WO2023019799A1 - 一种哈达玛积的实现方法、设备及存储介质 - Google Patents

一种哈达玛积的实现方法、设备及存储介质 Download PDF

Info

Publication number
WO2023019799A1
WO2023019799A1 PCT/CN2021/134194 CN2021134194W WO2023019799A1 WO 2023019799 A1 WO2023019799 A1 WO 2023019799A1 CN 2021134194 W CN2021134194 W CN 2021134194W WO 2023019799 A1 WO2023019799 A1 WO 2023019799A1
Authority
WO
WIPO (PCT)
Prior art keywords
hadamard product
microring
microring resonator
light intensity
processed
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.)
Ceased
Application number
PCT/CN2021/134194
Other languages
English (en)
French (fr)
Inventor
陈静静
吴睿振
黄萍
王凛
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.)
Suzhou Wave Intelligent Technology Co Ltd
Original Assignee
Suzhou Wave Intelligent Technology 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 Suzhou Wave Intelligent Technology Co Ltd filed Critical Suzhou Wave Intelligent Technology Co Ltd
Priority to US18/267,431 priority Critical patent/US11893479B1/en
Publication of WO2023019799A1 publication Critical patent/WO2023019799A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/067—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization

Definitions

  • the present application relates to the technical field of optoelectronic chips, in particular to a Hadamard product realization method, device and storage medium.
  • Photonic computing chips are the foundation and core of the modern electronic information industry. With the rapid development of globalization and technology, the amount of data to be processed is increasing rapidly, and the corresponding data processing models and algorithms are also increasing. As a result, the requirements for computing power and power consumption continue to increase. Photonic computing chips use photons as the information carrier, which has the advantages of high-speed parallelism and low power consumption, so it is considered to be the most promising solution for high-speed, large data volume, and artificial intelligence computing processing in the future.
  • Optical Neural Network Today's most common industrial solution for Optical Neural Network (ONN) is to set up proprietary devices, but it is generally only suitable for solving the multiplication and addition operation based on convolution operations. Although the largest number of operations in Artificial Neural Network (ANN) comes from convolution operations, there are a large number of Hadamard product (Hadamard product) in networks such as Long Short-Term Memory (LSTM). ) operation.
  • ANN Artificial Neural Network
  • Hadamard product Hadamard product
  • LSTM Long Short-Term Memory
  • the purpose of this application is to provide a Hadamard product realization method, device and storage medium, which can use the microring resonator structure to realize the simulation solution suitable for the Hadamard product in the optical neural network.
  • the specific plan is as follows:
  • a method for realizing Hadamard product comprising:
  • the microring resonator structure includes a plurality of microring resonator groups composed of two microring resonators with the same radius ;
  • a corresponding current is applied to the microring resonator structure, and a Hadamard product result is obtained according to the output light intensity.
  • the radii of each of the microring resonator groups are different.
  • the wavelength of the optical signal to be processed corresponds to the radius of the microring resonator group one by one;
  • the number of types of wavelengths of the optical signal to be processed is the same as the number of the microring resonator groups.
  • the microring resonator includes a straight waveguide and a microring waveguide.
  • the straight waveguides of all the microring resonators in the microring resonator structure are the same common straight waveguide;
  • the common straight waveguide has an input port and a through port.
  • the radius of each microring resonator group increases sequentially along the direction from the input port to the through port.
  • the transfer function between the light intensity passing through the through port and the light intensity at the input port is between 0 and 1.
  • the microring resonator structure while applying the corresponding current to the microring resonator structure, it also includes:
  • a transfer function of the light intensity of the through port and the light intensity of the input port is controlled to achieve a target transfer function.
  • the embodiment of the present application also provides a device for realizing Hadamard product, including a processor and a memory, wherein, when the processor executes the computer program stored in the memory, the above-mentioned Hadamard product as provided in the embodiment of the present application is realized. implementation method.
  • the embodiment of the present application also provides a computer-readable storage medium for storing a computer program, wherein when the computer program is executed by a processor, the implementation method of the above-mentioned Hadamard product as provided in the embodiment of the present application is realized.
  • a Hadamard product implementation method includes: acquiring multiple optical signals of different wavelengths to be processed; inputting the optical signals to be processed into a wavelength division multiplexer; The demultiplexer feeds the optical signal to be processed to the microring resonator structure; the microring resonator structure includes a plurality of microring resonator groups composed of two microring resonators with the same radius; applying to the microring resonator structure Corresponding current, according to the output light intensity, the result of Hadamard product is obtained.
  • the microring resonator is used as the basis for realizing the artificial neural network scheme, and the wavelength division multiplexer is used to feed the optical signal to be processed to the microring resonator structure, and the effective refractive index and phase of the microring resonator can be changed by current heating , according to the light intensity of the output optical signal, the result of Hadamard product can be obtained, and then the simulation solution suitable for Hadamard product in optical neural network is realized.
  • the present application also provides a corresponding device and a computer-readable storage medium for the implementation method of the Hadamard product, which further makes the above method more practical, and the device and the computer-readable storage medium have corresponding advantages.
  • Fig. 1 is the flowchart of the implementation method of the Hadamard product provided by the embodiment of the present application;
  • Fig. 2 is the result schematic diagram of the microring resonator provided by the embodiment of the present application.
  • FIG. 3 is a light intensity distribution diagram of the microring resonator provided in the embodiment of the present application under the condition of non-resonance;
  • FIG. 4 is a light intensity distribution diagram of the microring resonator provided in the embodiment of the present application under the condition of resonance;
  • FIG. 5 is a schematic diagram of the change of the transfer function of the microring resonator with the phase provided by the embodiment of the present application;
  • FIG. 6 is a schematic structural diagram of a device for implementing a Hadamard product using a wavelength division multiplexer and a microring resonator structure according to an embodiment of the present application.
  • This application provides a kind of realization method of Hadamard product (Hadamard product), as shown in Figure 1, comprises the following steps:
  • WDM Wavelength Division Multiplexing
  • the microring resonator structure includes a plurality of microring resonator groups composed of two microring resonators with the same radius;
  • the microring resonator is used as the basis for realizing the artificial neural network scheme, and the optical signal to be processed is fed to the microring resonator structure by using a wavelength division multiplexer.
  • Current heating can change the effective refractive index and phase of the microring resonator, and the result of the Hadamard product can be obtained according to the light intensity of the output optical signal, and then the simulation solution suitable for the Hadamard product in the optical neural network is realized.
  • the Hadamard product is a kind of matrix operation.
  • the Hadamard product operation is two matrices of the same order, and the elements at the corresponding positions are multiplied.
  • the microring resonator MRR may include a straight waveguide and a microring waveguide.
  • the MRR is an All-pass silicon-based MRR.
  • Fig. 3 and Fig. 4 respectively show the light intensity distribution diagrams of the All-pass type MRR in the case of non-resonance and resonance.
  • the optical signal enters from the input end.
  • the microring When the wavelength of the incident light satisfies the resonance condition, most of the optical signal of this wavelength is confined in the microring, and there is almost no output at the output end; if the resonance condition is not satisfied, the optical signal entering the microring is The light waves input by destructive interference will be directly output from the through end, so the microring has the most basic filtering function.
  • the resonance equation of the microring is:
  • ⁇ i is the wavelength of the optical signal
  • m is the integer multiple of the wavelength of the optical signal
  • R is the radius of the MRR
  • n eff is the effective refractive index of the light
  • the light satisfying the wavelength of the formula (2) satisfies the resonance condition and will be limited in the microring. It can be seen from the resonance equation (2) that different wavelengths correspond to different microring radii.
  • the MRR will be heated, resulting in a change in the effective refractive index n eff of the light, causing a shift in the resonance wavelength, resulting in the output of the light part confined in the microring from the through end.
  • ⁇ i is the phase of the MRR
  • r is the self-coupling coefficient
  • a defines the propagation loss of the loop and directional coupler.
  • the value range of the transfer function is [0,1].
  • phase ⁇ i in formula (1) is:
  • Fig. 5 shows the variation diagram of the All-pass microring transfer function T n ( ⁇ i ) with the phase ⁇ i .
  • the radii of each microring resonator group are different.
  • the wavelength of the optical signal to be processed corresponds to the radius of the microring resonator group one by one; the number of types of the wavelength of the optical signal to be processed is the same as the number of the microring resonator group.
  • a ⁇ B [a 1 b 1 a 2 b 2 a 3 b 3 ] (6)
  • This application realizes the calculation of the Hadamard product according to the properties of silicon-based MRR and WDM. As shown in Figure 6, it includes six MRRs, two of which are a group of the same radius, and the radius lengths are R 1 , R 2 , and R 3 .
  • the straight waveguides of all MRRs in the MRR structure may be the same shared straight waveguide; the shared straight waveguide has an input port and a through port.
  • the transfer function of the light intensity through the through port to the light intensity at the input port is between 0 and 1.
  • the laser emits optical signals with wavelengths of ⁇ 1 , ⁇ 2 , and ⁇ 3 respectively.
  • the light intensities of the three optical signals are all 1. It is then fed to the MRR structure via WDM.
  • the MRR structure consists of three MRR groups with different radius lengths. The radius lengths are R 1 , R 2 , and R 3 . The radius lengths can be increased sequentially along the direction from the input port to the through port, and satisfy:
  • the light intensities of the optical signals of ⁇ 1 , ⁇ 2 , and ⁇ 3 passing through the through end are respectively That is, a 1 b 1 , a 2 b 2 , and a 3 b 3 are the results of the Hadamard product of A and B.
  • the implementation method of the above-mentioned Hadamard product while applying the corresponding current to the microring resonator structure, it also includes: controlling the light intensity of the through port and the input port The transfer function of the light intensity reaches the target transfer function. In this way, the transfer function of the MRR structure is controlled as the target transfer function, which can ensure that the applied current value can accurately obtain the required Hadamard product result.
  • the embodiment of the present application also discloses a device for realizing Hadamard product, including a processor and a memory; wherein, when the processor executes the computer program stored in the memory, the method for realizing Hadamard product disclosed in the foregoing embodiments is realized.
  • the present application also discloses a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, the implementation method of the Hadamard product disclosed above is realized.
  • each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same or similar parts of each embodiment can be referred to each other.
  • the description is relatively simple, and for relevant details, please refer to the description of the method part.
  • RAM random access memory
  • ROM read-only memory
  • EEPROM electrically programmable ROM
  • EEPROM electrically erasable programmable ROM
  • registers hard disk, removable disk, CD-ROM, or any other Any other known storage medium.
  • the implementation method of a Hadamard product includes: acquiring multiple optical signals of different wavelengths to be processed; inputting the optical signals to be processed into a wavelength division multiplexer; using the wavelength division multiplexer Feed the optical signal to be processed to the microring resonator structure; the microring resonator structure includes a plurality of microring resonator groups consisting of two microring resonators with the same radius; apply a corresponding current to the microring resonator structure, According to the output light intensity, the result of Hadamard product is obtained.
  • the microring resonator is used as the basis for realizing the artificial neural network scheme, and the wavelength division multiplexer is used to feed the optical signal to be processed to the microring resonator structure, and the effective refractive index and phase of the microring resonator can be changed by current heating. According to the light intensity of the output optical signal, the result of the Hadamard product can be obtained, and then the simulation solution suitable for the Hadamard product in the optical neural network is realized.
  • the present application also provides a corresponding device and a computer-readable storage medium for the implementation method of the Hadamard product, which further makes the above method more practical, and the device and the computer-readable storage medium have corresponding advantages.

Landscapes

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

Abstract

一种哈达玛积的实现方法、设备及存储介质，该方法包括：获取多种不同波长的待处理光信号（S101）；将待处理光信号输入至波分复用器（S102）；利用波分复用器将待处理光信号馈送到微环谐振器结构；微环谐振器结构包括多个由两个半径相同的微环谐振器组成的微环谐振器组（S103）；向微环谐振器结构施加相应的电流，根据输出的光强得到哈达玛积的结果（S104）。这样以微环谐振器为实现人工神经网络方案的基础，利用波分复用器将待处理光信号馈送到微环谐振器结构，通过电流加热可以改变微环谐振器的有效折射率和相位，根据输出的光信号的光强即可得到哈达玛积的结果，进而实现了适用于光学神经网络中哈达玛积的模拟解决方案。

Description

一种哈达玛积的实现方法、设备及存储介质
本申请要求在2021年8月18日提交中国专利局、申请号为202110945843.9、发明名称为“一种哈达玛积的实现方法、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及光电芯片技术领域，特别是涉及一种哈达玛积的实现方法、设备及存储介质。
背景技术
芯片是现代电子信息产业的基础和核心。随着全球化以及科技的高速发展，需要处理的数据量在急剧增加，相应的数据处理模型和算法也在不断增加，带来的结果就是对算力和功耗的要求不断提高。光子计算芯片以光子为信息的载体具有高速并行、低功耗的优势，因此被认为是未来高速、大数据量、人工智能计算处理的最具有前景的方案。
现今光子神经网络(Optical Neural Network，ONN)方面最常见的工业解决方案是设置专有器件，但是其一般只适合于解决基于卷积运算的乘加运算部分。人工神经网络(Artificial Neural Network，ANN)中虽然最大量的运算来源于卷积运算，但是在长短期记忆人工神经网络(Long Short-Term Memory，LSTM)等网络中存在大量哈达玛积(Hadamard product)的运算。
因此，如何实现神经网络中哈达玛积的模拟运算，是本领域技术人员亟待解决的技术问题。
发明内容
有鉴于此，本申请的目的在于提供一种哈达玛积的实现方法、设备及存储介质，可以利用微环谐振器结构实现适用于光学神经网络中哈达玛积的模 拟解决方案。其具体方案如下：
一种哈达玛积的实现方法，包括：
获取多种不同波长的待处理光信号；
将所述待处理光信号输入至波分复用器；
利用所述波分复用器将所述待处理光信号馈送到微环谐振器结构；所述微环谐振器结构包括多个由两个半径相同的微环谐振器组成的微环谐振器组；
向所述微环谐振器结构施加相应的电流，根据输出光强得到哈达玛积的结果。
优选地，在本申请实施例提供的上述哈达玛积的实现方法中，每个所述微环谐振器组的半径不同。
优选地，在本申请实施例提供的上述哈达玛积的实现方法中，所述待处理光信号的波长与所述微环谐振器组的半径一一对应；
所述待处理光信号的波长的种类数量与所述微环谐振器组的个数相同。
优选地，在本申请实施例提供的上述哈达玛积的实现方法中，所述微环谐振器包括一根直波导和一个微环波导。
优选地，在本申请实施例提供的上述哈达玛积的实现方法中，所述微环谐振器结构中所有的所述微环谐振器的所述直波导为同一根共用直波导；
所述共用直波导具有输入端口和直通端口。
优选地，在本申请实施例提供的上述哈达玛积的实现方法中，各所述微环谐振器组的半径沿着所述输入端口向所述直通端口的方向依次变大。
优选地，在本申请实施例提供的上述哈达玛积的实现方法中，通过所述直通端口的光强与所述输入端口的光强的传递函数在0和1之间。
优选地，在本申请实施例提供的上述哈达玛积的实现方法中，在所述向所述微环谐振器结构施加相应的电流的同时，还包括：
控制所述直通端口的光强与所述输入端口的光强的传递函数达到目标传递函数。
本申请实施例还提供了一种哈达玛积的实现设备，包括处理器和存储器，其中，所述处理器执行所述存储器中存储的计算机程序时实现如本申请实施例提供的上述哈达玛积的实现方法。
本申请实施例还提供了一种计算机可读存储介质，用于存储计算机程序，其中，所述计算机程序被处理器执行时实现如本申请实施例提供的上述哈达玛积的实现方法。
从上述技术方案可以看出，本申请所提供的一种哈达玛积的实现方法，包括：获取多种不同波长的待处理光信号；将待处理光信号输入至波分复用器；利用波分复用器将待处理光信号馈送到微环谐振器结构；微环谐振器结构包括多个由两个半径相同的微环谐振器组成的微环谐振器组；向微环谐振器结构施加相应的电流，根据输出光强得到哈达玛积的结果。
本申请以微环谐振器为实现人工神经网络方案的基础，利用波分复用器将待处理光信号馈送到微环谐振器结构，通过电流加热可以改变微环谐振器的有效折射率和相位，根据输出的光信号的光强即可得到哈达玛积的结果，进而实现了适用于光学神经网络中哈达玛积的模拟解决方案。此外，本申请还针对哈达玛积的实现方法提供了相应的设备及计算机可读存储介质，进一步使得上述方法更具有实用性，该设备及计算机可读存储介质具有相应的优点。
附图说明
为了更清楚地说明本申请实施例或相关技术中的技术方案，下面将对实施例或相关技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据提供的附图获得其他的附图。
图1为本申请实施例提供的哈达玛积的实现方法的流程图；
图2为本申请实施例提供的微环谐振器的结果示意图；
图3为本申请实施例提供的微环谐振器在非谐振情况下光强分布图；
图4为本申请实施例提供的微环谐振器在谐振情况下光强分布图；
图5为本申请实施例提供的微环谐振器的传递函数随相位的变化示意图；
图6为本申请实施例提供的利用波分复用器和微环谐振器结构实现哈达玛积的装置结构示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请提供一种哈达玛积(Hadamard product)的实现方法，如图1所示，包括以下步骤：
S101、获取多种不同波长的待处理光信号；
S102、将待处理光信号输入至波分复用器(Wavelength Division Multiplexing，WDM)；
S103、利用波分复用器将待处理光信号馈送到微环谐振器结构；微环谐振器结构包括多个由两个半径相同的微环谐振器组成的微环谐振器组；
S104、向微环谐振器结构施加相应的电流，根据输出光强得到哈达玛积的结果。
在本申请实施例提供的上述哈达玛积的实现方法中，以微环谐振器为实现人工神经网络方案的基础，利用波分复用器将待处理光信号馈送到微环谐振器结构，通过电流加热可以改变微环谐振器的有效折射率和相位，根据输出的光信号的光强即可得到哈达玛积的结果，进而实现了适用于光学神经网络中哈达玛积的模拟解决方案。
需要说明的是，哈达玛积是一种矩阵的运算，定义设A，B∈C m×n，且A＝[a ij]，B＝[b ij]，A，B的哈达玛积记作AοB，计算方式如下：
Figure PCTCN2021134194-appb-000001
哈达玛积运算就是两个同阶的矩阵，对应位置的元素相乘。
进一步地，在具体实施时，在本申请实施例提供的上述哈达玛积的实现方法中，如图2所示，微环谐振器MRR可以包括一根直波导和一个微环波导。优选地，该MRR为All-pass型的硅基MRR。图3和图4分别示出了All-pass型的MRR在非谐振以及谐振情况下光强分布图。光信号从输入端进入，当入射光的波长满足谐振条件时，该波长的光信号大部分被限制在微环中，输出端几乎无输出；如果不满足谐振条件，进入微环的光信号就会发生相消干涉输入的光波直接从直通端输出，因此微环具有最基本的滤波作用。
光在微环中传输时，受到微环的限制较强，当它满足绕微环传输一圈时产生的光程差是光信号波长的整数倍m这个条件时，就会发生谐振，光信号的强度会不断加强，而使其产生相互作用而加强的条件称为谐振条件，微环的谐振方程为：
2πRn eff＝mλ i    (2)
其中，λ i为光信号波长，m为光信号波长的整数倍，R为MRR的半径，n eff为光的有效折射率，满足公式(2)的波长的光即满足谐振条件，会被限制在微环中。由谐振方程(2)可知不同的波长对应的微环半径也不同。当给MRR通过电流时会加热MRR，导致光的有效折射率n eff的变化，使谐振波长发生漂移，从而导致被限制在微环中的光部分从直通端输出。
通过直通端的通孔出射的光的强度与进入全通谐振器MRR的输入端口的光强的传递函数的表达式如下：
Figure PCTCN2021134194-appb-000002
其中，φ i为MRR的相位，r是自耦合系数，a定义了环和定向耦合器的传播损耗。传递函数的取值范围为[0,1]。
当输入光信号的振幅为E in(光强为|E in| 2)，则通过MRR输出的光强为：
|E out| 2＝T n(φ i)|E in| 2     (4)
公式(1)中相位φ i的表达式为：
Figure PCTCN2021134194-appb-000003
图5示出了All-pass微环传递函数T n(φ i)随相位φ i的变化图。
当给硅基MRR通过电流时会加热MRR，导致n eff的变化，从而导致相位φ i的变化，最终影响光强的传递函数T n(φ i)。也就是说，当输入光信号的振幅为E in(光强为|E in| 2)，通过给硅基微环施加电流加热，改变传递函数T n(φ i)，从而得到想要的输出光强|E out| 2。本申请正是根据硅基MRR的这种性质实现了哈达玛积的计算。
在具体实施时，在本申请实施例提供的上述哈达玛积的实现方法中，每个微环谐振器组的半径不同。较佳地，待处理光信号的波长与所述微环谐振器组的半径一一对应；待处理光信号的波长的种类数量与微环谐振器组的个数相同。
下面以A＝[a 1 a 2 a 3]，B＝[b 1 b 2 b 3]为例，根据定义Ａ，Ｂ的哈达玛积为：
AοB＝[a 1b 1 a 2b 2 a 3b 3]       (6)
本申请根据硅基MRR性质及WDM实现了哈达玛积的计算，如图6所示，包含六个MRR，其中两个为一组半径相同，半径长度分别为R 1，R 2，R 3。
在具体实施时，MRR结构中所有MRR的直波导可以为同一根共用直波导；共用直波导具有输入端口和直通端口。通过直通端口的光强与输入端口的光强的传递函数在0和1之间。
如图6所示，激光器发射波长分别为λ 1，λ 2，λ 3的光信号，为了好理解，假设三种光信号的光强都为1。然后通过WDM馈送到MRR结构。MRR结构由三个不同半径长度的MRR组组成，半径长度分别为R 1，R 2，R 3，半径长度可以沿着输入端口向直通端口的方向依次变大，且满足：
2πR in eff＝mλ i i＝1，2，3     (7)
由公式(5)和公式(7)可知，如果不加热MRR，此时波长分别为λ 1，λ 2，λ 3的三种光信号都被限制在微环中，直通端没有光信号输出。如果分别向微环中加入电流，微环受热导致n eff的变化，使得相位φ i的变化满足：
Figure PCTCN2021134194-appb-000004
Figure PCTCN2021134194-appb-000005
Figure PCTCN2021134194-appb-000006
此时，从直通端通过的为λ 1，λ 2，λ 3的光信号的光强分别为
Figure PCTCN2021134194-appb-000007
Figure PCTCN2021134194-appb-000008
即a 1b 1，a 2b 2，a 3b 3，就是A，B的哈达玛积的结果。
进一步地，在具体实施时，在本申请实施例提供的上述哈达玛积的实现方法中，在向微环谐振器结构施加相应的电流的同时，还包括：控制直通端 口的光强与输入端口的光强的传递函数达到目标传递函数。这样控制MRR结构的传递函数为目标传递函数，可以确保施加的电流值大小能准确得到所需哈达玛积的结果。
相应地，本申请实施例还公开了一种哈达玛积的实现设备，包括处理器和存储器；其中，处理器执行存储器中存储的计算机程序时实现前述实施例公开的哈达玛积的实现方法。
关于上述方法更加具体的过程可以参考前述实施例中公开的相应内容，在此不再进行赘述。
进一步地，本申请还公开了一种计算机可读存储介质，用于存储计算机程序；计算机程序被处理器执行时实现前述公开的哈达玛积的实现方法。
关于上述方法更加具体的过程可以参考前述实施例中公开的相应内容，在此不再进行赘述。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其它实施例的不同之处，各个实施例之间相同或相似部分互相参见即可。对于实施例公开的设备、存储介质而言，由于其与实施例公开的方法相对应，所以描述的比较简单，相关之处参见方法部分说明即可。
专业人员还可以进一步意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
结合本文中所公开的实施例描述的方法或算法的步骤可以直接用硬件、处理器执行的软件模块，或者二者的结合来实施。软件模块可以置于随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可 编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
综上，本申请实施例提供的一种哈达玛积的实现方法，包括：获取多种不同波长的待处理光信号；将待处理光信号输入至波分复用器；利用波分复用器将待处理光信号馈送到微环谐振器结构；微环谐振器结构包括多个由两个半径相同的微环谐振器组成的微环谐振器组；向微环谐振器结构施加相应的电流，根据输出光强得到哈达玛积的结果。这样以微环谐振器为实现人工神经网络方案的基础，利用波分复用器将待处理光信号馈送到微环谐振器结构，通过电流加热可以改变微环谐振器的有效折射率和相位，根据输出的光信号的光强即可得到哈达玛积的结果，进而实现了适用于光学神经网络中哈达玛积的模拟解决方案。此外，本申请还针对哈达玛积的实现方法提供了相应的设备及计算机可读存储介质，进一步使得上述方法更具有实用性，该设备及计算机可读存储介质具有相应的优点。
最后，还需要说明的是，在本文中，诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
以上对本申请所提供的哈达玛积的实现方法、设备及存储介质进行了详细介绍，本文中应用了具体个例对本申请的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本申请的方法及其核心思想；同时，对于本领域的一般技术人员，依据本申请的思想，在具体实施方式及应用范围上均会有改变之处，综上所述，本说明书内容不应理解为对本申请的限制。

Claims (10)

  1. 一种哈达玛积的实现方法，其特征在于，包括：
    获取多种不同波长的待处理光信号；
    将所述待处理光信号输入至波分复用器；
    利用所述波分复用器将所述待处理光信号馈送到微环谐振器结构；所述微环谐振器结构包括多个由两个半径相同的微环谐振器组成的微环谐振器组；
    向所述微环谐振器结构施加相应的电流，根据输出光强得到哈达玛积的结果。
  2. 根据权利要求1所述的哈达玛积的实现方法，其特征在于，每个所述微环谐振器组的半径不同。
  3. 根据权利要求2所述的哈达玛积的实现方法，其特征在于，所述待处理光信号的波长与所述微环谐振器组的半径一一对应；
    所述待处理光信号的波长的种类数量与所述微环谐振器组的个数相同。
  4. 根据权利要求3所述的哈达玛积的实现方法，其特征在于，所述微环谐振器包括一根直波导和一个微环波导。
  5. 根据权利要求4所述的哈达玛积的实现方法，其特征在于，所述微环谐振器结构中所有的所述微环谐振器的所述直波导为同一根共用直波导；
    所述共用直波导具有输入端口和直通端口。
  6. 根据权利要求5所述的哈达玛积的实现方法，其特征在于，各所述微环谐振器组的半径沿着所述输入端口向所述直通端口的方向依次变大。
  7. 根据权利要求6所述的哈达玛积的实现方法，其特征在于，通过所述直通端口的光强与所述输入端口的光强的传递函数在0和1之间。
  8. 根据权利要求7所述的哈达玛积的实现方法，其特征在于，在所述向所述微环谐振器结构施加相应的电流的同时，还包括：
    控制所述直通端口的光强与所述输入端口的光强的传递函数达到目标传递函数。
  9. 一种哈达玛积的实现设备，其特征在于，包括处理器和存储器，其中，所述处理器执行所述存储器中存储的计算机程序时实现如权利要求1至8任一项所述的哈达玛积的实现方法。
  10. 一种计算机可读存储介质，其特征在于，用于存储计算机程序，其中，所述计算机程序被处理器执行时实现如权利要求1至8任一项所述的哈达玛积的实现方法。
PCT/CN2021/134194 2021-08-18 2021-11-30 一种哈达玛积的实现方法、设备及存储介质 Ceased WO2023019799A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/267,431 US11893479B1 (en) 2021-08-18 2021-11-30 Hadamard product implementation method and device, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110945843.9A CN113392965B (zh) 2021-08-18 2021-08-18 一种哈达玛积的实现方法、设备及存储介质
CN202110945843.9 2021-08-18

Publications (1)

Publication Number Publication Date
WO2023019799A1 true WO2023019799A1 (zh) 2023-02-23

Family

ID=77622824

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/134194 Ceased WO2023019799A1 (zh) 2021-08-18 2021-11-30 一种哈达玛积的实现方法、设备及存储介质

Country Status (3)

Country Link
US (1) US11893479B1 (zh)
CN (1) CN113392965B (zh)
WO (1) WO2023019799A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113392965B (zh) 2021-08-18 2021-11-19 苏州浪潮智能科技有限公司 一种哈达玛积的实现方法、设备及存储介质
CN114564071A (zh) * 2022-01-29 2022-05-31 山东云海国创云计算装备产业创新中心有限公司 一种基于mrr的克罗内克积的实现方法、装置、设备及介质
EP4703843A1 (en) 2023-04-27 2026-03-04 Lightstandard Co., Ltd. Photonic computing method, photonic computing array, and photoelectric hybrid computing method and array

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108616686A (zh) * 2016-12-12 2018-10-02 南京理工大学 一种全哈达玛矩阵编码成像方法
CN109639359A (zh) * 2019-01-07 2019-04-16 上海交通大学 基于微环谐振器的光子神经网络卷积层芯片
US20200052792A1 (en) * 2016-10-18 2020-02-13 Nippon Telegraph And Telephone Corporation Encoding Device And Decoding Device
CN113031161A (zh) * 2021-02-26 2021-06-25 北京邮电大学 一种硅基微波光子信道化芯片
CN113392965A (zh) * 2021-08-18 2021-09-14 苏州浪潮智能科技有限公司 一种哈达玛积的实现方法、设备及存储介质

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8027587B1 (en) * 2007-08-21 2011-09-27 Sandia Corporation Integrated optic vector-matrix multiplier
WO2010108093A2 (en) * 2009-03-20 2010-09-23 Cornell University Apparatus and methods for wide temperature range operation of micrometer-scale silicon electro-optic modulators
WO2013169393A1 (en) * 2012-05-07 2013-11-14 Stc.Unm Biomarker sensing based on nanofluidic amplification and resonant optical detection
CN103487985B (zh) * 2013-09-27 2016-03-16 京东方科技集团股份有限公司 透光基板、阵列基板、彩膜基板及显示装置
US12039433B2 (en) * 2020-02-20 2024-07-16 Ohio University Optical multiply and accumulate unit
US20230152667A1 (en) * 2020-04-16 2023-05-18 The George Washington University Photonic tensor core matrix vector multiplier
WO2022047378A1 (en) * 2020-08-31 2022-03-03 The Board Of Trustees Of The Leland Stanford Junior University Frequency-domain arbitrary linear transformations for photons

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200052792A1 (en) * 2016-10-18 2020-02-13 Nippon Telegraph And Telephone Corporation Encoding Device And Decoding Device
CN108616686A (zh) * 2016-12-12 2018-10-02 南京理工大学 一种全哈达玛矩阵编码成像方法
CN109639359A (zh) * 2019-01-07 2019-04-16 上海交通大学 基于微环谐振器的光子神经网络卷积层芯片
CN113031161A (zh) * 2021-02-26 2021-06-25 北京邮电大学 一种硅基微波光子信道化芯片
CN113392965A (zh) * 2021-08-18 2021-09-14 苏州浪潮智能科技有限公司 一种哈达玛积的实现方法、设备及存储介质

Also Published As

Publication number Publication date
US11893479B1 (en) 2024-02-06
CN113392965B (zh) 2021-11-19
CN113392965A (zh) 2021-09-14
US20240046084A1 (en) 2024-02-08

Similar Documents

Publication Publication Date Title
Zhou et al. Chip-scale optical matrix computation for PageRank algorithm
WO2023019799A1 (zh) 一种哈达玛积的实现方法、设备及存储介质
Xu et al. Optical patching scheme for optical convolutional neural networks based on wavelength-division multiplexing and optical delay lines
Hu et al. Integrated lithium niobate photonic computing circuit based on efficient and high-speed electro-optic conversion
Shao et al. Generalized robust training scheme using genetic algorithm for optical neural networks with imprecise components
Huang et al. Programmable matrix operation with reconfigurable time-wavelength plane manipulation and dispersed time delay
CN115641494B (zh) 一种基于微环调制器的神经网络图像处理系统
Huang et al. Sophisticated deep learning with on-chip optical diffractive tensor processing
Zhang et al. Compact, efficient, and scalable nanobeam core for photonic matrix-vector multiplication
CN113673677B (zh) 一种非线性激活函数relu的实现方法、设备及介质
Cheng et al. Photonic emulator for inverse design
Liao et al. Matrix eigenvalue solver based on reconfigurable photonic neural network
Zhang et al. Redundancy-free integrated optical convolver for optical neural networks based on arrayed waveguide grating
Vatsavai et al. Photonic reconfigurable accelerators for efficient inference of cnns with mixed-sized tensors
Zelaya et al. The Goldilocks principle of learning unitaries by interlacing fixed operators with programmable phase shifters on a photonic chip
Huang et al. Diffraction‐driven parallel convolution processing with integrated photonics
Nikkhah et al. Reconfigurable nonlinear optical element using tunable couplers and inverse-designed structure
Wang et al. Reconfigurable versatile integrated photonic computing chip
Wu et al. Scaling up for end-to-end on-chip photonic neural network inference
Yi et al. Accelerating convolutional processing by harnessing channel shifts in arrayed waveguide gratings
Shi et al. Parallel photonic convolutional processing on-chip with cross-connect architecture and cyclic AWGs
Marchesin et al. Braided interferometer mesh for robust photonic matrix-vector multiplications with non-ideal components
Zhou et al. Adaptive reconfiguration and full control of light structures in optical fibers with programmable silicon photonics
CN113705774A (zh) 光学电路搭建方法、光学电路、光信号处理方法及装置
CN116822601A (zh) 一种结合波分复用和mzi级联网络的矩阵运算加速器

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21954034

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 18267431

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21954034

Country of ref document: EP

Kind code of ref document: A1