A photonic matrix-vector multiplier based on tunable bragg grating array

Abstract

This invention proposes a photonic matrix-vector multiplier based on a tunable Bragg grating array, relating to the field of optical information technology. It includes multiple input ports, an optical weighting modulation array, multiple summing waveguides, and multiple signal modulation units. Each input port is optically coupled sequentially to an optical weighting modulation unit in the corresponding column of the optical weighting modulation array via an input column waveguide. The optical weighting modulation array includes multiple optical weighting modulation units, each comprising a tunable Bragg grating integrated on a silicon-based waveguide and a driving electrode electrically connected to the tunable Bragg grating. Each summing waveguide is coupled to an optical weighting modulation unit in the corresponding row of the optical weighting modulation array. Optical powers weighted by different input column waveguides and different rows of optical weighting modulation units are linearly accumulated in parallel within the same summing waveguide. The signal modulation units are coupled to their corresponding summing waveguides. This application helps improve overall computational efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of optical information technology, and in particular to a photonic matrix vector multiplier based on a tunable Bragg grating array. Background Technology

[0002] Photonic computing, with its advantages of high parallelism, high bandwidth, and low latency, holds immense potential in artificial intelligence, big data processing, and high-speed signal processing. Existing electronic computing systems face the bottleneck of Moore's Law and the challenge of high energy consumption. The photonic matrix-vector multiplier (MVM) is the core technology for realizing optical neural networks (ONNs) and general-purpose optical computing.

[0003] Chinese Patent CN112988113B discloses a photonic matrix-vector multiplier. This photonic matrix-vector multiplier includes a light source, an input beam splitter, an input modulator array, a first multi-level diffraction network, an intermediate modulator array, a second multi-level diffraction network, and a detector array arranged sequentially. The input modulator array contains N first modulators; the intermediate modulator array contains N second modulators; and the detector array contains N detectors. The first multi-level diffraction network includes M levels of first photon distribution units, each level including a first beam splitter and a first phase modulator array. The second multi-level diffraction network includes M levels of second photon distribution units, each level including a second beam splitter and a second phase modulator array. However, this scheme requires the optical signal to pass through multiple levels of diffraction networks and a large number of phase modulators sequentially during matrix-vector multiplication. The cascaded structure introduces additional propagation delay and insertion loss, limiting the actual effective bandwidth and throughput, resulting in insufficient overall computational efficiency. Therefore, it is essential to provide a photon matrix vector multiplier based on a tunable Bragg grating array to improve overall computational efficiency and accuracy. Summary of the Invention

[0004] In view of this, the present invention proposes a photon matrix vector multiplier based on a tunable Bragg grating array.

[0005] This invention provides a photonic matrix-vector multiplier based on a tunable Bragg grating array, comprising multiple input ports, an optical weighting modulation array, multiple summing waveguides, and multiple signal modulation units, wherein...

[0006] The input port is used to receive multiple optical signals corresponding to each component in the input vector. Each input port is optically coupled to the optical weighting modulation unit in the corresponding column of the optical weighting modulation array through an input column waveguide.

[0007] The optical weighting modulation array includes multiple optical weighting modulation units. Each optical weighting modulation unit includes a tunable Bragg grating integrated on a silicon-based waveguide and a driving electrode electrically connected to the tunable Bragg grating. The tunable Bragg grating adjusts its reflectivity or transmittance based on the thermo-optical effect or carrier injection effect to perform amplitude-weighted modulation on the optical signal transmitted through the corresponding input column waveguide and obtain multiple weighted optical signals.

[0008] Each summing waveguide is coupled to the optical weighting modulation unit in the corresponding row of the optical weighting modulation array. The summing waveguide is used to sum the optical power output by each optical weighting modulation unit in the same row. The weighted optical signal is laterally coupled into the corresponding summing waveguide. In the same summing waveguide, the optical power weighted by different input column waveguides and different rows of optical weighting modulation units is linearly accumulated in parallel to form the optical power corresponding to the input vector at the output of each summing waveguide.

[0009] The signal modulation unit is coupled to the corresponding summing waveguide, and the signal modulation unit is used to convert the optical power output by the corresponding summing waveguide into an electrical signal.

[0010] Based on the above technical solutions, preferably, each input port corresponds one-to-one with the optical signal in the input vector, each input port receives only a single center wavelength optical signal, and the optical power of the optical signal represents the component value of the corresponding input vector, so that the optical signals from different input ports in the same summing waveguide are physically incoherent.

[0011] Based on the above technical solution, preferably, the optical weight modulation array includes M×N optical weight modulation units, where N represents the input vector dimension and M represents the output vector dimension;

[0012] N input waveguides are arranged in parallel along a first direction, and M summing waveguides are arranged in parallel along a second direction perpendicular to the first direction. The optical weighting modulation unit is set in the intersection area of ​​each input waveguide and each summing waveguide. The optical weighting modulation unit is optically coupled to the corresponding input waveguide and summing waveguide respectively through a directional coupler or a lateral coupling structure.

[0013] More preferably, the driving electrode includes a metal micro-heating electrode disposed on the surface of the Bragg grating, the metal micro-heating electrode being used for amplitude-weighted modulation according to the thermo-optic effect;

[0014] Alternatively, a PN junction may be formed in the silicon waveguide of the Bragg grating region, the PN junction being used for amplitude-weighted modulation via carrier injection or extraction.

[0015] More preferably, at least one input port is configured to receive a wavelength-multiplexed optical signal comprising multiple wavelength channels with equal frequency spacing, each wavelength channel representing a subcomponent of the input vector, wherein the tunable Bragg grating provides different transmittance or reflectance for different wavelength channels to perform parallel matrix-vector multiplication operations within the optical weighting modulation array.

[0016] More preferably, the tunable Bragg grating is disposed in a silicon-based waveguide and satisfies the Bragg condition:

[0017] λ B = 2 n eff Λ

[0018] in, λ B Indicates the Bragg wavelength. n eff Λ represents the effective refractive index of the waveguide and Λ represents the grating period.

[0019] More preferably, the signal modulation unit includes a photodetector and an analog-to-digital converter, each photodetector being electrically connected to a corresponding analog-to-digital converter. The photodetector is used to convert the optical power output from the summing waveguide into a corresponding analog electrical signal, and the analog-to-digital converter is used to convert the analog electrical signal into a digital signal.

[0020] A second aspect of this application provides a computation method for a photon matrix-vector multiplier based on a tunable Bragg grating array, the computation method comprising:

[0021] The multiple optical signals corresponding to each component in the input vector are injected into the corresponding input column waveguides from multiple input ports, so that each optical signal is transmitted along the input column waveguides and optically coupled sequentially to the optical weight modulation units of the corresponding column in the optical weight modulation array.

[0022] By applying control electrical signals to each optical weighting modulation unit through the driving electrodes electrically connected to the tunable Bragg grating, the tunable Bragg grating adjusts the reflectivity or transmittance of the optical signals transmitted through the corresponding input waveguides based on the thermo-optical effect or carrier injection effect, so as to perform amplitude weighted modulation of each optical signal and obtain multi-weighted optical signals.

[0023] The weighted optical signals output from each optical weighting modulation unit are coupled into the summing waveguide corresponding to their respective rows via lateral coupling. In the same summing waveguide, the optical powers weighted by different input column waveguides and different rows of optical weighting modulation units are linearly accumulated in parallel to form the optical power corresponding to the input vector at the output of each summing waveguide.

[0024] The optical power at the output of each summing waveguide is input to its corresponding signal modulation unit, which then converts the optical power into an electrical signal for output.

[0025] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory.

[0026] A fourth aspect of this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of an operation method for a photon matrix-vector multiplier based on a tunable Bragg grating array.

[0027] The photon matrix-vector multiplier based on a tunable Bragg grating array provided by this invention has the following advantages over existing technologies:

[0028] (1) Through a two-dimensional structure consisting of an input column waveguide, an optical weighting modulation array, and a summing waveguide, the weighting of each element of the matrix to the vector components and the summation of row vectors are directly performed on the chip, realizing complete matrix-vector multiplication operations. This can significantly improve the operation bandwidth and speed. Furthermore, the Bragg grating can be tuned in combination with the thermo-optical effect or the carrier injection effect to finely adjust the reflectivity / transmittance, thereby realizing continuous and adjustable weighting of the amplitude of the optical signal in each channel. At the same time, each optical weighting modulation unit can be independently controlled, and the weight of the entire matrix can be updated and reconstructed online to meet the weight requirements of different algorithms and tasks. To meet the dynamic configuration requirements of the matrix, the weighted optical power output from multiple optical weight modulation units in the same row is directly superimposed through a summing waveguide to achieve linear accumulation within the row vector. This avoids the serial summation delay and bandwidth bottleneck in the electronic domain and effectively avoids phase noise and calculation errors caused by coherent interference at the same frequency, further improving overall computational efficiency and accuracy. Matrix operations are completed in the optical domain inside the chip, reducing the number of long-distance high-speed electrical interconnects and large-scale electronic multiply-accumulate units, thereby reducing overall power consumption and interconnect complexity. This is more conducive to achieving low-power computing at high throughput.

[0029] (2) By specifically defining the driving electrode as a metal micro-heating electrode set on the surface of the Bragg grating, the effective refractive index of the Bragg grating is adjusted by using the thermo-optic effect, thereby realizing the continuous adjustable amplitude weighted modulation of the grating reflectivity or transmittance. This can obtain a large dynamic range of weight adjustment and high weight stability, which is convenient for fine calibration and long-term maintenance of optical weights. It is beneficial to improve the accuracy and consistency of matrix-vector multiplication results. The metal micro-heating structure has a simple process and is compatible with mainstream silicon photonics processes, which helps to realize large-scale array integration. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 A schematic diagram of the system architecture of the photon matrix-vector multiplier provided by the present invention;

[0032] Figure 2 A cross-sectional view of the structure of a single optical weighting modulation unit provided by the present invention;

[0033] Figure 3 A schematic diagram of waveguide coupling for a 2×2 photonic matrix vector multiplier array provided by the present invention;

[0034] Figure 4 The present invention provides reflection / transmission spectrum curves of the tunable Bragg grating under different control currents;

[0035] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention.

[0036] Explanation of reference numerals in the attached figures: 1. Electronic device; 11. Processor; 12. Communication bus; 13. User interface; 14. Network interface; 15. Memory. Detailed Implementation

[0037] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0038] refer to Figure 1 This invention provides a photonic matrix-vector multiplier based on a tunable Bragg grating array, comprising multiple input ports, an optical weighting modulation array, multiple summing waveguides, and multiple signal modulation units, wherein...

[0039] The input port is used to receive multiple optical signals corresponding to each component in the input vector. Each input port is optically coupled to the corresponding column of the optical weighting modulation unit in the optical weighting modulation array through an input column waveguide.

[0040] Each input port corresponds one-to-one with the optical signal in the input vector. Each input port receives only a single center wavelength optical signal. The optical power of the optical signal represents the component value of the corresponding input vector, so that the optical signals from different input ports in the same summing waveguide are physically incoherent.

[0041] In one example, N input ports are connected to an input vector X=[x1,x2,…,X]. n The components of the vector are in a one-to-one correspondence, meaning each input port is responsible for receiving only one specific component of the vector. In the wavelength dimension, each port... j It contains only one specific center wavelength channel. , representing vector components The intensity of the signal. This encoding method ensures that the optical signals from different columns (different components) are physically incoherent in each row of the summing waveguide, thus enabling matrix summation by direct linear accumulation of optical power, effectively avoiding phase noise and calculation errors caused by coherent interference at the same frequency.

[0042] Furthermore, at least one input port is configured to receive a wavelength-multiplexed optical signal comprising multiple wavelength channels with equal frequency spacing, each wavelength channel representing a subcomponent of the input vector. The tunable Bragg grating provides different transmittance or reflectance for different wavelength channels to perform parallel matrix-vector multiplication operations within the optical weighted modulation array.

[0043] In this embodiment, by configuring at least one input port to receive a wavelength-multiplexed optical signal containing multiple wavelength channels with equal frequency spacing, and assigning each wavelength channel a subcomponent of the input vector, a tunable Bragg grating provides different transmittances or reflectivities for different wavelength channels, achieving independent amplitude-weighted modulation of multiple wavelength channels within the same spatial waveguide. Thus, within the optical weighted modulation array, not only can matrix-vector multiplication operations be performed in the spatial dimension (i.e., between different input column waveguides and different summing waveguides), but additional parallelism is also introduced in the wavelength dimension. This significantly improves the throughput and integration of matrix-vector multiplication without significantly increasing chip area or the number of waveguides.

[0044] Because tunable Bragg gratings offer selective spectral responses and independently adjustable reflectivity or transmittance for wavelength channels with equal frequency intervals, and the weight settings for each wavelength channel do not interfere with each other, the weight values ​​for different wavelength channels can be flexibly configured as needed. This enables parallel computation of different sub-vectors or multiple sets of input vectors, thereby improving the system's reconfigurability and functional density. Simultaneously, wavelength multiplexing technology carries multiple computation signals in the spectral dimension, reducing reliance on additional physical channels and electrical interconnects. This helps reduce system power consumption and interconnect complexity, enhancing the scalability and application value of large-scale optical matrix computing systems.

[0045] The optical weighting modulation array includes multiple optical weighting modulation units. Each optical weighting modulation unit includes a tunable Bragg grating integrated on a silicon-based waveguide and a driving electrode electrically connected to the tunable Bragg grating. The tunable Bragg grating adjusts the reflectivity or transmittance based on the thermo-optical effect or the carrier injection effect to perform amplitude weighting modulation on the optical signal transmitted through the corresponding input column waveguide and obtain multiple weighted optical signals.

[0046] The optical weighted modulation array comprises M×N optical weighted modulation units, where N represents the input vector dimension and M represents the output vector dimension.

[0047] N input waveguides are arranged in parallel along a first direction, and M summing waveguides are arranged in parallel along a second direction perpendicular to the first direction. An optical weighting modulation unit is set in the intersection region of each input waveguide and each summing waveguide. The optical weighting modulation unit is optically coupled to the corresponding input waveguide and summing waveguide respectively through a directional coupler or a lateral coupling structure.

[0048] Each optical weighting modulation unit is integrated into a two-dimensional array via a side-coupled or broadcast-and-weighted structure. Specifically, from the input port... j The optical signal propagates in the vertical column waveguide and, through a directional coupler, couples some of its energy into the corresponding M optical weighting modulation units in that column. The modulated optical signal is then injected into the horizontal row summing waveguide via lateral coupling. This parallel coupling structure avoids the cumulative insertion loss and multiple reflection noise caused by the optical signal passing through the grid step by step in the traditional series structure, thus ensuring the signal integrity of the large-scale array.

[0049] Furthermore, the tunable Bragg grating is disposed in the silicon-based waveguide and satisfies the Bragg condition:

[0050] λ B = 2 n eff Λ

[0051] in, λ B Indicates the Bragg wavelength. n eff Λ represents the effective refractive index of the waveguide and Λ represents the grating period. By changing the voltage applied to the tunable Bragg grating, the refractive index of the silicon waveguide is altered using the thermo-optical effect, thereby changing the reflectivity and achieving weighting. Modulation. The transmittance of the optical matrix W and the tunable Bragg grating. Proportional, which can be expressed as:

[0052]

[0053] in, This represents the voltage applied to the corresponding tunable Bragg grating. The spectral shape function of a tunable Bragg grating is represented. Indicates weight With the square of the voltage V 2 The correlation function.

[0054] Furthermore, the driving electrode includes a metal micro-heating electrode disposed on the surface of the Bragg grating, which is used for amplitude-weighted modulation based on the thermo-optic effect;

[0055] Alternatively, a PN junction may be formed in a silicon waveguide in the Bragg grating region, and the PN junction may be used for amplitude-weighted modulation by carrier injection or extraction.

[0056] In this embodiment, by specifically defining the driving electrode as a metal micro-heating electrode disposed on the surface of the Bragg grating, the effective refractive index of the Bragg grating is adjusted using the thermo-optical effect. This achieves continuously adjustable amplitude-weighted modulation of the grating's reflectivity or transmittance, resulting in a large dynamic range for weight adjustment and high weight stability. This facilitates precise calibration and long-term maintenance of optical weights, improving the accuracy and consistency of matrix-vector multiplication results. Furthermore, the metal micro-heating structure is simple to manufacture, compatible with mainstream silicon photonics processes, and facilitates large-scale array integration.

[0057] By implementing the driving electrode as a PN junction formed in the silicon waveguide of the Bragg grating region, and altering the waveguide refractive index and loss through carrier injection or descent, high-speed electrically controlled modulation of the Bragg grating amplitude response can be achieved. This significantly improves the weight update speed and device response bandwidth, meeting the application requirements of online training and rapid reconstruction. The PN junction modulation method features low driving voltage and high modulation efficiency, making it suitable for large-scale integration and row-column independent control, which is beneficial for building low-power, high-speed, and reconfigurable optical matrix computing chips. In summary, the above two driving electrode structures provide diverse and selectable implementation paths for optical weight modulation units, ensuring process compatibility while considering performance indicators such as modulation accuracy, speed, and power consumption, further improving the overall performance and application flexibility of photonic matrix vector multipliers.

[0058] Each summing waveguide is coupled to the optical weighting modulation unit in the corresponding row of the optical weighting modulation array. The summing waveguide is used to sum the optical power output by each optical weighting modulation unit in the same row. The weighted optical signal enters the corresponding summing waveguide through lateral coupling. In the same summing waveguide, the optical power after being weighted by different input column waveguides and different rows of optical weighting modulation units is linearly accumulated in parallel to form the optical power corresponding to the input vector at the output of each summing waveguide.

[0059] Each signal modulation unit is coupled to a corresponding summing waveguide, and the signal modulation unit is used to convert the optical power output from the corresponding summing waveguide into an electrical signal.

[0060] The signal modulation unit includes a photodetector and an analog-to-digital converter. Each photodetector is electrically connected to a corresponding analog-to-digital converter. The photodetector is used to convert the optical power output from the summing waveguide into a corresponding analog electrical signal, and the analog-to-digital converter is used to convert the analog electrical signal into a digital signal.

[0061] In this embodiment, a two-dimensional structure consisting of an input column waveguide, an optical weighting modulation array, and a summing waveguide is used to directly perform the weighting of each element of the matrix with respect to the vector components and the summation of row vectors on the chip, realizing complete matrix-vector multiplication operations. This significantly improves the computational bandwidth and speed. Furthermore, the tunable Bragg grating, combined with thermo-optical effects or carrier injection effects, allows for fine adjustment of reflectivity / transmittance, thereby achieving continuously adjustable weighting of the optical signal amplitude of each channel. Simultaneously, each optical weighting modulation unit can be independently controlled, and the entire matrix weights can be updated and reconstructed online to meet the needs of different algorithms and tasks. To meet the dynamic configuration requirements of the weight matrix, the weighted optical power output from multiple optical weight modulation units in the same row is directly superimposed through a summing waveguide to achieve linear accumulation within the row vector. This avoids the serial summation delay and bandwidth bottleneck in the electronic domain and effectively avoids phase noise and calculation errors caused by coherent interference at the same frequency, further improving overall computational efficiency and accuracy. Matrix operations are completed in the optical domain inside the chip, reducing the number of long-distance high-speed electrical interconnects and large-scale electronic multiply-accumulate units, thereby reducing overall power consumption and interconnect complexity. This is more conducive to achieving low-power computing at high throughput.

[0062] In one example, the photonic matrix-vector multiplier includes: N input ports, M*N optical weighting modulation units, M summing waveguides, and M photodetectors and analog-to-digital converters, where...

[0063] N input ports are used to couple N wavelength division multiplexing (WDM) input vector signals. Each of the N input optical ports corresponds to one component of the input vector. x jThe input light source can be a continuous light source or a high-speed modulated laser, and the light intensity can be encoded through an external modulator or direct modulation.

[0064] M*N optical weighted modulation units constitute a two-dimensional array, where M is the output dimension and N is the input dimension. Each input port is connected to a column waveguide extending vertically, and multiple directional couplers distribute optical power to the multiple optical weighted modulation units in that column. Each optical weighted modulation unit is integrated between its corresponding column waveguide and row waveguide, connected to the row waveguide via lateral coupling. All optical weighted modulation units constitute the optical weighted modulation array, and the weight matrix W is determined by the transmission / reflection characteristics of each optical weighted modulation unit.

[0065] Each optical weighting modulation unit consists of a tunable Bragg grating (TBG) integrated on a silicon-based waveguide. Its reflectivity or transmittance is precisely controlled through thermo-optical effects or carrier injection effects, thereby modulating the amplitude of the incident light signal, i.e., setting the matrix weights. W ij .

[0066] For the first optical weight modulation array i The row summation waveguide collects the optical signals modulated by all optical weighting modulation units in that row, realizing:

[0067]

[0068] in, Indicates the first i The weighted optical signal is obtained from summing waveguides, where M represents the number of summing waveguides, corresponding to the number of rows in the matrix and the dimension of the output vector; N represents the number of input optical signal channels, corresponding to the number of columns in the matrix and the dimension of the input vector. i Indicates the first i The index of a summing waveguide, j Indicates the first j Index of each input port / column, Indicates the first j The input optical signal, i.e., the input vector of the _th ... j One component; Indicates that it is located at the th i line, number j The optical weights of the columns are modulated by the corresponding optical weight modulation units.

[0069] Further, see Figure 2 A single tunable Bragg grating may include: a silicon waveguide with a Bragg grating region and a modulation structure, wherein,

[0070] The silicon waveguide and Bragg grating region involves fabricating a single-mode silicon waveguide on a silicon-based insulator photonic integration platform, and etching a periodic structure in a certain section of the waveguide to form a Bragg grating.

[0071] One way to implement the modulation structure is to deposit a metal heating electrode above the waveguide, and then generate localized heating by passing a controlled current to change the refractive index of silicon (thermo-optic effect). Another way is to form a P-N junction injection region in the waveguide area, and change the refractive index and absorption characteristics by injecting charge carriers.

[0072] Each optical weighting modulation unit is connected to an independent driving circuit, and achieves individual weighting through DAC output voltage or current. Independent adjustment. During actual calibration, the control voltage / current can be scanned and the output transmitted or reflected power measured to establish... The corresponding relationships are stored in the control system for subsequent rapid setting of matrix weights.

[0073] M summing waveguides are coupled to a weighted modulation unit array. Each summing waveguide collects all TBG-modulated optical signals from the same row i, and performs matrix multiplication accumulation by photonic parallel summation (based on passive optical coupling or simple waveguide coupling).

[0074] M photodetectors and an analog-to-digital converter are used to convert the final optical summation result into an electrical signal output.

[0075] In one example Figure 3 A schematic diagram of waveguide coupling for a 2×2 photonic matrix vector multiplier array according to the present invention is shown.

[0076] In this embodiment, the matrix-vector multiplier includes two input optical signals X1 and X2 and two output optical signals Y1 and Y2. Inputs X1 and X2 are introduced into the chip via two input waveguides, and after passing through a waveguide beam splitter / coupling structure, they are coupled to four optical weighting modulation units located in the central region. The four optical weighting modulation units are preferably tunable Bragg gratings (TBGs), each corresponding to a weight W. 11 W 12 W 21 W 22 This forms a 2×2 weight array.

[0077] The optical signal from input X1 is distributed to the optical weighting modulation unit W through the upper input waveguide. 11 and W 21 The optical signal from input X2 is distributed to the weighting unit W through the lower input waveguide. 12 and W 22 The optical weighting modulation unit performs amplitude modulation on the corresponding input optical signal coupled in, and its transmission or reflection intensity represents the matrix weights, respectively. The modulated optical signal is then converged to two output summing waveguides through a waveguide coupling structure. The upper summing waveguide corresponds to output Y1, and the lower summing waveguide corresponds to output Y2.

[0078] Therefore, the upper output summing waveguide will come from the weighting unit W. 11 and W 12 The optical power is accumulated in parallel to obtain:

[0079] Y1=W 11 ·X1+W 12 ·X2;

[0080] The output summation waveguide below will come from the weighting unit W. 21 and W 22 The optical power is accumulated in parallel to obtain:

[0081] Y2=W 21 ·X1+W 22 ·X2;

[0082] By independently adjusting the driving voltages of the four optical weighting modulation units, programmable loading of any 2×2 real-valued matrix can be achieved on the chip, thereby completing the input vector [X1,X2]. T The matrix-vector multiplication operation with the weight matrix W outputs a vector [Y1, Y2]. T .

[0083] Please see Figure 4 , Figure 4 The transmission spectrum curves of the tunable Bragg grating under different control currents are shown:

[0084] When the control current I=0, the transmission spectrum of the tunable Bragg grating exhibits a fixed shape (curve A), with obvious reflection / transmission characteristics near the Bragg wavelength.

[0085] As the control current gradually increases, the Bragg wavelength, grating bandwidth, and depth change, causing the corresponding transmission curve to drift towards longer or shorter wavelengths, accompanied by changes in transmittance (curves B and C).

[0086] When the input signal wavelength is fixed λ single At that time, as the control current increases, λ single The transmittance or reflectance at a certain point changes monotonically, thereby achieving weighting. Continuously adjustable.

[0087] Based on the above method, this application discloses an operation method for a photon matrix-vector multiplier based on a tunable Bragg grating array, the operation method including:

[0088] Step S1: Inject the multi-channel optical signals corresponding to each component in the input vector into the corresponding input column waveguides from multiple input ports, so that each optical signal is transmitted along the input column waveguides and optically coupled sequentially to the optical weight modulation units of the corresponding column in the optical weight modulation array.

[0089] Step S2: Apply control electrical signals to each optical weighting modulation unit through the driving electrode electrically connected to the tunable Bragg grating, so that the tunable Bragg grating adjusts the reflectivity or transmittance of the optical signal transmitted through the corresponding input column waveguide based on the thermo-optical effect or carrier injection effect, so as to perform amplitude weighted modulation of each optical signal and obtain multi-weighted optical signals.

[0090] Step S3: The weighted optical signal output from each optical weighting modulation unit is coupled into the summing waveguide corresponding to its row via lateral coupling. In the same summing waveguide, the optical power after being weighted by different input column waveguides and different row optical weighting modulation units is linearly accumulated in parallel to form the optical power corresponding to the input vector at the output of each summing waveguide.

[0091] Step S4: Input the optical power at the output of each summing waveguide to its corresponding signal modulation unit, and the signal modulation unit converts the optical power into an electrical signal for output.

[0092] In this embodiment, each component of the input vector is mapped to multiple optical signals and injected into the corresponding input column waveguides. Each optical signal is then optically coupled sequentially to the corresponding column optical weighting modulation unit along a predetermined path, achieving ordered mapping and efficient import of the input vector into on-chip optical signals. This facilitates a regular and scalable matrix operation channel layout in a two-dimensional array structure. By applying control signals to the tunable Bragg grating through driving electrodes, the reflectivity or transmittance is precisely adjusted using thermo-optical effects or carrier injection effects. This enables continuously adjustable weighting of the amplitudes of each optical signal, allowing for directly programmable setting and reconstruction of matrix weights in the optical domain, thus improving the flexibility and reconfigurability of matrix-vector multiplication operations.

[0093] By performing parallel linear accumulation of optical powers from different input columns modulated by optical weighting modulation units in different rows within the same summing waveguide, hardware-level parallel summation of the multiplication and addition operations of each row of the matrix with the input vector is achieved in the optical domain. This eliminates the need for complex electronic addition networks, significantly reducing computational latency and interconnection overhead, and improving overall computational throughput and energy efficiency. The optical power output from each summing waveguide is input to the corresponding signal modulation unit and converted into an electrical signal output, allowing the optical domain computation results to be easily interfaced with subsequent electronic circuits, data conversion, and control units, forming a hybrid signal processing architecture combining optical computation and electrical readout. This computational method achieves precisely adjustable weights, highly integrated structure, and user-friendly photonic matrix-vector multiplication operations while ensuring high computational speed and high parallelism.

[0094] Please see Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 As shown, electronic device 1 may include: at least one processor 11, at least one network interface 14, user interface 13, memory 15, and at least one communication bus 12.

[0095] The communication bus 12 is used to enable communication between these components.

[0096] The user interface 13 may include a display screen and a camera. Optionally, the user interface 13 may also include a standard wired interface and a wireless interface.

[0097] The network interface 14 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0098] The processor 11 may include one or more processing cores. The processor 11 connects to various parts of the server via various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 15, and by calling data stored in the memory 15. Optionally, the processor 11 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 11 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 11.

[0099] The memory 15 may include random access memory (RAM) or read-only memory. Optionally, the memory 15 may include non-transitory computer-readable storage medium. The memory 15 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 15 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the various method embodiments described above, etc.; the data storage area may store data involved in the various method embodiments described above, etc. Optionally, the memory 15 may also be at least one storage device located remotely from the aforementioned processor 11. Figure 5 As shown, the memory 15, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an operation program based on a photon matrix vector multiplier of a tunable Bragg grating array.

[0100] exist Figure 5In the electronic device 1 shown, the user interface 13 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 11 can be used to call the operation program of a photon matrix vector multiplier based on a tunable Bragg grating array stored in the memory 15. When executed by one or more processors, the electronic device performs one or more methods as described in the above embodiments.

[0101] A non-transitory computer-readable storage medium stores instructions that, when executed by one or more processors, cause a computer to perform one or more methods as described in the above embodiments.

[0102] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0103] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0105] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0107] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A photonic matrix-vector multiplier based on a tunable Bragg grating array, characterized in that, It includes multiple input ports, an optical weighted modulation array, multiple summing waveguides, and multiple signal modulation units, among which, The input port is used to receive multiple optical signals corresponding to each component in the input vector. Each input port is optically coupled to the optical weighting modulation unit in the corresponding column of the optical weighting modulation array through an input column waveguide. Each input port corresponds one-to-one with the optical signal in the input vector. Each input port receives only a single center wavelength optical signal. The optical power of the optical signal represents the component value of the corresponding input vector, so that the optical signals from different input ports in the same summing waveguide are physically incoherent. The optical weighting modulation array includes multiple optical weighting modulation units. Each optical weighting modulation unit includes a tunable Bragg grating integrated on a silicon-based waveguide and a driving electrode electrically connected to the tunable Bragg grating. The tunable Bragg grating adjusts its reflectivity or transmittance based on the thermo-optical effect or carrier injection effect to perform amplitude-weighted modulation on the optical signal transmitted through the corresponding input column waveguide and obtain multiple weighted optical signals. The optical weighting modulation array includes M×N optical weighting modulation units, where N represents the input vector dimension and M represents the output vector dimension. N input waveguides are arranged in parallel along a first direction, and M summing waveguides are arranged in parallel along a second direction perpendicular to the first direction. The optical weighting modulation unit is set in the intersection area of ​​each input waveguide and each summing waveguide. The optical weighting modulation unit is optically coupled to the corresponding input waveguide and summing waveguide respectively through a directional coupler or a lateral coupling structure. Each optical weighting modulation unit is integrated into a two-dimensional array via lateral coupling or a broadcast-weighting structure, originating from the input port. j The optical signal propagates in the vertical waveguide and a portion of its energy is coupled into the input port via a directional coupler. j The corresponding target column has M optical weighting modulation units, and the modulated optical signal is then injected into the horizontal row summing waveguide through lateral coupling. Each summing waveguide is coupled to the optical weighting modulation unit in the corresponding row of the optical weighting modulation array. The summing waveguide is used to sum the optical power output by each optical weighting modulation unit in the same row. The weighted optical signal is laterally coupled into the corresponding summing waveguide. In the same summing waveguide, the optical power weighted by different input column waveguides and different rows of optical weighting modulation units is linearly accumulated in parallel to form the optical power corresponding to the input vector at the output of each summing waveguide. The signal modulation unit is coupled to the corresponding summing waveguide, and the signal modulation unit is used to convert the optical power output by the corresponding summing waveguide into an electrical signal. The driving electrode includes a metal micro-heating electrode disposed on the surface of the Bragg grating, and the metal micro-heating electrode is used for amplitude weighted modulation according to the thermo-optic effect; Alternatively, a PN junction may be formed in the silicon waveguide of the Bragg grating region, the PN junction being used for amplitude-weighted modulation via carrier injection or extraction.

2. The photon matrix-vector multiplier based on a tunable Bragg grating array as described in claim 1, characterized in that, At least one input port is configured to receive a wavelength-multiplexed optical signal comprising multiple wavelength channels with equal frequency spacing, each wavelength channel representing a subcomponent of the input vector, wherein the tunable Bragg grating provides different transmittance or reflectance for different wavelength channels to perform parallel matrix-vector multiplication operations within the optical weighted modulation array.

3. The photon matrix-vector multiplier based on a tunable Bragg grating array as described in claim 1, characterized in that, The tunable Bragg grating is disposed in a silicon-based waveguide and satisfies the Bragg condition: λ B = 2 n eff L in, λ B Indicates the Bragg wavelength. n eff Λ represents the effective refractive index of the waveguide and Λ represents the grating period.

4. A photon matrix-vector multiplier based on a tunable Bragg grating array as described in claim 1, characterized in that, The signal modulation unit includes a photodetector and an analog-to-digital converter. Each photodetector is electrically connected to a corresponding analog-to-digital converter. The photodetector is used to convert the optical power output from the summing waveguide into a corresponding analog electrical signal, and the analog-to-digital converter is used to convert the analog electrical signal into a digital signal.

5. A computational method for a photon matrix-vector multiplier based on a tunable Bragg grating array, employing the photon matrix-vector multiplier based on a tunable Bragg grating array as described in any one of claims 1 to 4, characterized in that, The calculation method includes: The multiple optical signals corresponding to each component in the input vector are injected into the corresponding input column waveguides from multiple input ports, so that each optical signal is transmitted along the input column waveguides and optically coupled sequentially to the optical weight modulation units of the corresponding column in the optical weight modulation array. By applying control electrical signals to each optical weighting modulation unit through the driving electrodes electrically connected to the tunable Bragg grating, the tunable Bragg grating adjusts the reflectivity or transmittance of the optical signals transmitted through the corresponding input waveguides based on the thermo-optical effect or carrier injection effect, so as to perform amplitude weighted modulation of each optical signal and obtain multi-weighted optical signals. The weighted optical signals output from each optical weighting modulation unit are coupled into the summing waveguide corresponding to their respective rows via lateral coupling. In the same summing waveguide, the optical powers weighted by different input column waveguides and different rows of optical weighting modulation units are linearly accumulated in parallel to form the optical power corresponding to the input vector at the output of each summing waveguide. The optical power at the output of each summing waveguide is input to its corresponding signal modulation unit, which then converts the optical power into an electrical signal for output.

6. An electronic device, characterized in that, The device includes a processor (11), a memory (15), a user interface (13), and a network interface (14). The memory (15) is used to store instructions. The user interface (13) and the network interface (14) are used to communicate with other devices. The processor (11) is used to execute the instructions stored in the memory (15) so that the electronic device (1) performs the operation method as described in claim 5.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the operation method described in claim 5.

Images