Optical neural network device and optical apparatus
By integrating the diffraction structure with the waveguide structure, the problem of unstable calculation results in optical neural network devices under small-to-medium scale computing needs is solved, achieving more stable optical path transmission and accurate optical calculation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Utility models(China)
- Current Assignee / Owner
- APPOTRONICS CORP LTD
- Filing Date
- 2025-05-27
- Publication Date
- 2026-07-24
AI Technical Summary
For small- to medium-scale computing needs, diffractive optical neural network devices have unstable calculation results due to different diffraction regions being set on different diffraction plates, making them susceptible to external vibrations and changes in temperature and humidity.
Multiple diffraction structures are integrated with waveguide structures to form a stable optical transmission path. By adjusting parameters such as phase, intensity, direction, and polarization of the optical signal through reconfigurable diffraction structures, the stability and accuracy of optical calculations can be achieved.
This improves the stability of computational results from optical neural network devices, avoids the effects of external vibrations and temperature and humidity changes, and enhances the stability of optical path transmission and the accuracy of computation.
Smart Images
Figure CN224553801U_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of optical chip technology, and more specifically, to an optical neural network device and optical apparatus. Background Technology
[0002] Optical Neural Networks (ONNs) are a type of neural network that uses optical components for computation. They utilize light to transmit and process data, replacing traditional electronic signals, and use light to execute neural network operations. Research shows that all-optical neural networks, using light for computation, have advantages over electronic neural networks in that they are faster and more energy-efficient. They also feature parallel processing, low heat generation, and high scalability, making all-optical neural networks a potential choice for efficient computation in artificial intelligence.
[0003] While all-optical neural networks theoretically offer advantages in high speed and low power consumption, they still face several challenges. For diffractive optical neural network devices with small to medium-scale computing needs, different diffraction regions are fabricated on different diffraction plates, and the diffractive neural network is implemented using multiple diffraction plates. In this approach, multiple diffraction plates are highly susceptible to disturbances from external vibrations and changes in temperature and humidity, leading to unstable computational results. Utility Model Content
[0004] This application proposes an optical neural network device and optical apparatus to improve the above-mentioned defects.
[0005] In a first aspect, this application provides an optical neural network device, comprising: a waveguide structure for propagating optical signals; at least one first diffraction structure; at least one second diffraction structure; both the first and second diffraction structures are integrally formed with the waveguide structure, and both the first and second diffraction structures are used to adjust at least one of the phase, intensity, direction, polarization, and mode of the optical signal; the optical signal enters the waveguide structure through the first diffraction structure, the optical signal processed by the first diffraction structure is transmitted to the second diffraction structure, and the optical signal processed by the second diffraction structure is emitted from the waveguide structure.
[0006] Optionally, in one possible implementation, it further includes: at least one third diffraction structure; the third diffraction structure is integrally formed with the waveguide structure, and the third diffraction structure is used to adjust at least one of the phase, intensity, direction, polarization, and mode of the optical signal; the optical signal enters the waveguide structure through the first diffraction structure, the optical signal processed by the first diffraction structure is transmitted to the third diffraction structure, the optical signal processed by the third diffraction structure is transmitted to the second diffraction structure, and the optical signal processed by the second diffraction structure exits from the waveguide structure.
[0007] Optionally, in one possible implementation, the first diffraction structure is located inside or on the surface of the waveguide structure, the second diffraction structure is located inside or on the surface of the waveguide structure, and the third diffraction structure is located inside or on the surface of the waveguide structure.
[0008] Optionally, in one possible implementation, at least one of the first diffraction structure, the second diffraction structure, and the third diffraction structure is a reconfigurable diffraction structure, and the diffraction characteristics of the reconfigurable diffraction structure change when irradiated by a controlled beam.
[0009] Optionally, in one possible implementation, the wavelength of the control beam corresponding to the reconfigurable diffraction structure is different from the wavelength of the target optical signal, wherein the target optical signal is the optical signal irradiating the reconfigurable diffraction structure.
[0010] Optionally, in one possible implementation, an adjustment module is also included, which is used to generate the control beam.
[0011] Optionally, in one possible implementation, the difference between the wavelength of the control beam corresponding to the reconfigurable diffraction structure and the wavelength of the target optical signal exceeds 2 nm.
[0012] Optionally, in one possible implementation, the first diffraction structure is a polarizing holographic liquid crystal grating or a target metal nanostructure, wherein the target metal nanostructure is capable of exciting surface plasmon resonance.
[0013] Optionally, in one possible implementation, the first diffraction structure is a holographic structure, an embossed structure, or a liquid crystal structure; the second diffraction structure is a holographic structure, an embossed structure, or a liquid crystal structure; and the third diffraction structure is a holographic structure, an embossed structure, or a liquid crystal structure.
[0014] Secondly, this application provides an optical device, including: the aforementioned optical neural network device, and a control module, wherein the control module is used to process the optical signal emitted by the waveguide structure to obtain target data.
[0015] This application provides an optical neural network device, comprising: a waveguide structure for propagating optical signals; at least one first diffraction structure; at least one second diffraction structure; both the first and second diffraction structures are integrally formed with the waveguide structure, and both the first and second diffraction structures are used to adjust at least one of the phase, intensity, direction, polarization, and mode of the optical signal; the optical signal enters the waveguide structure through the first diffraction structure, the optical signal processed by the first diffraction structure is transmitted to the second diffraction structure, and the optical signal processed by the second diffraction structure exits from the waveguide structure.
[0016] Compared to placing different diffraction regions on different diffraction plates, this application sets multiple diffraction structures on a waveguide structure and integrally forms them with the waveguide structure, making the optical path transmission of the optical neural network device more stable, that is, making the optical calculation results more accurate, and avoiding the influence of external vibration, temperature and humidity changes, etc.
[0017] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 A schematic diagram of the structure of the optical neural network device provided in an embodiment of this application is shown;
[0020] Figure 2 A schematic diagram of the structure of an optical neural network device provided in another embodiment of this application is shown;
[0021] Figure 3 A schematic diagram of the structure of the first diffraction structure provided in an embodiment of this application is shown;
[0022] Figure 4 A schematic diagram of the structure of an optical neural network device according to another embodiment of this application is shown;
[0023] Figure 5 A schematic diagram of the structure of an optical neural network device according to another embodiment of this application is shown;
[0024] Figure 6 A schematic diagram of the structure of an optical neural network device according to another embodiment of this application is shown;
[0025] Figure 7 A schematic diagram of the structure of an optical neural network device according to another embodiment of this application is shown;
[0026] Figure 8 A schematic diagram of the combined structure of an optical neural network device provided in another embodiment of this application is shown.
[0027] Explanation of reference numerals in the attached figures:
[0028] 11. Waveguide structure; 12. First diffraction structure; 13. Second diffraction structure; 14. Third diffraction structure. Detailed Implementation
[0029] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all of them. The components of the embodiments of the present application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without inventive effort are within the scope of protection of the present application.
[0030] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0031] Optical Neural Networks (ONNs) are a type of neural network that uses optical components for computation. They utilize light to transmit and process data, replacing traditional electronic signals, and use light to execute neural network operations. Research shows that all-optical neural networks, using light for computation, have advantages over electronic neural networks in that they are faster and more energy-efficient. They also feature parallel processing, low heat generation, and high scalability, making all-optical neural networks a potential choice for efficient computation in artificial intelligence.
[0032] While all-optical neural networks theoretically offer advantages in high speed and low power consumption, they still face several challenges. For diffractive optical neural network devices with small to medium-scale computing needs, different diffraction regions are fabricated on different diffraction plates, and the diffractive neural network is implemented using multiple diffraction plates. In this approach, multiple diffraction plates are highly susceptible to disturbances from external vibrations and changes in temperature and humidity, leading to unstable computational results.
[0033] Therefore, in this application embodiment, an optical neural network device and optical apparatus are provided to solve or partially solve the above-mentioned problems.
[0034] Please see Figure 1 It shows a schematic diagram of the structure of an optical neural network device 100 provided in an embodiment of this application, including:
[0035] Waveguide structure 11, which is used to propagate optical signals.
[0036] At least one first diffraction structure 12.
[0037] At least one second diffraction structure 13.
[0038] The first diffraction structure 12 and the second diffraction structure 13 are both integrally formed with the waveguide structure 11. The first diffraction structure 12 and the second diffraction structure 13 are both used to adjust at least one of the phase, intensity, direction, polarization and mode of the optical signal.
[0039] The optical signal enters the waveguide structure 11 through the first diffraction structure 12, and the optical signal processed by the first diffraction structure 12 is transmitted to the second diffraction structure 13. The optical signal processed by the second diffraction structure 13 is emitted from the waveguide structure 11.
[0040] It should be noted that the waveguide structure consists of a high refractive index core and a low refractive index cladding. The waveguide structure can confine light to a tiny channel (usually at the nanometer or micrometer scale) through the refractive index difference between the high refractive index core and the low refractive index cladding.
[0041] Diffractive structures are the core computational units of Diffractive Optical Neural Networks (DONNs). Diffractive structures are composed of micron or nanometer-scale optical modulation units and can perform mathematical operations (such as matrix multiplication and convolution) by utilizing the physical effects of light diffraction.
[0042] The first diffraction structure is used to diffract the optical signal entering the waveguide structure, and the second diffraction structure is used to diffract the optical signal transmitted within the waveguide structure and transmit it outside the waveguide structure. Both the first and second diffraction structures are used to adjust at least one of the phase, intensity, direction, polarization, and mode of the optical signal, thereby changing the information carried by the optical signal. This adjustment process is the optical calculation process.
[0043] like Figure 1 As shown, the optical neural network device includes three first diffraction devices 12 and one second diffraction device 13. Optical signals A1, A2, and A3 enter the waveguide structure 11 through their respective first diffraction devices 12, and are finally processed by the second diffraction device 13 to obtain optical signal B1 which exits the waveguide structure 11.
[0044] Compared to placing different diffraction regions on different diffraction plates, this application sets multiple diffraction structures on a waveguide structure and integrally forms them with the waveguide structure, making the optical path transmission of the optical neural network device more stable, that is, making the optical calculation results more accurate, and avoiding the influence of external vibration, temperature and humidity changes, etc.
[0045] In one alternative embodiment, please refer to Figure 2 The optical neural network device 100 also includes:
[0046] At least one third diffraction structure 14.
[0047] The third diffraction structure 14 is integrally formed with the waveguide structure 11, and the third diffraction structure 14 is used to adjust at least one of the phase, intensity, direction, polarization and mode of the optical signal.
[0048] The optical signal enters the waveguide structure 11 through the first diffraction structure 12, and the optical signal processed by the first diffraction structure 12 is transmitted to the third diffraction structure 14. The optical signal processed by the third diffraction structure 14 is transmitted to the second diffraction structure 13, and the optical signal processed by the second diffraction structure 13 is emitted from the waveguide structure 11.
[0049] By setting a third diffraction structure between the optical paths of the first and second diffraction structures, more complex optical neural network devices can be obtained for computing neural networks with more layers.
[0050] For an example, please refer to Figure 2The dashed arrows indicate the transmission path of the optical signal within the waveguide structure. Optical signal A1 enters the waveguide structure 11 through the first diffraction structure 12. After being processed by the first diffraction structure 12, the optical signal is transmitted to multiple third diffraction structures 14 connected in series (the multiple third diffraction structures are arranged sequentially in the transmission optical path). After being processed by the third diffraction structure 14, the optical signal is transmitted to the second diffraction structure 13. After being processed by the second diffraction structure 13, the optical signal B1 is emitted from the waveguide structure 11.
[0051] In one example, an optical signal enters a waveguide structure through a first diffraction structure, and the optical signal processed by the first diffraction structure is transmitted to multiple parallel third diffraction structures (the multiple third diffraction structures are arranged in parallel on the transmission optical path), the optical signal processed by the third diffraction structure is transmitted to a second diffraction structure, and the optical signal processed by the second diffraction structure is emitted from the waveguide structure.
[0052] In one alternative embodiment, the optical neural network device includes a first diffraction structure and multiple second diffraction structures, which can realize single-input multiple-output of optical signals.
[0053] In one alternative embodiment, the optical neural network device includes a plurality of first diffraction structures and a second diffraction structure, which can realize multiple inputs and single outputs of optical signals.
[0054] In one alternative embodiment, the optical neural network device includes a first diffraction structure and a second diffraction structure, enabling single-input single-output of optical signals.
[0055] It should be noted that this application does not limit the number or position of the first, second, and third diffraction structures; the specific number or position depends on the actual needs. The first, second, or third diffraction structures can process a single optical signal or multiple optical signals simultaneously.
[0056] In one alternative embodiment, the first diffraction structure is located inside or on the surface of the waveguide structure, the second diffraction structure is located inside or on the surface of the waveguide structure, and the third diffraction structure is located inside or on the surface of the waveguide structure.
[0057] In one optional embodiment, the first diffraction structure includes, but is not limited to, a holographic structure, an embossed structure, or a liquid crystal structure; the second diffraction structure includes, but is not limited to, a holographic structure, an embossed structure, or a liquid crystal structure; and the third diffraction structure includes, but is not limited to, a holographic structure, an embossed structure, or a liquid crystal structure.
[0058] For example, please refer to Figure 3It shows a first diffraction structure 12 with a surface relief structure, whose diffraction characteristics can be adjusted by modulating parameters such as the linewidth, morphology, height and spacing between adjacent features of different micro-nano structures.
[0059] One exemplary method for manufacturing the waveguide structure is a planar semiconductor process or a nanoimprint process.
[0060] In one alternative embodiment, please refer to Figure 4 At least one of the first diffraction structure 12, the second diffraction structure 13 and the third diffraction structure 14 is a reconfigurable diffraction structure, and the diffraction characteristics of the reconfigurable diffraction structure change when irradiated by a controlled beam.
[0061] It should be noted that by employing nonlinear optical materials, photochromic materials, carrier modulation, and plasma resonance, the diffraction characteristics of the structure can be further modulated by modulated light.
[0062] For example, the reconfigurable diffraction structure is a surface relief structure, and the material of this micro / nano relief structure is an optical nonlinear material (e.g., GaAs, WS2, etc.). Under the illumination of a control beam, the absorption coefficient of the material changes with the intensity of the control beam. Therefore, by changing the surface intensity distribution of the control beam, the grating diffraction efficiency within the irradiated area can be controlled, thereby achieving modulation of the intensity distribution of the diffracted light.
[0063] Furthermore, the wavelength of the control beam corresponding to the reconfigurable diffraction structure is different from the wavelength of the target optical signal, and the target optical signal is the optical signal that illuminates the reconfigurable diffraction structure.
[0064] Furthermore, the wavelength difference between the control beam corresponding to the reconfigurable diffraction structure and the wavelength of the target optical signal exceeds 2 nm.
[0065] Furthermore, the optical neural network device also includes an adjustment module for generating the control beam.
[0066] For example, such as Figure 4As shown, the dashed arrows indicate the transmission path of the optical signal within the waveguide structure. The third diffraction structure is a reconfigurable diffraction structure. Control beams C1 and C2 illuminate the third diffraction structure 14 respectively to adjust its diffraction characteristics. The optical signal A1 enters the waveguide structure 11 through the first diffraction structure 12. The optical signal processed by the first diffraction structure 12 is transmitted to two cascaded third diffraction structures 14 (the two third diffraction structures are set sequentially in the transmission optical path). The optical signal processed by the third diffraction structure 14 is transmitted to the second diffraction structure 13. The optical signal B1 processed by the second diffraction structure 13 is emitted from the waveguide structure 11.
[0067] When illuminated by a controlled beam, the diffraction characteristics of the reconfigurable diffraction structure change. In other words, the controlled beam can modulate the diffraction characteristics of the reconfigurable diffraction structure, thereby altering the information carried by the output signal light and realizing the modulated function of the optical neural network device. This application achieves device reconfigurability by introducing modulating light to modulate the diffraction characteristics of the diffraction structure.
[0068] In one alternative embodiment, the first diffraction structure is a polarizing holographic liquid crystal grating or a target metal nanostructure, wherein the target metal nanostructure is capable of exciting surface plasmon resonance.
[0069] It should be noted that the first diffraction structure is a polarizing holographic liquid crystal grating or a target metal nanostructure. That is, the first diffraction structure is sensitive to wavelength and direction. For the first diffraction structure, changing the wavelength of the input light will change the convergence position of the output light, which can realize the multi-condition operation of the device and also realize the reuse of optical neural network devices.
[0070] Please see Figure 5 , Figure 5 Optical neural network devices and Figure 4 The optical neural network devices in them are exactly the same. Figure 5 The optical signal A4 input to the first diffraction structure 12 and Figure 4 The wavelength of the optical signal A1 input to the first diffraction structure 12 is different, and the wavelength of the optical signal B2 after processing by the second diffraction structure 13 is different. Figure 4 The convergence position of the optical signal B1 obtained in the process is different, and thus the calculation results of the optical neural network device are also different, thereby enabling the reuse of the optical neural network device.
[0071] It should be noted that the beam can also be deflected by total internal reflection at the interface between the waveguide structure and the air. The waveguide structure can be strip-shaped or irregularly shaped. This application does not impose specific limitations on the shape of the waveguide structure, which can be determined according to specific needs.
[0072] For an example, please refer to Figure 6The dashed arrows indicate the transmission path of the optical signal within the waveguide structure. These dashed arrows are for illustrative purposes only, and the actual optical path depends on the specific structure. Optical signal A1 enters the waveguide structure 11 through the first diffraction structure 12. After being processed by the first diffraction structure 12, the optical signal is further processed by the third diffraction structure 14 and subjected to total internal reflection by the waveguide structure 11. Finally, optical signal B1, after being processed by the second diffraction structure 13, exits from the waveguide structure 11.
[0073] For an example, please refer to Figure 7 The dashed arrows indicate the transmission path of the optical signal within the waveguide structure. These dashed arrows are for illustrative purposes only, and the actual optical path depends on the specific structure. Optical signal A1 enters the waveguide structure 11 through the first diffraction structure 12. After being processed by the first diffraction structure 12, the optical signal undergoes total internal reflection by the waveguide structure 11 and is then processed by two second diffraction structures 13 to obtain two optical signals B1 and B2, which exit the waveguide structure 11 respectively.
[0074] It should be noted that multiple optical neural network devices can be used in combination; please refer to [link / reference]. Figure 8 The optical signal B1 output by the second-layer optical neural network device is transmitted to the first diffraction structure 12 of the first-layer optical neural network device for processing, and the optical signal B2 obtained by the second diffraction structure 13 of the first-layer optical neural network device is emitted out of the waveguide structure.
[0075] Specifically, in the second-layer optical neural network device, control beams C1 and C2 respectively adjust the diffraction characteristics of the third diffraction structure 14. Optical signal A1 enters the waveguide structure 11 through the first diffraction structure 12. After being processed by the first diffraction structure 12, the optical signal is further processed by the third diffraction structure 14 and subjected to total internal reflection by the waveguide structure 11. Finally, optical signal B1, processed by the second diffraction structure 13, exits from the waveguide structure 11. Similarly, in the second-layer optical neural network device, control beams C3 and C4 respectively adjust the diffraction characteristics of the first diffraction structure 12 and the third diffraction structure 14. Optical signals A2 and B1 respectively enter the waveguide structure 11 through their respective first diffraction structures 12. After being processed by the first diffraction structure 12, the optical signals are further processed by the third diffraction structure 14 and subjected to total internal reflection by the waveguide structure 11. Finally, optical signal B2, processed by the second diffraction structure 13, exits from the waveguide structure 11.
[0076] Therefore, the optical neural network device of this application has strong scalability and high flexibility in use. Multiple optical neural network devices can be used in series or in parallel.
[0077] In one alternative embodiment, this application also proposes an optical device, including: the aforementioned optical neural network device, and a control module, wherein the control module is used to process the optical signal emitted by the waveguide structure to obtain target data.
[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. An optical neural network device, characterized in that, include: A waveguide structure for propagating optical signals; At least one first diffraction structure; At least one second diffraction structure; Both the first diffraction structure and the second diffraction structure are integrally formed with the waveguide structure, and both the first diffraction structure and the second diffraction structure are used to adjust at least one of the phase, intensity, direction, polarization and mode of the optical signal. The optical signal enters the waveguide structure through the first diffraction structure, and the optical signal processed by the first diffraction structure is transmitted to the second diffraction structure. The optical signal processed by the second diffraction structure is emitted from the waveguide structure.
2. The optical neural network device according to claim 1, characterized in that, Also includes: At least one third diffraction structure; The third diffraction structure is integrally formed with the waveguide structure, and the third diffraction structure is used to adjust at least one of the phase, intensity, direction, polarization and mode of the optical signal. The optical signal enters the waveguide structure through the first diffraction structure, is processed by the first diffraction structure, is transmitted to the third diffraction structure, is processed by the third diffraction structure, is transmitted to the second diffraction structure, and is emitted from the waveguide structure after being processed by the second diffraction structure.
3. The optical neural network device according to claim 2, characterized in that, The first diffraction structure is located inside or on the surface of the waveguide structure, the second diffraction structure is located inside or on the surface of the waveguide structure, and the third diffraction structure is located inside or on the surface of the waveguide structure.
4. The optical neural network device according to claim 2, characterized in that, At least one of the first diffraction structure, the second diffraction structure, and the third diffraction structure is a reconfigurable diffraction structure, and the diffraction characteristics of the reconfigurable diffraction structure change when irradiated by a controlled beam.
5. The optical neural network device according to claim 4, characterized in that, The wavelength of the control beam corresponding to the reconfigurable diffraction structure is different from the wavelength of the target optical signal, which is the optical signal that illuminates the reconfigurable diffraction structure.
6. The optical neural network device according to claim 4, characterized in that, It also includes an adjustment module for generating the control beam.
7. The optical neural network device according to claim 5, characterized in that, The wavelength difference between the control beam corresponding to the reconfigurable diffraction structure and the wavelength of the target optical signal exceeds 2 nm.
8. The optical neural network device according to claim 1, characterized in that, The first diffraction structure is a polarizing holographic liquid crystal grating or a target metal nanostructure, wherein the target metal nanostructure can excite surface plasmon resonance.
9. The optical neural network device according to claim 2, characterized in that, The first diffraction structure is a holographic structure, an embossed structure, or a liquid crystal structure; the second diffraction structure is a holographic structure, an embossed structure, or a liquid crystal structure; and the third diffraction structure is a holographic structure, an embossed structure, or a liquid crystal structure.
10. An optical device, characterized in that, include: The optical neural network device according to any one of claims 1-9, and the control module, wherein the control module is used to process the optical signal emitted by the waveguide structure to obtain target data.