Optical neural network chip having reconfigurable information path, design method, and device

Abstract

An optical neural network chip (10) having a reconfigurable information path, a design method, and a device. The chip comprises: a chip substrate (100) and an optical diffraction layer (200) arranged on the chip substrate (100). The optical diffraction layer (200) comprises a plurality of reconfigurable diffraction modulation units. The refractive index of the optical diffraction layer (200) can be reconfigured on the basis of optical characteristics of, the number of, the number of layers of, and the spatial distribution of the reconfigurable diffraction modulation units, and a cluster modulation effect of the reconfigurable diffraction modulation units on an optical signal can be used to achieve multi-degree-of-freedom modulation of one or more optical signal wavefront propagation characteristics. Thus, the problems in the related art of the large physical size of modulators, a low integration level, and limited regulation sensitivity and precision of reconfigured optical neural network chips (10) are solved.

Description

Optical neural network chip with reconfigurable information path, design method and device

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application is based on the Chinese patent application with application number 202310819769.5 and application date of July 5, 2023, and claims the priority of the Chinese patent application. The entire content of the Chinese patent application is hereby introduced into this application as a reference. Technical Field

[0003] The present application relates to the technical field of optical neural network chips, and in particular to an optical neural network chip with reconfigurable information paths, a design method, and a device. Background Art

[0004] Optical neural networks use the interconnection and interaction of photons within optical networks to transmit and process optical signals. They are the key core technology for realizing a new generation of low-power, high-speed, high-computing-power high-end chips such as optical computing chips and brain-like chips. Among them, optical neural network chips use the propagation characteristics of photons within the chip to realize on-chip processing of optical signals and various computing functions. They are an important technical path to achieve the miniaturization, integration and practical application of neural network chips.

[0005] In related technologies, traditional optical neural networks mainly modulate the optical properties of input signals in waveguide devices, such as microring modulators and Mach-Zehnder interferometers, to achieve reconstruction of the chip's logical functions. However, due to the inherent technical limitations of this method, such as large modulator size, low integration density, and limited modulation freedom, its modulation scale, sensitivity, and accuracy are extremely limited. As a result, the prepared neural network chips, while achieving many excellent characteristics such as high computing power, also have problems such as limited neural network scale and single logical function. One of the core reasons is that once the chip is completed, its structure or optical properties cannot be changed arbitrarily, which greatly limits the practical application value of such chips.

[0006] Summary of the Invention

[0007] The present application provides an optical neural network chip, design method and device with reconfigurable information path to solve the problems of reconstructed optical neural network chips in related technologies, such as large modulator physical size, low integration, and limited control sensitivity and accuracy.

[0008] The first aspect of the present application provides an optical neural network chip with a reconfigurable information path, comprising: a chip substrate and an optical diffraction layer arranged on the chip substrate, wherein the optical diffraction layer includes a plurality of reconfigurable diffraction modulation units, and the optical diffraction layer dynamically reconstructs the refractive index based on the optical properties, quantity, number of layers and spatial distribution of the reconfigurable diffraction modulation units, and utilizes the cluster modulation effect of the reconfigurable diffraction modulation units on the optical signal to achieve multi-degree-of-freedom modulation of the wavefront propagation characteristics of one or more optical signals.

[0009] Optionally, the multi-degree-of-freedom modulation includes modulation of arbitrary energy distribution and / or arbitrary propagation path.

[0010] Optionally, the structure of the optical diffraction layer includes one or more of depth, scale and spatial arrangement.

[0011] Optionally, the physical model of the optical diffraction layer includes one or more of an optical hole diffraction layer and an optical medium column diffraction layer.

[0012] Optionally, the optical hole diffraction layer includes rectangular gratings and / or circular holes, and the optical medium column diffraction layer includes rectangular columns and / or circular columns.

[0013] Optionally, the physical form of the reconfigurable diffraction modulation unit includes one or more of the period, width, height, length, shape, scale, material and refractive index of neurons.

[0014] Optionally, the material includes a phase change material and / or a ferroelectric material.

[0015] Optionally, the optical neural network chip includes one or more of a diffraction neural network chip, an interference neural network chip, and a scattering neural network chip.

[0016] A second embodiment of the present application provides a device, including an optical neural network chip with reconfigurable information path as described in the above embodiment.

[0017] A third aspect of the present application provides a method for designing an optical neural network chip with reconfigurable information paths. The method is used for designing an optical neural network chip with reconfigurable information paths as described in the above embodiments, and the method includes the following steps: obtaining design requirements for an optical diffraction layer; determining the inherent optical properties, quantity, number of layers, and spatial distribution of reconfigurable diffraction modulation units in the optical diffraction layer according to the design requirements; arranging a plurality of reconfigurable diffraction modulation units on a chip substrate, and dynamically reconstructing the refractive index of the optical diffraction layer based on the inherent optical properties, quantity, number of layers, and spatial distribution of the reconfigurable diffraction modulation units, wherein the cluster modulation effect of the reconfigurable diffraction modulation units on optical signals is utilized to realize multi-degree-of-freedom modulation of the wavefront propagation characteristics of one or more optical signals.

[0018] Therefore, this application has at least the following beneficial effects:

[0019] This application mainly adopts a diffraction layer composed of on-chip diffraction neurons with adjustable refractive index as a wavefront modulation layer, and specially designs the optical properties and physical form of the neurons to achieve multi-level modulation of the wavefront of the input signal beam of the optical neural network chip, and then reconstruct its multi-parameter optical properties such as energy and propagation path, thereby achieving the purpose of reconstructing the logical function of the optical neural network chip and enhancing the chip's processing ability for complex and dynamic optical input information.

[0020] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0022] FIG1 is a structural diagram of an optical neural network chip with reconfigurable information paths according to an embodiment of the present application;

[0023] FIG2 is a structural diagram of a reconfigurable optical neural network chip based on a reconfigurable diffraction layer according to an embodiment of the present application;

[0024] FIG3 is a flow chart of a method for designing an optical neural network chip with reconfigurable information paths according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0026] Optical neural networks use the interconnection and interaction of photons within optical networks to transmit and process optical signals. They are the key core technology for realizing a new generation of low-power, high-speed, high-computing-power high-end chips such as optical computing chips and brain-like chips. Optical neural network chips use the propagation characteristics of photons within the chip to realize on-chip processing of optical signals and various computing functions. They are an important technical path to achieve the miniaturization, integration and practical application of neural network chips.

[0027] Currently, due to the failure of Moore's Law and the inherent limitations of the traditional von Neumann architecture, as well as the explosive development of emerging applications such as artificial intelligence, cloud computing, and autonomous driving, the computing power and energy consumption of electronic chip networks have reached a difficult bottleneck. At the same time, the excellent properties of photons, such as high bandwidth, large capacity, low power consumption, and low crosstalk, make the exploration of photons as information computing carriers particularly important. In this context, research on optical neural network chips has risen to the strategic level of breaking through the bottlenecks of existing electronic chip computing architectures and seizing the country's position at the forefront of global information technology.

[0028] However, while existing research on neural network chips has achieved many excellent characteristics such as high computing power, it has also encountered many difficulties, such as the limited scale of the chip neural network and single logical function. One of the core reasons is that once the chip is prepared, its structure or optical properties cannot be changed arbitrarily, which greatly limits the practical application value of such chips.

[0029] Traditional optical neural networks primarily modulate the optical properties of input signals within waveguide devices, such as microring modulators and Mach-Zehnder interferometers. However, due to inherent technical limitations of this approach, such as large modulator size, low integration density, and limited modulation freedom, the modulation scale, sensitivity, and accuracy are severely limited. To address these challenges, developing high-efficiency, high-degree-of-freedom, on-chip programmable optical neural network chips with reconfigurable logic functions is urgently considered a necessary step and a key strategic trend towards achieving multifunctional, universal optical neural network chips.

[0030] The objective of the embodiments of the present application is to provide an optical neural network chip based on reconfigurable information paths, which mainly uses a diffraction layer composed of on-chip diffraction neurons with adjustable refractive index as a wavefront modulation layer to realize the path reconstruction function of the input optical signal.

[0031] The following describes an optical neural network chip with reconfigurable information paths, a design method, and an apparatus according to an embodiment of the present application with reference to the accompanying drawings. Specifically, FIG1 is a structural diagram of an optical neural network chip with reconfigurable information paths provided by an embodiment of the present application.

[0032] As shown in FIG1 , the optical neural network chip 10 with reconfigurable information paths includes: a chip substrate 100 and an optical diffraction layer 200 .

[0033] Among them, the optical diffraction layer 200 is arranged on the chip substrate 100, and the optical diffraction layer 200 includes a plurality of reconfigurable diffraction modulation units. The optical diffraction layer 200 dynamically reconstructs the refractive index based on the optical properties, quantity, number of layers and spatial distribution of the reconfigurable diffraction modulation units, and utilizes the cluster modulation effect of the reconfigurable diffraction modulation units on the optical signal to achieve multi-degree-of-freedom modulation of the wavefront propagation characteristics of one or more optical signals.

[0034] Among them, the optical neural network chip 10 includes one or more of a diffraction neural network chip, an interference neural network chip and a scattering neural network chip. It should be noted that the optical neural network chip is not limited to the above chips and is not specifically limited.

[0035] The physical form of the reconfigurable diffraction modulation unit includes one or more of the period, width, height, length, shape, scale, material and refractive index of the neuron.

[0036] Among them, the materials include phase change materials and / or ferroelectric materials, which can be selected according to actual needs. Phase change materials refer to substances that change the state of matter and can provide latent heat under constant temperature, such as water. Ferroelectric materials refer to a class of materials with ferroelectric effect, such as electro-optical crystals, piezoelectric materials, etc., without specific limitation.

[0037] The structure of the optical diffraction layer 200 includes one or more of depth, scale, and spatial arrangement.

[0038] As shown in FIG2( b ), multi-degree-of-freedom modulation includes modulation of arbitrary energy distribution and / or arbitrary propagation path.

[0039] It can be understood that the embodiments of the present application utilize the characteristics of high neuron density in the optical diffraction layer and independent modulation of the optical properties of each neuron to perform high-resolution dynamic modulation of the wavefront of the input signal light beam in the diffraction layer, thereby achieving large-angle and high-precision modulation of its propagation path and energy distribution; the modulated light signal is used as the dynamic input of the next-level optical neural network to achieve reconstruction of the logical functions of the optical neural network chip, thereby greatly enhancing the chip's processing capabilities for complex and dynamic optical input information.

[0040] Specifically, as shown in (a) and (b) of Figure 2, multiple reconfigurable diffraction modulation units perform multi-pole control on the input optical signal wavefront and propagation path, as shown by wavefront 1, wavefront 2, wavefront 3, and path 1, path 2, path 3, respectively, to achieve the control function of the input signal propagation path of the neural network chip; the optical properties of the reconfigurable diffraction modulation unit itself and its number, number of layers, spatial distribution, etc. jointly define the reconfigurable characteristics of the entire control layer.

[0041] It should be noted that, as shown in (c) of Figure 2, the optical properties and physical morphology of the diffraction neurons of the reconfigurable diffraction modulation unit, such as the neuron period Λ, width w, length L, scale N, material or refractive index n, directly define the control ability of the modulation unit. The high density, large scale and scalability of the diffraction layer neurons determine its potentially extremely strong light field control ability. By establishing a mapping relationship between rich neural network target functional characteristics and reconfigurable diffraction modulation physical characteristics, the reconfigurable function of the optical neural network chip can be completed.

[0042] In the embodiment of the present application, the physical model of the optical diffraction layer 200 includes one or more of an optical hole diffraction layer and an optical medium column diffraction layer.

[0043] The optical hole diffraction layer includes rectangular gratings and / or circular holes, and the optical medium column diffraction layer includes rectangular columns and / or circular columns. The physical structure can be selected or adjusted according to actual conditions and is not specifically limited.

[0044] It can be understood that the embodiment of the present application is based on the reconstruction control layer of the high integration and high sensitivity information path of the reconfigurable diffraction layer, which can be realized by a variety of different physical models, including but not limited to optical hole diffraction layers of various shapes (such as rectangular gratings, circular holes, etc.), optical medium column diffraction layers (such as rectangular columns, cylinders, etc.) or other types of two-dimensional and three-dimensional physical structures, etc., without specific limitation.

[0045] According to the optical neural network chip with reconfigurable information paths proposed in the embodiments of the present application, the neurons in the diffraction layer are small in size, high in density, and have rich neuronal morphologies. Each neuron group can be independently and accurately modulated. By utilizing the cluster modulation effect of neurons on optical signals, the diffraction layer can achieve multi-degree-of-freedom modulation of the propagation characteristics of the input optical signal, thereby achieving high integration and high sensitivity reconstruction of the neural network logic functions of the neural network chip, greatly enhancing its processing capabilities for complex and dynamic optical input signals.

[0046] The following will be combined with Figure 2 to explain in detail the optical neural network chip with reconfigurable information path, as follows:

[0047] The embodiments of the present application can utilize the characteristics of high neuron density in the optical diffraction layer and independent modulation of the optical properties of each neuron to perform high-resolution dynamic modulation of the wavefront of the input signal light beam in the diffraction layer, thereby achieving large-angle and high-precision modulation of its propagation path and energy distribution; the modulated light signal is used as the dynamic input of the next-level optical neural network to realize the reconstruction of the logical functions of the optical neural network chip, thereby greatly enhancing the chip's processing capabilities for complex and dynamic optical input information.

[0048] As shown in (a) and (b) of Figure 2, the reconfigurable diffraction modulation unit performs multi-pole control on the wavefront and propagation path of the input optical signal, as shown by wavefront 1, wavefront 2, wavefront 3, and path 1, path 2, and path 3. Among them, multiple reconfigurable diffraction modulation units respectively perform multi-level modulation on the corresponding input optical signal paths to realize the control function of the input signal propagation path of the neural network chip. The optical properties of the reconfigurable diffraction modulation unit itself and its number, number of layers, spatial distribution, etc. jointly define the reconfigurable characteristics of the entire control layer.

[0049] As shown in (c) of Figure 2, a specific implementation form of the reconfigurable diffraction modulation unit is embodied. The optical properties and physical form of its diffraction neurons, such as the neuron's period Λ, width w, length L, scale N, material or refractive index n, directly define the modulation unit's control capability.

[0050] In summary, the embodiment of the present application provides a reconfigurable design for the optical diffraction layer, that is, a targeted design of the physical parameters of the reconfigurable diffraction modulation unit for the target logical function of the neural network chip, which can realize the arbitrary allocation of the chip input optical signal propagation path. The high density, large scale and scalability of the diffraction layer neurons determine its potentially extremely strong light field control capability. By establishing a mapping relationship between rich neural network target functional characteristics and reconfigurable diffraction modulation physical characteristics, the reconfigurable function of the optical neural network chip can be completed, which helps to accelerate the realization of a universal, multi-functional, highly integrated, low-power, and high-efficiency optical neural network system for practical application scenarios.

[0051] It should be noted that different neuron morphologies (such as different materials, different structures, etc.) and different diffraction layer structures (such as different diffraction layer depths, scales, spatial arrangements, etc.) can all constitute other application examples based on this application. In addition, for other types of reconfigurable optical neural networks, the reconfigurable technology paths of their neurons may be different, but based on the reconfigurable diffraction layer, the input optical signal wavefront is modulated to achieve the reconstruction of the logical functions of the optical neural network chip, thereby promoting the realization of a universal and multifunctional optical neural network chip.

[0052] An embodiment of the present application also provides a device including an optical neural network chip with reconfigurable information path as in the above embodiment.

[0053] FIG3 is a flow chart of a method for designing an optical neural network chip with reconfigurable information paths provided in an embodiment of the present application.

[0054] As shown in FIG3 , the design method of the optical neural network chip with reconfigurable information path is used for designing the optical neural network chip with reconfigurable information path of the above embodiment, and includes the following steps:

[0055] In step S101 , the design requirements of the optical diffraction layer are obtained.

[0056] It can be understood that the embodiment of the present application obtains the design requirements of the optical diffraction layer so as to subsequently determine the relevant content of the reconfigurable diffraction modulation unit in the optical diffraction layer according to the design requirements.

[0057] It should be noted that the design requirement of the optical diffraction layer may be to adjust the refractive index of the optical diffraction layer, without any specific limitation.

[0058] In step S102 , the optical properties, quantity, number of layers, and spatial distribution of the reconfigurable diffraction modulation units in the optical diffraction layer are determined according to design requirements.

[0059] It can be understood that the embodiment of the present application determines the optical properties, quantity, number of layers and spatial distribution of the reconfigurable diffraction modulation unit in the optical diffraction layer according to design requirements, so as to facilitate the subsequent reconstruction of the refractive index of the optical diffraction layer based on the relevant content of the reconfigurable diffraction modulation unit.

[0060] In step S103, a plurality of reconfigurable diffraction modulation units are arranged on the chip substrate, and the refractive index of the optical diffraction layer is dynamically reconstructed based on the optical properties, quantity, number of layers and spatial distribution of the reconfigurable diffraction modulation units. The cluster modulation effect of the reconfigurable diffraction modulation units on the optical signal is utilized to realize multi-degree-of-freedom modulation of the wavefront propagation characteristics of one or more optical signals.

[0061] It can be understood that the embodiment of the present application sets up multiple reconfigurable diffraction modulation units on the chip substrate, and dynamically reconstructs the refractive index of the optical diffraction layer based on the optical characteristics, quantity, number of layers and spatial distribution of the reconfigurable diffraction modulation units themselves, and utilizes the cluster modulation effect of the reconfigurable diffraction modulation units on the optical signal to achieve multi-degree-of-freedom modulation of the wavefront propagation characteristics of one or more optical signals, and achieves large-angle range and high-precision modulation of their propagation path and energy distribution.

[0062] According to the design method of an optical neural network chip with reconfigurable information paths proposed in the embodiments of the present application, the refractive index of the diffraction layer composed of high-density neurons is reconstructed and the wavefront of the light beam is precisely modulated to achieve precise control of optical properties such as the light beam energy distribution and the propagation path; by specially designing the scale of the diffraction layer and the physical form and dynamic refractive index distribution characteristics of its neurons, high integration and high-resolution regulation of the light beam propagation characteristics by the diffraction layer can be achieved, thereby effectively improving the reconfigurable scale and reducing the reconfigurable implementation cost.

[0063] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0064] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this application, "N" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0065] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing a custom logical function or process step, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a different order than shown or discussed, including performing functions in a substantially simultaneous manner or in a reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0066] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array, a field programmable gate array, etc.

[0067] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0068] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. An optical neural network chip with reconfigurable information path, characterized in that: include: A chip substrate and an optical diffraction layer arranged on the chip substrate, wherein: The optical diffraction layer includes a plurality of reconfigurable diffraction modulation units, which dynamically reconstruct the refractive index based on the optical properties, quantity, number of layers and spatial distribution of the reconfigurable diffraction modulation units, and utilize the cluster modulation effect of the reconfigurable diffraction modulation units on the optical signal to achieve multi-degree-of-freedom modulation of the wavefront propagation characteristics of one or more optical signals.

2. The optical neural network chip with reconfigurable information path according to claim 1, characterized in that: The multi-degree-of-freedom modulation includes modulation of arbitrary energy distribution and / or arbitrary propagation path.

3. The optical neural network chip with reconfigurable information path according to claim 1, characterized in that: The structure of the optical diffraction layer includes one or more of depth, scale and spatial arrangement.

4. The optical neural network chip with reconfigurable information path according to claim 1 or 3, characterized in that: The physical model of the optical diffraction layer includes one or more of an optical hole diffraction layer and an optical medium column diffraction layer.

5. The optical neural network chip with reconfigurable information path according to claim 4, characterized in that: The optical hole diffraction layer includes rectangular gratings and / or circular holes, and the optical medium column diffraction layer includes rectangular columns and / or circular columns.

6. The optical neural network chip with reconfigurable information path according to claim 1, characterized in that: The physical form of the reconfigurable diffractive modulation unit includes one or more of the period, width, height, length, shape, scale, material and refractive index of the neuron.

7. The optical neural network chip with reconfigurable information path according to claim 6, characterized in that: The material includes a phase change material and / or a ferroelectric material.

8. The optical neural network chip with reconfigurable information path according to claim 1, characterized in that: The optical neural network chip includes one or more of a diffraction neural network chip, an interference neural network chip and a scattering neural network chip.

9. A device, characterized in that: An optical neural network chip with reconfigurable information path comprising the one described in any one of claims 1-8.

10. A design method for an optical neural network chip with reconfigurable information path, characterized in that: The method is used for designing an optical neural network chip with reconfigurable information path as claimed in any one of claims 1 to 8, wherein the method comprises the following steps: Obtain design requirements for the optical diffraction layer; Determining the optical properties, quantity, number of layers and spatial distribution of the reconfigurable diffraction modulation units in the optical diffraction layer according to the design requirements; A plurality of reconfigurable diffraction modulation units are arranged on the chip substrate, and based on the optical characteristics of the reconfigurable diffraction modulation units, The refractive index of the optical diffraction layer is dynamically reconstructed by adjusting the property, quantity, number of layers and spatial distribution of the reconfigurable diffraction modulation unit, wherein the cluster modulation effect of the reconfigurable diffraction modulation unit on the optical signal is utilized to realize multi-degree-of-freedom modulation of the wavefront propagation characteristics of one or more optical signals.

Images