A large-scale reconfigurable three-dimensional integrated optical neural network chip

Abstract

The application discloses a large-scale and reconfigurable three-dimensional integrated optical neural network chip, and belongs to the field of optical neural networks, and comprises an optical information input layer, a reconfigurable hidden layer, a detection output layer and a feedback control structure; the reconfigurable hidden layer comprises a plurality of cascaded waveguide structures, wherein each waveguide structure comprises an independent modulation area and a continuous coupling area, the optical waveguides of the independent modulation area are independently transmitted and independently modulated by a light modulator; after reaching the continuous coupling area, the evanescent waves of the waveguide array are continuously coupled with each other in the transmission direction and are integrally modulated by the light modulator; and the feedback control structure adjusts a light modulator loading signal according to the detected output light intensity spatial distribution information, so that the reconfigurable hidden layer identifies the current input signal. The application can greatly improve the actual computing power of the current on-chip integrated optical neural network, and also provides a solution to the problems of low reconfigurability and large system size of the current three-dimensional spatial diffraction optical neural network.

Description

Technical Field

[0001] This invention belongs to the field of optical neural networks, and more specifically, relates to a large-scale reconfigurable three-dimensional integrated optical neural network chip. Background Technology

[0002] With the rapid increase in GPU computing speed and power, artificial intelligence has become a mainstay of the new technological revolution. However, the size and performance of electrical transistors are approaching their limits and will be unable to meet the ever-increasing data volume and energy consumption constraints in the future. Light, due to its advantages of wide bandwidth, high frequency, and low energy consumption, has attracted widespread attention in the fields of communication and computing. In particular, the multi-dimensional resources and parallel transmission characteristics of light are widely used in optical neural networks based on parallel matrix multiplication, and integrated optical neural networks are a promising candidate to replace GPUs.

[0003] Currently, on-chip integrated optical neural networks are mainly based on waveguide structures and tunable optical devices, such as those based on Mach 60. Zehnder interferometer (Mach) There are two main types of networks: Zehnder Interferometers (MZI) networks and Microring Resonators (MRR) arrays. This approach offers high integration, but the network complexity and total number of devices increase rapidly with the total number of neurons, limiting scalability and often restricting it to below 100×100.

[0004] Another approach is based on free-space optical diffraction, which utilizes each pixel of spatial light as an optical neuron. Although the number of pixels in a spatial light modulator can easily reach 1000×1000, its integration still faces many challenges. For example, the approach of etching multiple rows of slit arrays on a silicon slab to achieve two-dimensional multi-plane diffraction can only load and process one-dimensional information, which greatly reduces the large-scale advantage of diffraction neural networks, and reconstructing the effective phase shift of each slit is still very difficult. Summary of the Invention

[0005] To address the shortcomings and improvement needs of existing technologies, this invention provides a large-scale reconfigurable three-dimensional integrated optical neural network chip. Its purpose is to fully utilize the three-dimensional waveguide structure in space to realize large-scale neuron interconnection in the neural network, and at the same time apply an optical modulator to realize the adjustable connection weights between neurons, thereby constructing a large-scale reconfigurable three-dimensional integrated optical neural network and improving the scalability and flexibility of the integrated optical neural network.

[0006] To achieve the above objectives, the present invention provides a large-scale reconfigurable three-dimensional integrated optical neural network chip, comprising: an optical information input layer, a reconfigurable hidden layer, a detector output layer, and a feedback control structure; The reconfigurable hidden layer comprises an m-level waveguide array, corresponding to the m fully connected layers in an optical neural network. Each waveguide array includes an independent modulation region and a continuous coupling region. The independent modulation region comprises N×M independent transmission waveguides and N×M optical modulators. The N×M independent transmission waveguides transmit the input optical signal, and no two waveguides are coupled to each other. The input optical signal is modulated by the N×M optical modulators, corresponding to a reconfigurable (N×M)×(N×M) diagonal matrix. The continuous coupling region comprises an N×M waveguide array, with the coupling relationship between the N×M inputs and N×M outputs of the waveguide array, corresponding to a fixed N×M unitary matrix. The reconfigurable hidden layer receives the input optical signal from the optical signal input layer and sequentially modulates the intensity distribution of the input optical signal through the independent modulation region and the continuous coupling region to obtain the output optical signal. The optical information input layer is used to generate an input optical signal carrying input information that is incident on the reconfigurable hidden layer; The output layer is used to detect the output optical signal.

[0007] A feedback control structure, connected to the detector output layer and the reconfigurable hidden layer, generates a feedback control signal based on the power distribution information of the output optical signal. This feedback control signal is then used to adjust the loading signal of the optical modulator, thereby modulating the optical field signal in the multi-layered cascaded N×M waveguide array. This modulates the reconfigurable hidden layer, enabling it to recognize the spatial distribution information of the current input signal's light intensity. The continuous coupling region in the reconfigurable hidden layer can also be partially modulated by the optical modulator to achieve partial modulation of the complex amplitude of the optical signal within the waveguide, corresponding to a partially reconfigurable unitary matrix.

[0008] In one embodiment, the optical modulator can be a thermoelectrode, any other optical modulator (thermally tunable, electrically tunable), or a phase change material. For thermoelectrodes and other arbitrary optical modulators, the optical signal passing through the waveguide is modulated by applying other signals; for phase change materials, the transmission or reflection characteristics of the waveguide are adjusted by changing the phase state of the phase change material, thereby achieving effective control of the optical signal.

[0009] In one embodiment, the optical information input layer includes: a light source, a spatial light modulator, and a first beam converter; The light source is used to generate the original light spot; The spatial light modulator is used to load input vector information; The first beam converter is used to shrink the original beam spot loaded with input vector information and then couple it into the reconfigurable hidden layer.

[0010] In one embodiment, the detection output layer includes: a second beam converter and a charge-coupled element; The beam converter is used to amplify the output optical signal; The charge-coupled device is used to detect the power distribution of the output optical signal.

[0011] In one embodiment, the feedback control structure includes: a field-programmable gate array (FPGA) and a digital-to-analog / analog-to-digital converter (DAC); The field-programmable gate array (FPGA) is used to execute the control algorithm; The digital-to-analog / analog-to-digital converter is connected to the field-programmable gate array (FPGA) and is used to convert the digital level generated by the FPGA into an analog level and control the optical field signal and input information in the waveguide of the reconfigurable hidden layer; and to convert the analog level of the detector array into a digital level and then send it to the FPGA for processing.

[0012] This invention provides a large-scale reconfigurable three-dimensional integrated optical neural network chip, comprising: an optical information input layer, a reconfigurable hidden layer, a detector output layer, and a feedback control structure; The reconfigurable hidden layer comprises a multi-layered cascaded waveguide array. Each cascaded waveguide array includes an independent modulation region and a continuous coupling region. The independent modulation region comprises N×M independent transmission waveguides and N×M optical modulators. The N×M independent transmission waveguides are used to transmit the input optical signal. No two waveguides are coupled to each other. Each optical modulator independently modulates the transmitted light in each independent transmission waveguide, corresponding to a reconfigurable (N×M)×(N×M) diagonal matrix. The continuous coupling region comprises a waveguide array composed of N×M waveguides, corresponding to a fixed (N×M)×(N×M) unitary matrix. By cascading the above-mentioned independent modulation region and continuous coupling region to form a single-layer neural network architecture, the input optical signal is alternately modulated to obtain the output optical signal to be tested. The optical information input layer is used to generate an input optical signal carrying input image information and incident orthogonally on the reconfigurable hidden layer. The detection output layer is used to detect the power of the output optical signal to be tested; The feedback control structure, connected to the detection output layer and the reconfigurable hidden layer, is used to generate a feedback control signal based on the power distribution information of the output optical signal under test, and to adjust the loading signal of the optical modulator using the feedback control signal, thereby modulating the optical field signal in the multi-layer cascaded waveguide array, so that the reconfigurable hidden layer can reconfigure the input optical signal to obtain the output optical signal under test.

[0013] In one embodiment, the three-dimensional integrated optical neural network is realized by inscribing optical waveguides in a glass substrate material through a femtosecond laser modification process.

[0014] In one embodiment, the optical information input layer includes: an illumination source, a spatial light modulator, and a first beam converter; The lighting source is used to generate the original light spot; The spatial light modulator is used to load input vector information; The first beam converter is used to reduce the size of the original beam spot carrying the information and then couple it into the reconfigurable hidden layer; During operation, the illumination source transmits light through a spatial light modulator to load input vector information, and then passes through a first beam converter to reduce the light spot of the loaded information, thereby realizing a reconfigurable hidden layer composed of a beam-coupled input waveguide array containing input vector information.

[0015] In one embodiment, the probe output layer includes a second beam converter and a charge-coupled element. After the light is output from the reconfigurable hidden layer, it passes through the second beam converter to expand the original output spot to fully cover the charge-coupled element.

[0016] In one embodiment, the independent modulation region includes N×M root waveguides with independent transmission and modulation, corresponding to N×M neurons. N×M optical modulators modulate the complex amplitude of the optical signal in the N×M root waveguides one-to-one, corresponding to a reconfigurable (N×M)×(N×M) diagonal matrix.

[0017] In one embodiment, the continuous coupling region includes N×M root waveguides that are coupled together for overall modulation, corresponding to N×M neurons. The continuously coupled waveguide array reaches a certain transmission length to realize the coupling between the optical fields of any input waveguide and any output waveguide, corresponding to a fixed (N×M)×(N×M) unitary matrix.

[0018] In one embodiment, N×M optical modulators in independent modulation zones are used to control the optical field signals in N×M corresponding waveguides, corresponding to a reconfigurable (N×M)×(N×M) diagonal matrix; based on the waveguide array arrangement and spacing of the continuous coupling zone, the coupling relationship between any N×M inputs and N×M outputs of the waveguide array can be calculated, corresponding to a fixed (N×M)×(N×M) unitary matrix; The signal loaded onto the optical modulator in the independent modulation zone is controlled by a feedback control structure to achieve adjustment of the reconfigurable (N×M)×(N×M) diagonal matrix.

[0019] In one embodiment, the feedback control structure includes: Field-programmable gate arrays (FPGAs) are used to execute control algorithms; A digital-to-analog (DAC) / analog-to-digital (ADC) converter, connected to the FPGA, is used to convert digital levels generated by the FPGA into analog levels and control signals on the optical modulator, or to convert analog levels of the probe output layer into digital levels and then send them to the FPGA for processing.

[0020] In one embodiment, the optical modulator is a thermoelectrode, a thermo-optic modulator, an electro-optic modulator, or a phase change material. For a thermoelectrode, a thermo-optic modulator, or an electro-optic modulator, the optical signal passing through the waveguide is modulated by applying other signals. For a phase change material, the transmission or reflection characteristics of the waveguide are adjusted by changing the phase state of the phase change material, thereby achieving effective control of the optical signal.

[0021] In one embodiment, the N×M waveguide arrays are arranged in a square grid array, or in a circular or polygonal grid array.

[0022] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects: (1) The reconfigurable three-dimensional integrated optical neural network chip based on tunable waveguide array provided by the present invention includes a multi-level waveguide array with reconfigurable hidden layers, corresponding to multiple fully connected layers of the neural network; each level of the waveguide array includes an independent modulation region and a continuous coupling region. In the independent modulation region, the optical signals in N×M independent transmission waveguides are modulated by N×M optical modulators, corresponding to a reconfigurable (N×M)×(N×M) diagonal matrix; in the continuous coupling region, the coupling relationship between the N×M inputs and N×M outputs of the waveguide array corresponds to a fixed (N×M)×(N×M) unitary matrix (which can also be modulated by optical modulators to achieve reconfigurability). The feedback control structure generates a feedback control signal according to the power distribution of the output optical signal, and uses the feedback control signal to adjust the loading signal of the optical modulator, thereby controlling the optical modulator of the reconfigurable hidden layer, thereby adjusting the optical field signal in the waveguide to train the reconfigurable hidden layer so as to be able to identify the current input signal. Therefore, this invention realizes reconfigurable matrix multiplication calculation of input information through a three-dimensional integrated tunable waveguide array, thereby realizing a large-scale reconfigurable neural network, which greatly improves the scalability and flexibility of integrated optical neural networks. At the same time, it also proposes a solution to the problems of low reconfigurability and large system size of current three-dimensional spatial diffraction optical neural networks.

[0023] (2) The large-scale reconfigurable three-dimensional integrated optical neural network chip provided by the present invention realizes the improvement of actual computing power from on-chip integrated quadratic level to three-dimensional integrated cubic level, laying the foundation for realizing high-computing-power integrated optical neural networks.

[0024] (3) The large-scale reconfigurable three-dimensional integrated optical neural network chip provided by the present invention is based on the architecture of tunable waveguide array, which can realize large-scale on-chip integration. The large-scale reconfigurable three-dimensional integrated optical neural network chip provided by the present invention can be realized by writing optical waveguides in glass substrate material using femtosecond laser modification process. Attached Figure Description

[0025] Figure 1 This is a large-scale reconfigurable three-dimensional integrated optical neural network chip provided in the embodiments of the present invention, with a three-layer 3D structure. A structural diagram for example, N=3, M=3.

[0026] Figure 2 The large-scale reconfigurable three-dimensional integrated optical neural network chip provided in this invention example is based on a three-layer 3D... A side view of a structure with 3 (N=3, M=3) as an example.

[0027] Figure 3 The large-scale reconfigurable three-dimensional integrated optical neural network chip provided in this invention example is based on a three-layer 3D... A top view of a structure with 3 (N=3, M=3) as an example.

[0028] Figure 4 This is an experimental setup diagram of the large-scale reconfigurable three-dimensional integrated optical neural network chip provided in this invention example. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0030] This invention discloses a large-scale reconfigurable three-dimensional integrated optical neural network chip, comprising: Optical information input layer I is used to generate an input optical signal carrying input information that is incident on reconfigurable hidden layer II; The reconfigurable hidden layer II is used to modulate the intensity distribution of the input optical signal using the waveguide array within it. It includes an M-level waveguide array, corresponding to the M fully connected layers in an optical neural network; each waveguide array includes an independent modulation region 4 and a continuous coupling region 5. The independent modulation region 4 modulates the optical signal within N×M independent transmission waveguides using N×M optical modulators, corresponding to a reconfigurable (N×M)×(N×M) diagonal matrix; the continuous coupling region 5 represents the coupling relationship between the N×M inputs and N×M outputs of the waveguide array, corresponding to a fixed (N×M)×(N×M) unitary matrix. Output layer III is used to detect the output optical signal.

[0031] Feedback control structure IV, connected to the detector output layer III and the reconfigurable hidden layer II, is used to generate a feedback control signal based on the power distribution information of the output optical signal under test, and to adjust the loading signal of the optical modulator using the feedback control signal, thereby modulating the optical field signal in the multi-layer cascaded waveguide array, thereby training the reconfigurable hidden layer II, which can identify the spatial distribution information of the light intensity of the current input signal.

[0032] Figure 1 The structure shown only depicts a three-layer cascaded 3×3 (N=3, M=3) waveguide array. The number of cascaded layers and waveguides can be expanded according to this scheme.

[0033] Figure 2 and Figure 3 They are respectively Figure 1 Side and top views of the structural schematic diagram.

[0034] In one embodiment, the optical modulator can be a thermoelectrode, any other optical modulator (thermally tunable, electrically tunable), or a phase change material. For thermoelectrodes and other arbitrary optical modulators, the optical signal passing through the waveguide is modulated by applying other signals; for phase change materials, the transmission or reflection characteristics of the waveguide are adjusted by changing the phase state of the phase change material, thereby achieving effective control of the optical signal.

[0035] Specifically, assuming the number of waveguide arrays in the reconfigurable hidden layer II is N×M, the spatial light modulator 2 can divide the spatial distribution of the input light into N×M pixels, with each pixel represented by the complex amplitude of the light at that location. Therefore, the input can be written as an (N×M)×1 vector E. in The transmission of light within the independent modulation region 4 can be represented by a reconfigurable (N×M)×(N×M) diagonal matrix M. i Using (i=1,2…) to describe it, the transmission of light within the continuous coupling region 5 can be regarded as a fixed (N×M)×(N×M) unitary matrix U. i (i=1,2…), the function of optocoupler 7 is to take the intensity I of the output optical signal. out Therefore, the information received by the m-layer three-dimensional integrated optical neural network detector can be represented as: I out =|M m U m M m-1 …M2U2M1U1E in | 2The entire system can be viewed as first performing matrix multiplication on the input vector, and then generating a squared nonlinearity, which is also a component of a neural network. Therefore, the system can be regarded as an optical neural network.

[0036] In one embodiment, the optical information input layer I includes: The spatial light modulator 2 loads the initial input signal by encoding the phase of the input light source and transfers it to the spatial distribution of the input light; The beam converter 3 is positioned after the spatial light modulator 2 and is used to adjust the spot size of the beam after the input signal is applied so that it matches the size of the tunable waveguide array of the reconfigurable hidden layer II, thereby achieving high-efficiency coupling input.

[0037] In one embodiment, the probe output layer III includes: Beam converter 6 enlarges the spot size of the beam of the output signal of reconfigurable hidden layer II, which facilitates the next step of detection; The charge-coupled device 7 is used to detect the power distribution of the output optical signal.

[0038] In one embodiment, a feedback control structure IV is connected to the probe output layer III and the reconfigurable hidden layer II, and is used to generate a feedback control signal based on the power distribution information of the output optical signal of the waveguide under test, and to adjust the loading signal of the optical modulator using the feedback control signal. The feedback control structure IV includes: Field-programmable gate arrays (FPGAs) are used to execute control algorithms; A digital-to-analog / analog-to-digital converter, connected to the FPGA, is used to convert the digital level generated by the FPGA into an analog level and control the optical field signal and input information in the waveguide of the reconfigurable hidden layer II; or to convert the analog level of the detector array into a digital level and send it to the FPGA for processing.

[0039] Figure 4 The experimental setup diagram shown includes all the components mentioned in the above specific embodiments.

[0040] The entire three-dimensional integrated optical neural network provided by this invention can operate in two modes: pre-training mode and online training mode. Pre-training mode requires obtaining the waveguide array coupling matrix M and the modulation effect of the optical modulator on the light from simulation or experiment beforehand, constructing a simulation model, and optimizing the signal of each optical modulator using a gradient descent algorithm in a computer to achieve the target output. Then, the optimized signal is configured onto the optical modulator, thus completing the construction of the optical neural network. Online training mode refers to directly using the transmission of the actual physical system instead of the mathematical model transmission of the aforementioned methods, providing stronger robustness to the defects of the physical system.

[0041] In summary, the key technical point of this application lies in the design of the reconfigurable hidden layer, which enables arbitrary allocation of the propagation path of the input optical signal of the chip. The high density, high flexibility and scalability of the three-dimensional integrated optical waveguide determine its potentially strong optical field manipulation capability. By establishing a rich mapping relationship between the target functional characteristics of the neural network and the reconfigurable modulation physical characteristics, the reconfigurable function of the optical neural network chip can be completed, which helps to accelerate the realization of a high-computing-power, high-integration, and multifunctional optical neural network system.

[0042] The waveguide array and one type of optical modulator—the thermoelectrode—of the present invention can be fabricated using femtosecond laser processing technology.

[0043] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A large-scale reconfigurable three-dimensional integrated optical neural network chip, characterized in that, include: Optical information input layer (I), reconfigurable hidden layer (II), detector output layer (III), and feedback control structure (IV); The reconfigurable hidden layer (II) comprises a multi-layer cascaded waveguide array; each cascaded waveguide array includes an independent modulation region and a continuous coupling region. The independent modulation region includes N×M independent transmission waveguides and N×M optical modulators. The N×M independent transmission waveguides are used to transmit the input optical signal. No two waveguides are coupled to each other. The optical modulators, corresponding one-to-one, independently modulate the transmitted light in each independent transmission waveguide, corresponding to a reconfigurable (N×M)×(N×M) diagonal matrix. The continuous coupling region comprises a waveguide array composed of N×M waveguides, corresponding to a fixed (N×M)×(N×M) unitary matrix. Through the single-layer neural network architecture composed of the cascaded independent modulation region and continuous coupling region, the input optical signal is alternately modulated to obtain the output optical signal to be tested. The optical information input layer (I) is used to generate an input optical signal carrying input image information and incident on the reconfigurable hidden layer (II); The detection output layer (III) is used to detect the power of the output optical signal to be tested; The feedback control structure (IV), connected to the probe output layer (III) and the reconfigurable hidden layer (II), is used to generate a feedback control signal based on the power distribution information of the output optical signal under test, and to adjust the loading signal of the optical modulator using the feedback control signal, thereby modulating the optical field signal in the multi-layer cascaded waveguide array, so that the reconfigurable hidden layer (II) can reconfigure the input optical signal to obtain the output optical signal under test.

2. The large-scale reconfigurable three-dimensional integrated optical neural network chip as described in claim 1, characterized in that, The three-dimensional integrated optical neural network is realized by inscribing optical waveguides in a glass substrate material through a femtosecond laser modification process.

3. The large-scale reconfigurable three-dimensional integrated optical neural network chip as described in claim 1, characterized in that, The optical information input layer (I) includes: an illumination source (1), a spatial light modulator (2), and a first beam converter (3); The lighting source (1) is used to generate the original light spot; The spatial light modulator (2) is used to load input vector information; The first beam converter (3) is used to shrink the original beam spot of the loaded information and then couple it into the reconfigurable hidden layer (II); During operation, the illumination source (1) transmits input vector information through the spatial light modulator (2), and then passes through the first beam converter (3) to reduce the light spot of the loaded information, thereby realizing a reconfigurable hidden layer (II) composed of a beam coupled to the input waveguide array containing the input vector information.

4. The large-scale reconfigurable three-dimensional integrated optical neural network chip as described in any one of claims 1 to 3, characterized in that, The detection output layer (III) includes a second beam converter (6) and a charge-coupled element (7). After the light is output from the reconfigurable hidden layer (II), it passes through the second beam converter (6) to expand the original output spot to fully cover the charge-coupled element (7).

5. The large-scale reconfigurable three-dimensional integrated optical neural network chip as described in any one of claims 1 to 3, characterized in that, The independent modulation region (4) includes N×M root waveguides with independent transmission and modulation, corresponding to N×M neurons. N×M optical modulators modulate the complex amplitude of the optical signal in the N×M root waveguides one-to-one, corresponding to a reconfigurable (N×M)×(N×M) diagonal matrix.

6. The large-scale reconfigurable three-dimensional integrated optical neural network chip as described in any one of claims 1 to 3, characterized in that, The continuous coupling region (5) includes N×M root waveguides that are coupled together for overall modulation, corresponding to N×M neurons. The continuously coupled waveguide array reaches a certain transmission length to realize the coupling between the optical fields of any input waveguide and any output waveguide, corresponding to a fixed (N×M)×(N×M) unitary matrix.

7. The large-scale reconfigurable three-dimensional integrated optical neural network chip as described in claim 1, characterized in that, include: The N×M optical modulators in the independent modulation region (4) are used to control the optical field signals in the N×M corresponding waveguides, which correspond to a reconfigurable (N×M)×(N×M) diagonal matrix. Based on the waveguide array arrangement and spacing of the continuous coupling region (5), the coupling relationship between any N×M inputs and N×M outputs of the waveguide array can be calculated, corresponding to a fixed (N×M)×(N×M) unitary matrix. The signal loaded on the optical modulator of the independent modulation region (4) is controlled by the feedback control structure (IV) to achieve the adjustment of the reconfigurable (N×M)×(N×M) diagonal matrix.

8. The large-scale reconfigurable three-dimensional integrated optical neural network chip as described in claim 7, characterized in that, The feedback control structure (IV) includes: Field-programmable gate arrays (FPGAs) are used to execute control algorithms; A digital-to-analog / analog-to-digital converter, connected to the FPGA, is used to convert the digital level generated by the FPGA into an analog level and control the signal on the optical modulator, or to convert the analog level of the probe output layer (III) into a digital level and then send it to the FPGA for processing.

9. The large-scale reconfigurable three-dimensional integrated optical neural network chip as described in claim 1, characterized in that, Optical modulators can be thermoelectrodes, thermo-optic modulators, electro-optic modulators, or phase change materials. For thermoelectrodes, thermo-optic modulators, and electro-optic modulators, other signals are applied to modulate the optical signal passing through the waveguide. For phase change materials, the transmission or reflection characteristics of the waveguide are adjusted by changing the phase state of the phase change material, thereby achieving effective control of the optical signal.

10. The large-scale reconfigurable three-dimensional integrated optical neural network chip as described in claim 1, characterized in that, The N×M waveguide arrays are arranged in a uniform square grid array, or in a circular or polygonal grid array distribution.

Images