y1=X1K11+X2K12+…+XNK1NConvolution operation result Y1=X 1 K 11 +X 2 K 12 +...+X N K 1N .

例如,预设卷积矩阵包括5个卷积核C1~C5。多模干涉耦合器4分别根据5个卷积核C1~C5调整4束激光Laser’1~Laser’4的强度,得到5组光信号。For example, the preset convolution matrix includes 5 convolution kernels C 1 -C 5 . The multimode interference coupler 4 adjusts the intensities of the four laser beams Laser' 1 -Laser' 4 respectively according to the five convolution kernels C 1 -C 5 to obtain five groups of optical signals.

根据卷积核C1调整4个激光Laser’1~Laser’4的强度,对激光Laser’1~Laser’4上负载的待处理数据进行卷积运算,输出第一组光信号Laser”1。同理,输出光信号Laser”2~Laser”5。其中,5个卷积核与5组光信号一一对应,每个卷积核包括4个元素。Adjust the intensities of the four lasers Laser' 1 - Laser' 4 according to the convolution kernel C 1 , perform convolution operations on the data to be processed loaded on the lasers Laser' 1 - Laser' 4 , and output the first group of optical signals Laser" 1 . Similarly, output optical signals Laser" 2 -Laser" 5 . Among them, 5 convolution kernels correspond to 5 groups of optical signals one by one, and each convolution kernel includes 4 elements.

需要说明的是,每个卷积核包括的元素数量与多模干涉耦合器输入端口数量相同。卷积核具有多种形式。在上述实施例中,卷积核以一维横向量的形式表示,卷积核还可以多维矩阵的形式表示。但无论卷积核以何种形式表示,卷积运算过程均可以通过上述式(1)表示。在卷积核为多维矩阵的情况下,可将多维矩阵转化为一维横向量。It should be noted that the number of elements included in each convolution kernel is the same as the number of input ports of the multimode interference coupler. Convolution kernels come in many forms. In the above embodiments, the convolution kernel is expressed in the form of a one-dimensional horizontal vector, and the convolution kernel can also be expressed in the form of a multi-dimensional matrix. However, no matter what form the convolution kernel is expressed in, the convolution operation process can be expressed by the above formula (1). When the convolution kernel is a multi-dimensional matrix, the multi-dimensional matrix can be converted into a one-dimensional horizontal vector.

图3示意性示出了本公开实施例的多模干涉耦合器的结构示意图。Fig. 3 schematically shows a schematic structural diagram of a multimode interference coupler according to an embodiment of the present disclosure.

如图3所示,多模干涉耦合器4包括N个输入波导41、M个输出波导42和多模干涉区43。N个输入波导41和M个输出波导42分别以预设间距安装在多模干涉区43两侧。改变输入波导41和输出波导42的安装位置可实现对预设卷积矩阵的调整。多模干涉区43的长L、宽W和高H可以被设计,以实现对预设卷积矩阵的调整。As shown in FIG. 3 , the multimode interference coupler 4 includes N input waveguides 41 , M output waveguides 42 and a multimode interference region 43 . N input waveguides 41 and M output waveguides 42 are respectively installed on both sides of the multimode interference region 43 at a preset distance. Changing the installation positions of the input waveguide 41 and the output waveguide 42 can realize the adjustment of the preset convolution matrix. The length L, width W and height H of the multi-mode interference region 43 can be designed to realize the adjustment of the preset convolution matrix.

图4示意性示出了本公开另一实施例的多模干涉耦合器的结构示意图。Fig. 4 schematically shows a structural diagram of a multimode interference coupler according to another embodiment of the present disclosure.

如图4所示,多模干涉区43包括多个多模干涉单元和多个调控件,多个调控件分别用于调整多个多模干涉单元的折射率。多个多模干涉单元和多个调控件分别一一对应。As shown in FIG. 4 , the multimode interference area 43 includes multiple multimode interference units and multiple adjustment parts, and the multiple adjustment parts are respectively used to adjust the refractive index of the multiple multimode interference units. A plurality of multi-mode interference units and a plurality of adjustment controls are in one-to-one correspondence.

在本公开实施例中,调控多模干涉单元的方法包括多种。可以在每个多模干涉单元内制作电极,在电极上施加不同的电压,通过载流子注入的方式实现对折射率的调整;也可以使用相变材料或者铁电材料制作这些多模干涉区域实现对折射率的独立进行调控;还可以通过独立设计每个多模干涉单元的厚度、掺杂浓度等方式实现对每个多模干涉单元折射率的独立调整。In the embodiments of the present disclosure, there are multiple methods for regulating the multi-mode interference unit. Electrodes can be made in each multi-mode interference unit, and different voltages can be applied to the electrodes to adjust the refractive index through carrier injection; phase-change materials or ferroelectric materials can also be used to make these multi-mode interference areas The independent adjustment of the refractive index can be realized; the independent adjustment of the refractive index of each multi-mode interference unit can also be realized by independently designing the thickness and doping concentration of each multi-mode interference unit.

通过对多模干涉单元的折射率进行独立精细调控,可以实现对卷积矩阵的精细调整。Fine tuning of the convolution matrix can be achieved by independently fine-tuning the refractive index of the multimode interference unit.

图5示意性示出了本公开另一实施例的多模干涉耦合器的结构示意图。Fig. 5 schematically shows a structural diagram of a multimode interference coupler according to another embodiment of the present disclosure.

如图5所示,多模干涉耦合器4在N个输入波导41、M个输出波导42和多模干涉区43的基础上,还包括N个偏振控制器44。As shown in FIG. 5 , the multimode interference coupler 4 further includes N polarization controllers 44 on the basis of N input waveguides 41 , M output waveguides 42 and multimode interference regions 43 .

在光信号输入至输入波导41之前,将光信号先输入至偏振控制器44。偏振控制器44分别控制激光的偏振态,以实现对卷积矩阵的调整。Before the optical signal is input to the input waveguide 41 , the optical signal is first input to the polarization controller 44 . The polarization controller 44 separately controls the polarization state of the laser light, so as to realize the adjustment of the convolution matrix.

探测器阵列5将接收到光信号转换为电信号,从而将卷积结果转换到电域。The detector array 5 converts the received optical signal into an electrical signal, thereby converting the convolution result into the electrical domain.

探测器阵列5包括M个光电探测器51,M个光电探测器51分别接收M组光信号,并探测M组光信号的强度,得到卷积运算结果。The detector array 5 includes M photodetectors 51, and the M photodetectors 51 respectively receive M groups of optical signals, detect the intensity of the M groups of optical signals, and obtain convolution operation results.

本公开还提供另一种实施例,延时模块3还可以是色散介质和分束器。在延时模块3为色散介质的情况下,激光发生模块1产生具有梳状光谱的激光,该激光包括N个激光分量。每个激光分量的波长不同,波长相邻的任意两个激光分量之间的波长间隔相等。The present disclosure also provides another embodiment, the delay module 3 may also be a dispersion medium and a beam splitter. In the case that the delay module 3 is a dispersive medium, the laser generating module 1 generates laser light with a comb spectrum, and the laser light includes N laser components. The wavelength of each laser component is different, and the wavelength interval between any two laser components with adjacent wavelengths is equal.

例如,激光发生模块1产生包括4个激光分量Laser1~Laser4,4个激光分量Laser1~Laser4的波长分别为λ1、λ2、λ3和λ4。其中,λ1、λ2、λ3和λ4中任意波长相邻的两个激光分量之间的波长间隔相同。可理解地,λ1、λ2、λ3和λ4之间的波长关系满足Δλ=λ12=λ23=λ34For example, the laser generating module 1 generates four laser components Laser 1 - Laser 4 , and the wavelengths of the four laser components Laser 1 - Laser 4 are λ 1 , λ 2 , λ 3 and λ 4 respectively. Wherein, the wavelength interval between two adjacent laser components of any wavelength among λ 1 , λ 2 , λ 3 and λ 4 is the same. Understandably, the wavelength relationship among λ 1 , λ 2 , λ 3 and λ 4 satisfies Δλ=λ 122334 .

激光发生模块1可以是多波长激光器,也可以是N个激光器,还可以是主动锁模激光器、光频梳、高速直调激光器以及激光器与电光调制器的组合。The laser generating module 1 can be a multi-wavelength laser, or N lasers, or an active mode-locked laser, an optical frequency comb, a high-speed directly modulated laser, or a combination of a laser and an electro-optic modulator.

色散介质包括但不限于色散补偿光纤、啁啾光纤光栅、普通单模光纤和多模光纤中的至少一种。色散介质对N个激光分量进行不同程度的延时。The dispersion medium includes but not limited to at least one of dispersion compensating fiber, chirped fiber grating, common single-mode fiber and multimode fiber. The dispersive medium delays the N laser components to different degrees.

本公开提供一种实现波长相邻的任意两个激光分量之间具有一个数据输入周期的延迟的示例性方法。但本公开对具体的激光延时方法不作限定。The present disclosure provides an exemplary method to achieve a delay of one data input period between any two laser components of adjacent wavelengths. However, the present disclosure does not limit the specific laser delay method.

例如,可以设置使激光分量与色散介质之间满足:For example, it can be set to satisfy the relationship between the laser component and the dispersive medium:

Figure BDA0003418647360000101
Figure BDA0003418647360000101

其中,Δλ为波长相邻的任意两个激光分量之间的波长间隔,fk为待处理数据的输入频率,

Figure BDA0003418647360000102
为色散介质的色散系数和L为色散介质的长度。可理解地,波长相邻的任意两个激光分量之间的延迟量为1/f。Among them, Δλ is the wavelength interval between any two laser components with adjacent wavelengths, f k is the input frequency of the data to be processed,
Figure BDA0003418647360000102
is the dispersion coefficient of the dispersive medium and L is the length of the dispersive medium. Understandably, the retardation between any two laser components with adjacent wavelengths is 1/f.

需要说明的是,在延时模块3为分束器和N-1个光延时线的情况下,需要先将激光根据波长分束为N束,再通过光延时线进行延时,将延时后的N束激光输入至多模干涉耦合器4。在延时模块3为色散介质和分束器的情况下,直接对调制后的激光进行群速度色散,将色散后的激光分束为N束,将N束激光输入至多模干涉耦合器4。It should be noted that, in the case that the delay module 3 is a beam splitter and N-1 optical delay lines, it is necessary to split the laser beam into N beams according to the wavelength, and then delay the time through the optical delay line. The delayed N beams of laser light are input to the multimode interference coupler 4 . When the delay module 3 is a dispersive medium and a beam splitter, the modulated laser beam is directly subjected to group velocity dispersion, the dispersed laser beam is split into N beams, and the N beams of laser light are input to the multimode interference coupler 4 .

本公开还提供另一种实施例,激光发生模块1可以是单波长激光器,单波长激光器产生单个波长激光。延时模块3为多个光延时线。The present disclosure also provides another embodiment, the laser generating module 1 may be a single-wavelength laser, and the single-wavelength laser generates laser light with a single wavelength. The delay module 3 is a plurality of optical delay lines.

在光延时线对激光进行延时之前,还需要通过光耦合器,将调制后的单波长激光功分为N束,光延时线对N束激光分别进行延时。Before the optical delay line delays the laser light, the modulated single-wavelength laser power needs to be divided into N beams through an optical coupler, and the optical delay line delays the N beams of laser light respectively.

图6示意性示出了根据本公开另一实施例的基于多模干涉的卷积运算装置的示意图。Fig. 6 schematically shows a schematic diagram of a convolution operation device based on multi-mode interference according to another embodiment of the present disclosure.

如图6所示,基于多模干涉的卷积运算装置100包括激光发生模块1、数据输入模块2、延时模块3和N个光耦合器6、R个多模干涉耦合器4和R个探测器阵列5,R个多模干涉耦合器4并联。As shown in Figure 6, the multimode interference-based convolution computing device 100 includes a laser generation module 1, a data input module 2, a delay module 3, N optical couplers 6, R multimode interference couplers 4 and R A detector array 5 and R multimode interference couplers 4 are connected in parallel.

激光发生模块1、数据输入模块2、延时模块3和探测器阵列5参见上一基于多模干涉的卷积运算装置的实施例,在此不再赘述。For the laser generating module 1 , the data input module 2 , the delay module 3 and the detector array 5 , please refer to the previous embodiment of the multi-mode interference-based convolution operation device, which will not be repeated here.

延时模块3对N个激光分量延时后得到N束第二激光,N个光耦合器6分别将N束第二激光功分为R束。通过每一光耦合器6功分的R束激光分别输入R个多模干涉耦合器4,实现多个卷积核的并行卷积计算。The delay module 3 delays the N laser components to obtain N second laser beams, and the N optical couplers 6 divide the N second laser beams into R beams respectively. R beams of laser light divided by 6 power points of each optical coupler are respectively input into R multimode interference couplers 4 to realize parallel convolution calculation of multiple convolution kernels.

本公开提供了一种详细的卷积运算方法,适用于上述的卷积运算装置,图7示意性示出了根据本公开实施例的卷积运算方法的流程图。The present disclosure provides a detailed convolution operation method, which is suitable for the above-mentioned convolution operation device. FIG. 7 schematically shows a flowchart of the convolution operation method according to an embodiment of the present disclosure.

如图7所示,卷积运算方法至少包括以下步骤:As shown in Figure 7, the convolution operation method at least includes the following steps:

S1,通过激光发生模块产生第一激光。S1, generating the first laser light through the laser generating module.

S2,通过数据输入模块将待处理数据调制到第一激光上。S2. Modulate the data to be processed onto the first laser through the data input module.

S3,通过延时模块对调制后的第一激光进行延时和分束,得到N束第二激光。S3. Delaying and beam-splitting the modulated first laser light through a delay module to obtain N beams of second laser light.

S4,通过至少一个多模干涉耦合器根据预设卷积矩阵同时对N束第二激光进行多模干涉,输出M组光信号,M为正整数。S4. Simultaneously perform multi-mode interference on N beams of second laser light through at least one multi-mode interference coupler according to a preset convolution matrix, and output M groups of optical signals, where M is a positive integer.

S5,通过探测器阵列包括M个光电探测器,M个光电探测器分别接收M组光信号,并探测M组光信号的强度,得到卷积运算结果。S5. The detector array includes M photodetectors, and the M photodetectors respectively receive M groups of optical signals, and detect the intensity of the M groups of optical signals, to obtain a convolution operation result.

需要说明的是,本公开的实施例中多模干涉的卷积运算方法与本公开的实施例中多模干涉的卷积运算装置部分是相对应的,多模干涉的卷积运算方法的描述具体参考多模干涉的卷积运算装置部分,在此不再赘述。It should be noted that the convolution operation method of multi-mode interference in the embodiment of the present disclosure corresponds to the convolution operation device part of the multi-mode interference in the embodiment of the present disclosure, and the description of the convolution operation method of multi-mode interference For details, refer to the convolution operation device part of the multi-mode interference, which will not be repeated here.

对于本公开的实施例,还需要说明的是,在不冲突的情况下,本公开的实施例及实施例中的特征可以相互组合以得到新的实施例。Regarding the embodiments of the present disclosure, it should also be noted that, under the condition of no conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other to obtain new embodiments.

最后应说明的是,以上实施例仅用以说明本公开的技术方案而非限制,尽管参照较佳实施例对本公开进行了详细说明,本领域的普通技术人员应当理解,可以对本公开的技术方案进行修改或等同替换,而不脱离本公开技术方案的精神和范围。Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure and not to limit them. Although the present disclosure has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present disclosure can be Modifications or equivalent replacements may be made without departing from the spirit and scope of the technical solutions of the present disclosure.

Claims (10)

1.一种基于多模干涉的卷积运算装置,其特征在于,包括:1. A convolution operation device based on multimode interference, characterized in that, comprising: 激光发生模块,用于产生第一激光;a laser generating module, configured to generate the first laser; 数据输入模块,用于将待处理数据调制到所述第一激光上;a data input module, configured to modulate data to be processed onto the first laser; 延时模块,用于对调制后的第一激光进行延时和分束,得到N束第二激光;a delay module, configured to delay and split the modulated first laser light to obtain N beams of second laser light; 至少一个多模干涉耦合器,用于根据预设卷积矩阵同时对所述N束第二激光进行多模干涉,输出M组光信号;以及At least one multimode interference coupler, configured to simultaneously perform multimode interference on the N beams of second laser light according to a preset convolution matrix, and output M groups of optical signals; and 探测器阵列包括M个光电探测器,所述M个光电探测器用于分别接收所述M组光信号,并探测所述M组光信号的强度,得到卷积运算结果;The detector array includes M photodetectors, and the M photodetectors are used to respectively receive the M groups of optical signals and detect the intensity of the M groups of optical signals to obtain a convolution operation result; 其中,M和N为正整数。Wherein, M and N are positive integers. 2.根据权利要求1所述的卷积运算装置,其特征在于,所述多模干涉耦合器包括N个输入端口和M个输出端口;2. The convolution operation device according to claim 1, wherein the multimode interference coupler comprises N input ports and M output ports; 所述多模干涉耦合器,用于通过所述N个输入端口分别接收所述N束第二激光,并分别根据N组预设分光比将所述N束第二激光分束至所述M个输出端口,得到M组光信号;The multimode interference coupler is configured to respectively receive the N beams of second laser light through the N input ports, and split the N beams of second laser light to the M according to N sets of preset beam splitting ratios respectively. output ports to obtain M groups of optical signals; 其中,所述N组预设分光比与所述N束第二激光一一对应。Wherein, the N sets of preset splitting ratios correspond to the N beams of second laser light one by one. 3.根据权利要求1所述的卷积运算装置,其特征在于,所述预设卷积矩阵包括M个卷积核;3. The convolution operation device according to claim 1, wherein the preset convolution matrix includes M convolution kernels; 所述多模干涉耦合器,用于分别根据所述M个卷积核调整所述N束第二激光的强度,得到M组光信号;The multimode interference coupler is configured to adjust the intensity of the N beams of second laser light according to the M convolution kernels respectively, to obtain M groups of optical signals; 其中,所述M个卷积核与所述M组光信号一一对应。Wherein, the M convolution kernels are in one-to-one correspondence with the M groups of optical signals. 4.根据权利要求1所述的卷积运算装置,其特征在于,所述多模干涉耦合器包括N个输入波导、M个输出波导和多模干涉区;4. The convolution operation device according to claim 1, wherein the multimode interference coupler comprises N input waveguides, M output waveguides and a multimode interference region; 所述N个输入波导和M个输出波导分别以预设间距安装在所述多模干涉区两侧。The N input waveguides and the M output waveguides are respectively installed on both sides of the multi-mode interference area at preset intervals. 5.根据权利要求4所述的卷积运算装置,其特征在于,所述多模干涉区包括多个多模干涉单元和多个调控件,所述多个调控件分别用于调整所述多个多模干涉单元的折射率。5. The convolution operation device according to claim 4, wherein the multimode interference area comprises a plurality of multimode interference units and a plurality of adjustment parts, and the plurality of adjustment parts are respectively used to adjust the multimode interference The refractive index of a multimode interference element. 6.根据权利要求2-5任一项所述的卷积运算装置,其特征在于,所述多模干涉耦合器还包括N个偏振控制器;6. The convolution operation device according to any one of claims 2-5, wherein the multimode interference coupler further comprises N polarization controllers; 所述N个偏振控制器,用于分别控制所述N束第二激光的偏振态。The N polarization controllers are used to respectively control the polarization states of the N beams of second laser light. 7.根据权利要求1所述的卷积运算装置,其特征在于,在所述至少一个多模干涉耦合器包括多个多模干涉耦合器的情况下,所述多个多模干涉耦合器并联。7. The convolution operation device according to claim 1, wherein, when the at least one multimode interference coupler comprises a plurality of multimode interference couplers, the plurality of multimode interference couplers are connected in parallel . 8.根据权利要求1所述的卷积运算装置,其特征在于,所述数据输入模块包括:8. The convolution operation device according to claim 1, wherein the data input module comprises: 任意波形产生器,用于将待处理数据转换为待处理信号;Arbitrary waveform generator for converting data to be processed into signals to be processed; 电光调制器,用于接收待所述待处理信号和所述第一激光,将所述待处理信号调制到所述第一激光上。An electro-optic modulator, configured to receive the signal to be processed and the first laser light, and modulate the signal to be processed onto the first laser light. 9.根据权利要求8所述的卷积运算装置,其特征在于,所述延时模块包括分束器和至少N-1个光延时线,所述待处理数据包括连续输入的N位数据,每位数据持续的时间为一个输入周期;9. The convolution computing device according to claim 8, wherein the delay module includes a beam splitter and at least N-1 optical delay lines, and the data to be processed includes continuously input N-bit data , the duration of each bit of data is one input cycle; 所述分束器用于将调制后的第一激光分为N束,得到N个激光分量;The beam splitter is used to divide the modulated first laser light into N beams to obtain N laser components; 所述至少N-1个光延时线,用于对所述N个激光分量分别进行延时,得到N束第二激光;The at least N-1 optical delay lines are used to respectively delay the N laser components to obtain N beams of second laser light; 其中,所述N束第二激光的N-1束第二激光被依次延时一个所述输入周期,以使所述N位数据同时输入所述多模干涉耦合器。Wherein, N−1 second laser beams of the N second laser beams are sequentially delayed by one input period, so that the N-bit data is simultaneously input into the multimode interference coupler. 10.一种基于多模干涉的卷积运算方法,其特征在于,基于权利要求1-9中任意一项所述的基于多模干涉的卷积运算装置,所述方法包括:10. A convolution operation method based on multi-mode interference, characterized in that, based on the multi-mode interference-based convolution operation device according to any one of claims 1-9, the method comprises: 通过激光发生模块产生第一激光;generating the first laser light through the laser generating module; 通过数据输入模块将待处理数据调制到所述第一激光上;Modulating data to be processed onto the first laser through a data input module; 通过延时模块对调制后的第一激光进行延时和分束,得到N束第二激光;Delaying and splitting the modulated first laser light through a delay module to obtain N beams of second laser light; 通过至少一个多模干涉耦合器根据预设卷积矩阵同时对所述N束第二激光进行多模干涉,输出M组光信号;Simultaneously perform multimode interference on the N beams of second laser light through at least one multimode interference coupler according to a preset convolution matrix, and output M groups of optical signals; 通过探测器阵列包括M个光电探测器,所述M个光电探测器分别接收所述M组光信号,并探测所述M组光信号的强度,得到卷积运算结果;The detector array includes M photodetectors, the M photodetectors respectively receive the M groups of optical signals, and detect the intensity of the M groups of optical signals to obtain a convolution operation result; 其中,M和N为正整数。Wherein, M and N are positive integers.
CN202111558032.XA 2021-12-17 2021-12-17 Convolution operation device and convolution operation method based on multi-mode interference Pending CN116266771A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111558032.XA CN116266771A (en) 2021-12-17 2021-12-17 Convolution operation device and convolution operation method based on multi-mode interference

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111558032.XA CN116266771A (en) 2021-12-17 2021-12-17 Convolution operation device and convolution operation method based on multi-mode interference

Publications (1)

Publication Number Publication Date
CN116266771A true CN116266771A (en) 2023-06-20

Family

ID=86744027

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111558032.XA Pending CN116266771A (en) 2021-12-17 2021-12-17 Convolution operation device and convolution operation method based on multi-mode interference

Country Status (1)

Country Link
CN (1) CN116266771A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118659834A (en) * 2024-08-19 2024-09-17 山东云海国创云计算装备产业创新中心有限公司 On-chip coherent light detection architecture and detection method thereof
CN119312004A (en) * 2024-12-17 2025-01-14 中国科学院半导体研究所 Reconfigurable optical convolution operation device and optical convolution operation method

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118659834A (en) * 2024-08-19 2024-09-17 山东云海国创云计算装备产业创新中心有限公司 On-chip coherent light detection architecture and detection method thereof
CN119312004A (en) * 2024-12-17 2025-01-14 中国科学院半导体研究所 Reconfigurable optical convolution operation device and optical convolution operation method

Similar Documents

Publication Publication Date Title
WO2022016894A1 (en) Photonic neural network
WO2021248810A1 (en) Calculation device, calculation method and calculation system
US20240086698A1 (en) Two-dimensional photonic neural network convolutional acceleration chip based on series connection structure
CN112101540B (en) Optical neural network chip and its calculation method
CN116266771A (en) Convolution operation device and convolution operation method based on multi-mode interference
CN114815959A (en) A method and device for accelerating photon tensor calculation based on wavelength division multiplexing
US11880116B2 (en) Optical digital/analog converter
CN114157391B (en) Beamforming device and beamforming method thereof
CN115222035B (en) Photon neural network convolution acceleration chip
US20220414442A1 (en) Optical computing apparatus and system, and computing method
CN115472603A (en) A monolithic photonic integrated chip implementing convolutional neural networks
JP2025508561A (en) Optical Encoder
CN116484931B (en) Photon matrix multiplication operation device and operation method for neural network
CN114565091A (en) Optical neural network device, chip and optical implementation method for neural network calculation
CN221079153U (en) Optical digital-to-analog converter and optical digital-to-analog conversion device
CN116266770A (en) Data processing device and data processing method thereof
CN116384459A (en) Object recognition device and object recognition method
CN115185136B (en) Double-optical-frequency comb generating device and method with adjustable repetition frequency
CN115906977B (en) An all-optical convolver based on dual-comb
CN116306858A (en) Neural network device and its operation method based on multi-mode interference
CN112799464B (en) Matrix vector multiplier based on multi-wavelength light source and operation method thereof
CN112232487B (en) Optical neural network chip and calculation method thereof
JP3201564B2 (en) Optical multiplexing circuit
CN114610110A (en) Optical computing device
CN112783260A (en) Optical computing device, optical computing method and computing system

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