A two-dimensional photon coherent convolution operation chip and its application system

Abstract

The present invention discloses a two-dimensional photonic coherent convolution operation chip and its application system, belonging to the field of optoelectronic integration technology. In the present invention, an intensity modulator, a coherent delay weighting unit, and a first-stage delay waveguide for completing convolution operations are integrally integrated through photonic integration technology. The first-stage delay waveguide is used to achieve the first-stage time interleaving of intensity-modulated optical signals; the coherent delay weighting unit including an amplitude controller, a phase shifter, and a second-stage delay waveguide is used to achieve the second-stage time interleaving of signals to be convolved, the convolution kernel matrix coefficient weighting, and coherent summation to obtain a first-stage coherent summation optical signal. The first-stage coherent summation optical signal is fused in the optical domain to obtain a second-stage coherent summation optical signal. The second-stage coherent summation optical signal and a reference optical signal are fused in the optical domain to obtain a coherent summation optical signal, and the coherent summation optical signal completes photoelectric conversion to achieve convolution summation operation. By using light as the information carrier, the present invention can greatly improve the rate and energy efficiency ratio of convolution operations.

Description

Technical Field

[0001] The present invention relates to a coherent convolution operation chip for a convolutional neural network, specifically a two-dimensional photonic coherent convolution operation chip based on time interleaving and its application system, belonging to the field of photonic integration technology. Background Art

[0002] Matrix operation, as the main operation in a convolutional neural network, occupies most of the computing power of the convolutional neural network. Currently, the training and testing of mainstream neural network models mainly use electronic integrated chips such as CPUs, GPUs, and TPUs. Limited by the computing structure with separate program space and data space in electronic chips, the microscopic quantum characteristics and macroscopic high-frequency response characteristics of electronic chips, there are problems such as unstable data load between computing units and high power consumption in the convolutional neural network operation based on electronic chips. In addition, with the rapid development of artificial intelligence technology, the demand for computing power of neural network models is increasing rapidly. However, with the failure of Moore's law, the performance of future artificial intelligence application models based on electronic chips will be restricted. Photonic technology with photons as information carriers has characteristics such as large bandwidth, low loss, and parallelizability. Combining photonic technology with traditional neural networks is expected to give full play to the advantages of both technologies and break through the technical development bottlenecks of traditional electronic neural networks, such as high power consumption, long delay, and limited speed. However, most of the currently proposed photonic convolution operation methods are based on wavelength division multiplexing technology, and the implementation basis of wavelength division multiplexing technology is a multi-wavelength light source. The implementation methods of multi-wavelength light sources are often complex and costly. Summary of the Invention

[0003] The technical problem to be solved by the present invention is: to overcome the deficiencies of the prior art, based on photonic integration technology, use a first-level delay waveguide to achieve the first-level time interleaving of intensity-modulated optical signals; use a coherent delay weighting unit including an amplitude controller, a phase shifter, and a second-level delay waveguide to achieve the second-level time interleaving of the signal to be convolved, convolution kernel matrix coefficient weighting, and coherent summation to obtain a first-level coherent summation optical signal. The first-level coherent summation optical signal is fused in the optical domain to obtain a second-level coherent summation optical signal. The second-level coherent summation optical signal and the reference optical signal are fused in the optical domain to obtain a coherent summation optical signal. After photoelectric conversion by a photodetector, the weighted summation operation is realized. Without wavelength division multiplexing, it is applicable to multi-dimensional data convolution operations.

[0004] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0005] A two-dimensional photonic coherent convolution operation chip, which is integrally integrated by an intensity modulator, K coherent delay weighting units, and K - 1 first-level delay waveguides; where:

[0006] The optical input end of the entire chip is one optical input end of the first 1×2 optical coupler, which is used to receive external optical signals; one optical output end of the first 1×2 optical coupler is connected to one optical input end of the first 2×1 optical coupler to provide a reference optical signal, and the other optical output end is connected to the optical input end of the intensity modulator to provide a carrier optical signal;

[0007] The electrical input end of the intensity modulator is used to receive an external signal to be convolved. The signal to be convolved intensity-modulates the optical signal input to the intensity modulator through the intensity modulator to obtain an intensity-modulated optical signal; the optical output ends of the intensity modulator and the first K-2 first-order delay waveguides are connected to the optical input end of an optical coupler. The two optical output ends of the optical coupler are used as the two sub-output ends connecting to the corresponding intensity modulator and the first K-2 first-order delay waveguides. The intensity modulator and the K-1 first-order delay waveguides are connected in series through one sub-output end. The other sub-output end of the intensity modulator is connected to the input end of the first coherent delay weighting unit. The other sub-output ends of the first K-2 first-order delay waveguides and the output end of the last first-order delay waveguide are sequentially connected to the input ends of the K-1 coherent delay weighting units; the coherent delay weighting unit realizes the convolution kernel matrix coefficient weighting and coherent summation of the intensity-modulated optical signal to obtain the corresponding first-order coherent summation optical signal; the K first-order coherent summation optical signals output from the optical output ends of the K coherent delay weighting units are fused in the optical domain through one or more optical couplers to obtain a second-order coherent summation optical signal. The second-order coherent summation optical signal is sent to the other optical input end of the first 2×1 optical coupler to be fused with the reference optical signal in the optical domain to obtain a coherent summation optical signal. The optical output end of the first 2×1 optical coupler is the optical output end of the entire chip, and the coherent summation optical signal is output through the optical output end of the chip;

[0008] The coherent delay weighting unit is composed of L amplitude controllers, L phase shifters, a signal distribution unit, and a signal summation unit. Among them, the signal distribution unit is a one-to-two tree structure formed by cascading L-1 1×2 optical couplers and L-1 delay waveguides respectively located at one output end of the 1×2 optical couplers, including 1 input end and L output ends. The signal summation unit is a two-in-one tree structure formed by cascading L-1 2×1 optical couplers, including L input ends and 1 output end; the L output ends of the signal distribution unit, the L amplitude controllers, the L phase shifters, and the input ends of the signal summation unit are connected in sequence one by one. The L output ends of the signal summation unit obtain L time-interleaved sub-intensity-modulated optical signals; the convolution kernel matrix control signal controls the L amplitude controllers and the L phase shifters to respectively weight the L time-interleaved sub-intensity-modulated optical signals according to the convolution kernel matrix coefficients to obtain L time-interleaved sub-weighted modulated optical signals; the signal summation unit realizes the coherent summation of the L time-interleaved sub-weighted modulated optical signals to obtain a first-order coherent summation optical signal; where L = 2 Z, where Z is the number of levels of the tree structure of the signal distribution unit and the signal receiving unit.

[0009] Further, in the coherent delay weighting unit, the delay waveguide lengths corresponding to different levels in the signal distribution unit are Δlr = 2 (Z-z) cΔt / n w , where c is the speed of light in vacuum, n w is the effective refractive index of the waveguide delay line, Δt = 1 / S M is the duration of a single symbol of the signal to be convolved, S M is the symbol rate of the signal to be convolved, z = 1, 2, 3, …, Z, corresponding to the levels of the tree structure of the signal distribution unit. Further, the phase shifter is a thermally controlled phase shifter or an electrically controlled phase shifter; the amplitude controller is a Mach-Zehnder modulator, a microring modulator, an optical amplifier or an optical attenuator.

[0010] Further, the convolution kernel matrix control signal controls L amplitude controllers and L phase shifters to weight L time-interleaved sub-intensity-modulated optical signals according to the convolution kernel matrix coefficients respectively, specifically:

[0011] Determine the transmission characteristics of the amplitude controller and the phase of the phase shifter respectively according to the magnitude and positive / negative sign of the convolution kernel matrix coefficients. In each coherent delay weighting unit, the L amplitude controllers and the corresponding L phase shifters correspond to one row of coefficients in the convolution kernel matrix. In K coherent delay weighting units, the K×L amplitude controllers and the corresponding K×L phase shifters correspond to a two-dimensional convolution kernel matrix of size K×L.

[0012] Further, the length of the first-stage delay waveguide is ΔL = (P - L + 1)cΔt / n w , where P is the number of columns of the two-dimensional data matrix to be convolved. Further, the chip is integrated based on the III-V material integration process or the silicon-based integration process.

[0013] Further, the intensity modulator is a Mach-Zehnder modulator or a microring modulator.

[0014] Further, the signal to be convolved is a one-dimensional time series obtained after flattening the two-dimensional data to be convolved. The two-dimensional data to be convolved is obtained by matrix transformation of the original two-dimensional data. The specific transformation process is:

[0015] The original two-dimensional data A Q×O is slid and segmented in the column direction with a step of P - L + 1 into H sub-two-dimensional data B Q×P , and each sub-two-dimensional data is a two-dimensional data to be convolved, where Q is the number of rows of the original two-dimensional data, O is the number of columns of the original two-dimensional data, and P is the number of columns of the two-dimensional data matrix to be convolved.

[0016] Further, the original two-dimensional data is obtained by decomposing three-dimensional or multi-dimensional original data.

[0017] Based on the same principle, the present invention also provides a convolution operation application system for a two-dimensional photonic coherent convolution operation chip, including:

[0018] The two-dimensional photonic coherent convolution operation chip, a light source, a signal source to be convolved, a convolution kernel control unit, an optical amplifier, a photodetector, and an acquisition and processing unit; wherein, the output end of the signal source to be convolved is connected to the electrical input end of the two-dimensional photonic coherent convolution operation chip, the optical output end of the light source is connected to the optical input end of the two-dimensional photonic coherent convolution operation chip, and the convolution kernel control unit is used to respectively determine the transmission coefficient of the control amplitude controller and the phase of the phase shifter according to the magnitude and positive and negative signs of the convolution kernel matrix coefficients; the input end of the optical amplifier is connected to the optical output end of the two-dimensional photonic coherent convolution operation chip for amplifying the coherent summation signal; the optical input end of the photodetector is connected to the optical output end of the optical amplifier for completing the optoelectronic conversion of the coherent summation signal; the acquisition and processing unit is connected to the electrical output end of the photodetector for reconstructing the electrical output signal to obtain the characteristic signal after the two-dimensional convolution operation of the signal to be convolved.

[0019] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0020] 1) The present invention is based on a single wavelength to carry the signal to be convolved, uses coherent technology to realize the weighting of the real-domain convolution kernel matrix coefficients, and uses the reference optical signal to realize the amplitude bias loading of the coherent summation signal. Compared with the wavelength division multiplexing technology, it does not require multi-wavelength optical signals, and the scheme is simple and compact.

[0021] 2) The present invention realizes two-stage time interleaving of the single-wavelength weighted modulation optical signal through two-stage delay waveguides, and can realize the two-dimensional convolution kernel convolution acceleration operation of two-dimensional data within a single signal period, solving the data redundancy problem of traditional methods, and the scheme is simple and efficient.

[0022] 3) The present invention connects the coherent delay weighting units in a series-parallel combination manner through two-stage delay waveguides to realize the multiplexing of the modulation optical signal delay waveguides, reducing the chip size and the signal transmission loss caused by the delay waveguides, thereby improving the energy utilization efficiency of the chip. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic structural diagram of a two-dimensional photonic coherent convolution operation chip according to an exemplary embodiment of the present invention.

[0024] Figure 2 It is a schematic structural diagram of a coherent delay weighting unit in a two-dimensional photonic coherent convolution operation chip according to an exemplary embodiment of the present invention.

[0025] Figure 3 This is a schematic structural diagram of a specific embodiment of a two-dimensional photonic coherent convolution operation chip of the present invention.

[0026] Figure 4 This is a schematic diagram of the matrix transformation process from the original two-dimensional data to the two-dimensional data to be convolved in a specific embodiment of a two-dimensional photonic coherent convolution operation chip of the present invention.

[0027] Figure 5 This is a schematic diagram of the flattening process of the two-dimensional data matrix to be convolved in a specific embodiment of a two-dimensional photonic coherent convolution operation chip of the present invention. Figure 5 In [diagram], A is the two-dimensional data matrix to be convolved and the convolution kernel matrix. Figure 5 In [diagram], B is a schematic diagram of the one-dimensional flattening process of the two-dimensional data matrix to be convolved. Figure 5 In [diagram], C is the two-dimensional feature data matrix reconstructed from one-dimensional feature data.

[0028] Figure 6 This is a diagram showing the relationship between the time series and wavelength of the intensity-modulated optical signal output by the intensity modulator in a specific embodiment of a two-dimensional photonic coherent convolution operation chip of the present invention.

[0029] Figure 7 This is a schematic diagram of the spectra of each working node in a specific embodiment of a two-dimensional photonic coherent convolution operation chip of the present invention. Figure 7 In [diagram], A is a diagram showing the relationship between the time series of the weighted modulation optical signals output by the 4 phase shifters in the first coherent delay weighting unit and each phase shifter. Figure 7 In [diagram], B is a diagram showing the relationship between the time series and wavelength of the first-level coherent summation optical signal obtained at the output end of the first coherent delay weighting unit. Figure 7 In [diagram], C is a diagram showing the relationship between the time series of the weighted modulation optical signals output by the 4 phase shifters in the fourth coherent delay weighting unit and each phase shifter. Figure 7 In [diagram], D is a diagram showing the relationship between the time series and wavelength of the first-level coherent summation optical signal obtained at the output end of the fourth coherent delay weighting unit. Figure 7 In [diagram], E is a diagram showing the relationship between the time series and wavelength of the 4 first-level coherent summation optical signals and the reference optical signal obtained at the output ends of the 4 coherent delay weighting units. Figure 7 In [diagram], F is a diagram showing the relationship between the time series and wavelength of the coherent summation optical signal.

[0030] Figure 8 This is a schematic diagram of the matrix transformation process from two-dimensional feature data to the original two-dimensional feature data in a specific embodiment of a two-dimensional photonic convolution operation chip of the present invention. Detailed implementation mode

[0031] Aiming at the deficiencies of the prior art, the idea of the present invention is based on photon integration technology. The first-level time interleaving of intensity-modulated optical signals is realized by using a first-level delay waveguide; the second-level time interleaving, convolution kernel matrix coefficient weighting, and coherent summation of the signals to be convolved are realized by using a coherent delay weighting unit including an amplitude controller, a phase shifter, and a second-level delay waveguide, reducing the requirement for multi-wavelength signals in the traditional method.

[0032] Figure 1 Shown is a schematic structural diagram of a two-dimensional photon coherent convolution operation chip according to the present invention, as Figure 1 shown. The photon components integrated on the two-dimensional photon coherent convolution operation chip are: an intensity modulator, K coherent delay weighting units, and K - 1 first-level delay waveguides (K = 4 is shown in the figure); the photon components are connected through 1×2 optical couplers (OCs), 2×1 optical couplers, and optical waveguides; the 1×2 optical coupler includes 1 optical input end and 2 optical output ends, and the 2×1 optical coupler includes 2 optical input ends and 1 optical output end; wherein the optical input end of the first 1×2 optical coupler is the optical input end of the entire chip, used to receive external optical signals, 1 optical output end is connected to one optical input end of the first 2×1 optical coupler, used to provide a reference optical signal, and the other optical output end is connected to the optical input end of the intensity modulator, used to provide a carrier optical signal;

[0033] The intensity modulator has one electrical input terminal, one optical input terminal and one optical output terminal. The electrical input terminal is used to receive an external signal to be convolved. The signal to be convolved intensity-modulates the carrier optical signal input to the modulator through the intensity modulator to obtain an intensity-modulated optical signal. The optical output terminal is connected to the optical input terminal of the second 1×2 optical coupler. The second 1×2 optical coupler divides the intensity-modulated optical signal into two paths, and uses the two optical output terminals of the second 1×2 optical coupler as the two sub-output terminals of the intensity modulator. Among them, one sub-output terminal of the intensity modulator, that is, one optical output terminal of the second 1×2 optical coupler, is connected to the optical input terminal of the first first-order delay waveguide, which is used to realize the first first-order delay of the intensity-modulated optical signal to obtain the first first-order delay intensity-modulated optical signal. The other sub-output terminal of the intensity modulator, that is, the other optical output terminal of the second 1×2 optical coupler, is connected to the optical input terminal of the first coherent delay weighting unit, which is used to realize the convolution kernel matrix coefficient weighting and coherent summation of the intensity-modulated optical signal through the first coherent delay weighting unit to obtain the first first-order coherent summation optical signal. Similarly, the optical output terminals of the first K-2 first-order delay waveguides are connected to the optical input terminal of an optical coupler, and the two optical output terminals of the optical coupler are used as the two sub-output terminals connected to the corresponding first K-2 first-order delay waveguides. The K-1 first-order delay waveguides are connected in series through one sub-output terminal, and the other sub-output terminals of the first K-2 first-order delay waveguides and the output terminal of the last first-order delay waveguide are sequentially connected to the input terminals of the K-1 coherent delay weighting units. Taking the connection of the first first-order delay waveguide as an example, specifically, the optical output terminal of the first first-order delay waveguide is connected to the optical input terminal of the third 1×2 optical coupler. One optical output terminal of the third 1×2 optical coupler is connected to the optical input terminal of the second first-order delay waveguide, which is used to realize the second first-order delay of the intensity-modulated optical signal to obtain the second first-order delay intensity-modulated optical signal. The other optical output terminal of the third 1×2 optical coupler is connected to the optical input terminal of the second coherent delay weighting unit, which is used to realize the convolution kernel matrix coefficient weighting and coherent summation of the intensity-modulated optical signal through the second coherent delay weighting unit to obtain the second first-order coherent summation optical signal, and so on. Finally, K first-order coherent summation optical signals are obtained at the optical output terminals of all the coherent delay weighting units. The first-order coherent summation optical signals are fused in the optical domain through multiple 2×1 optical couplers to obtain a second-order coherent summation optical signal. The second-order coherent summation optical signal is sent to the other optical input terminal of the first 2×1 optical coupler to be fused with the reference optical signal in the optical domain to obtain a coherent summation optical signal. The optical output terminal of the first 2×1 optical coupler is the optical output terminal of the entire chip, and the coherent summation optical signal is output through the optical output terminal of the chip.

[0034] The schematic structural diagram of the coherent delay weighting unit is as Figure 2As shown, it consists of L amplitude controllers (IM), L phase shifters (PS) (L = 4 is shown in the figure), a signal distribution unit and a signal summation unit. Among them, the signal distribution unit is a cascaded one-to-two tree structure, including 1 input terminal and L output terminals, and is composed of L - 1 1×2 optical couplers and L - 1 delay waveguides respectively located at one output terminal of the 1×2 optical couplers, realizing the first-level time interleaving and distribution of the intensity-modulated optical signal, and obtaining L time-interleaved sub-intensity-modulated optical signals; the L time-interleaved sub-intensity-modulated optical signals are respectively input into the corresponding amplitude controllers. The L amplitude controllers and the L phase shifters are in one-to-one correspondence. The output terminal of the amplitude controller is connected to the input terminal of the corresponding phase shifter. The convolution kernel matrix control signal controls the L amplitude controllers and the L phase shifters to weight the L time-interleaved sub-intensity-modulated optical signals according to the convolution kernel matrix coefficients respectively, obtaining L time-interleaved sub-weighted modulated optical signals; the signal summation unit is a cascaded two-to-one tree structure, composed of L - 1 2×1 optical couplers, including L input terminals and 1 output terminal. The L input terminals are connected to the output terminals of the L phase shifters in one-to-one correspondence. The output terminal of the signal summation unit is the output terminal of the coherent delay weighting unit. The signal summation unit realizes the coherent summation of the L time-interleaved sub-weighted modulated optical signals, obtaining the first-level coherent summation optical signal; among them, L = 2 Z , where Z is the number of stages of the tree structures of the signal distribution unit and the signal receiving unit (Z = 2 is shown in the figure);

[0035] The chip can be integrated based on the III-V material integration process or the silicon-based integration process.

[0036] A convolution operation application system based on a two-dimensional photon coherent convolution operation chip, a specific embodiment is as Figure 3 shown, which includes: the above two-dimensional photon coherent convolution operation chip, a light source, a signal source to be convolved, a convolution kernel matrix control unit, an optical amplifier, a photodetector, and an acquisition and processing unit.

[0037] First, a light source generates an optical signal and sends it into an intensity modulator through the optical input end of a two-dimensional photonic coherent convolution operation chip. The convolution signal to be convolved output by the convolution signal source is sent into the intensity modulator through the electrical input end of the two-dimensional photonic coherent convolution operation chip. The convolution signal to be convolved is loaded onto the optical signal through the intensity modulator to obtain an intensity-modulated optical signal. Among them, the convolution signal to be convolved is a one-dimensional time series obtained after the two-dimensional data to be convolved is processed by matrix flattening; the intensity-modulated optical signal is sent into a second 1×2 optical coupler and the intensity-modulated optical signal is divided into two paths. One path is sent into a series of first-stage delay waveguides to perform first-stage delay respectively to obtain a first-stage delay intensity-modulated optical signal, and the other path is sent into a first coherent delay weighting unit to realize the convolution kernel matrix coefficient weighting and coherent summation of the intensity-modulated optical signal to obtain a first-stage coherent summation optical signal; the first-stage delay intensity-modulated optical signals delayed by each first-stage delay waveguide are sequentially sent into a second coherent delay weighting unit to a fourth coherent delay weighting unit to realize the convolution kernel matrix coefficient weighting and coherent summation of the intensity-modulated optical signal to obtain a first-stage coherent summation optical signal. As shown in the figure, finally, four first-stage coherent summation optical signals are obtained at the optical output ends of the 4 coherent delay weighting units; the four first-stage coherent summation optical signals are fused in the optical domain through three 2×1 optical couplers to obtain a second-stage coherent summation optical signal. The second-stage coherent summation optical signal is sent into another optical input end of the first 2×1 optical coupler to be fused with the reference optical signal in the optical domain to obtain a coherent summation optical signal. The optical output end of the first 2×1 optical coupler is the optical output end of the entire chip, and the coherent summation optical signal is output through the optical output end of the chip; the output coherent summation optical signal is amplified by an optical amplifier and then photoelectric detection is completed through a photodetector to obtain an electrical output signal. This signal can be reconstructed through acquisition processing to obtain the characteristic signal after the two-dimensional convolution operation of the convolution signal to be convolved.

[0038] For the convenience of public understanding, the technical solution of the present invention will be further described in detail through a specific embodiment below:

[0039] First, a light source outputs an optical signal with a wavelength of λ, and the signal intensity of the optical signal is A. The optical signal is sent into the optical input end of the first 1×2 optical coupler of the two-dimensional photonic coherent convolution operation chip through the fiber-optic waveguide coupling technology. One optical output end of the first 1×2 optical coupler is connected to one optical input end of the first 2×1 optical coupler to provide a reference optical signal, and the other optical output end is connected to the optical input end of the intensity modulator. The convolution signal to be convolved output by the convolution signal source modulates the optical signal through the intensity modulator, and the convolution signal to be convolved is respectively loaded onto the optical signal. The convolution signal sequence can be expressed as x(i) = [x(1), x(2), x(3), …, x(R)], where i represents the discrete time serial number, R = QP is the length of the convolution signal to be convolved, the convolution signal to be convolved is a one-dimensional signal obtained after the two-dimensional data to be convolved is processed by matrix flattening, and the two-dimensional data to be convolved is obtained by matrix transformation of the original two-dimensional data. The transformation process is asFigure 4 As shown, the original two-dimensional data A Q×O is slid and segmented in the column direction in steps of P - N + 1 into H sub two-dimensional data B Q×P , and each sub two-dimensional data is a two-dimensional signal to be convolved. Among them, Q is the number of rows of the original two-dimensional data, O is the number of columns of the original two-dimensional data, P is the number of columns of the two-dimensional signal matrix to be convolved, N is the number of columns of the two-dimensional convolution kernel matrix, and in this embodiment, N = 4. The two-dimensional signal to be convolved is as shown in A in Figure 5 , and it is a matrix with Q rows and P columns. The specific operation of matrix flattening is to convert a two-dimensional or multi-dimensional matrix into a one-dimensional matrix, and the process is as shown in B in Figure 5 . The intensity-modulated optical signal is obtained. The intensity-modulated optical signal S Mod can be represented by a matrix as follows:

[0040] S Mod = Ax(i) = [Ax(1) Ax(2) … Ax(R)] 1×R (1)

[0041] The time series and wavelength relationship diagram of the corresponding intensity-modulated optical signal is as shown in Figure 6 . The intensity-modulated optical signal output by the intensity modulator is sent to the optical input end of the second 1×2 optical coupler. The second 1×2 optical coupler divides the intensity-modulated optical signal into two paths. One optical output end of the second 1×2 optical coupler is connected to the optical input end of the first first-order delay waveguide. The first first-order delay waveguide realizes the first first-order delay of the intensity-modulated optical signal to obtain the first first-order delay intensity-modulated optical signal. The length of the first-order delay waveguide is ΔL = (P - 3)cΔt / n w , where P is the number of columns of the two-dimensional signal matrix to be convolved; the other optical output end is connected to the optical input end of the first coherent delay weighting unit. Through the first coherent delay weighting unit, the convolution kernel matrix coefficient weighting and coherent summation of the intensity-modulated optical signal are realized to obtain the first first-order coherent summation optical signal. The coherent delay weighting unit is composed of 4 amplitude controllers, 4 phase shifters, 1 first-order delay waveguide, 2 second-order delay waveguides, 3 1×2 optical couplers and 3 2×1 optical couplers; among them, 3 1×2 optical couplers and 3 delay waveguides form a signal distribution unit, which includes 1 optical input end and 4 optical output ends, and realizes the first-order time interleaving and distribution of the intensity-modulated optical signal to obtain 4 time-interleaved sub-intensity-modulated optical signals; the convolution kernel matrix control signal controls 4 amplitude controllers and 4 phase shifters to weight the 4 time-interleaved sub-intensity-modulated optical signals according to the convolution kernel matrix coefficients respectively to obtain 4 time-interleaved sub-weighted modulated optical signals. 3 2×1 optical couplers form a signal summation unit, which realizes the coherent summation of 4 time-interleaved sub-weighted modulated optical signals to obtain the first-order coherent summation optical signal. The time series and relationship diagram of the weighted modulated optical signals output by the 4 phase shifters in the first coherent delay weighting unit and each phase shifter are as shown inFigure 7 As shown by A in [reference], the time series and wavelength relationship diagram of the first-level coherent summation optical signal obtained at the output end of the first coherent delay weighting unit is as follows Figure 7 As shown by B in [reference]. The first-level delay waveguide is a delay waveguide with a length of Δl1 = 2cΔt / n w The second-level delay waveguide is a delay waveguide with a length of Δl2 = cΔt / n w where c is the speed of light in vacuum, and n w is the effective refractive index of the waveguide delay line, Δt = 1 / S M is the duration of a single symbol of the signal to be convolved, and S M is the symbol rate of the signal to be convolved. According to the magnitude and positive / negative sign of the convolution kernel matrix coefficients, the transmission characteristics of the amplitude controller and the phase of the phase shifter are determined respectively. In each coherent delay weighting unit, the 4 amplitude controllers correspond to the coefficients of one row in the corresponding 4×4 convolution kernel matrix of the phase shifters. Let the convolution kernel matrix M con can be expressed as:

[0042]

[0043] w represents the absolute value of the elements of the convolution kernel matrix; let the phase shift amount M pm can be expressed as:

[0044]

[0045] S represents the magnitude of the phase shift amount, and its value is 0 or π, corresponding to the positive and negative signs of the elements of the convolution kernel matrix respectively.

[0046] Repeat the above connection method three times. Finally, four first-level coherent summation optical signals are obtained at the optical output ends of the 4 coherent delay weighting units. The time series and relationship diagram of the weighted modulation optical signals output by the 4 phase shifters in the 4th coherent delay weighting unit are as shown by Figure 7 C in [reference], and the time series and wavelength relationship diagram of the first-level coherent summation optical signal obtained at the output end of the 4th coherent delay weighting unit is as shown by Figure 7 D in [reference]. The 4 first-level coherent summation optical signals are fused in the optical domain through three 2×1 optical couplers to obtain a second-level coherent summation optical signal. The second-level coherent summation optical signal is fed into another optical input end of the first 2×1 optical coupler to be fused with the reference optical signal in the optical domain to obtain a coherent summation optical signal. The time series and wavelength relationship diagram of the 4 first-level coherent summation optical signals output by the 4 coherent delay weighting units are as shown by Figure 7 E in [reference], and the time series and wavelength relationship diagram of the coherent summation optical signal is as shown by Figure 7As shown in F. The optical output end of the first 2×1 optical coupler is the optical output end of the entire chip, and the coherent sum optical signal is output through the optical output end of the chip; the output coherent sum optical signal is amplified by an optical amplifier and then photoelectrically detected by a photodetector to obtain an electrical output signal:

[0047]

[0048] Among them, S ca (r) is the result of the rth convolution operation, w mn is the convolution kernel matrix coefficient, S mn is the phase shift. After the acquisition processing unit acquires the signal, it can realize two-dimensional reconstruction of the signal in the digital domain by using the opposite method of matrix flattening to process the effective time series signal. The two-dimensional reconstructed data is as follows: Figure 5 As shown in C in FIG. 1 , the three gray columns are redundant data. After removing the redundant data, the two-dimensional feature signal after the two-dimensional convolution signal completes the convolution operation can be obtained. The above process is a specific embodiment description performed when the original data is not padded with zeros. When the original data is padded with zeros, the zero-padded data can be used as the original two-dimensional data for the same operation as above.

[0049] Finally, the H two-dimensional feature signals are passed through Figure 8 The method shown combines the original two-dimensional data into a feature signal corresponding to the original two-dimensional data, and then completes the convolution operation of the original two-dimensional data.

[0050] Finally, it should be noted that the above examples are only specific embodiments of the present invention. The present invention is not limited to the above examples, and many variations are possible. All variations that can be directly derived or associated with the content disclosed by a person skilled in the art should be considered as the protection scope of the present invention.

Claims

1. A two-dimensional photon coherent convolution operation chip, characterized in that, The chip is integrated with an intensity modulator, K coherent delay weighting units, and K-1 first-order delay waveguides; wherein: The optical input end of the entire chip is an optical input end of the first 1×2 optical coupler, which is used to receive an external optical signal; an optical output end of the first 1×2 optical coupler is connected to an optical input end of the first 2×1 optical coupler, which is used to provide a reference optical signal, and the other optical output end is connected to the optical input end of the intensity modulator, which is used to provide a carrier optical signal; The electrical input end of the intensity modulator is used to receive an external signal to be convolved, and the signal to be convolved intensity modulates the optical signal input to the intensity modulator through the intensity modulator to obtain an intensity modulated optical signal; the optical output ends of the intensity modulator and the first K-2 first-order delay waveguides are connected to the optical input end of an optical coupler, and the two optical output ends of the optical coupler are used as two sub-output ends connected to the corresponding intensity modulator and the first K-2 first-order delay waveguides, the intensity modulator and K-1 first-order delay waveguides are connected in series through a sub-output end, the other sub-output end of the intensity modulator is connected to the input end of the first coherent delay weighting unit, and the other sub-output end of the first K-2 first-order delay waveguides and the last one The output end of the delay waveguide is connected to the input end of K-1 coherent delay weighting units in sequence respectively; the coherent delay weighting unit implements the convolution kernel matrix coefficient weighting and coherent summation of the intensity modulated optical signal to obtain the corresponding first-order coherent sum optical signal; the K first-order coherent sum optical signals outputted from the optical output ends of the K coherent delay weighting units are fused in the optical domain through one or more optical couplers to obtain a second-order coherent sum optical signal, the second-order coherent sum optical signal is sent to the other optical input end of the first 2×1 optical coupler to be fused in the optical domain with the reference optical signal to obtain a coherent sum optical signal, the optical output end of the first 2×1 optical coupler is the optical output end of the entire chip, and the coherent sum optical signal is outputted through the optical output end of the chip; The coherent delay weighting unit consists of L amplitude controllers, L phase shifters, a signal distribution unit and a signal summation unit. Among them, the signal distribution unit is a one-to-two tree structure formed by cascading L-1 1×2 optical couplers and L-1 delay waveguides respectively located at one output end of the 1×2 optical couplers, including 1 input end and L output ends. The signal summation unit is a two-in-one tree structure formed by cascading L-1 2×1 optical couplers, including L input ends and 1 output end. The L output ends of the signal distribution unit, the L amplitude controllers, the L phase shifters and the input ends of the signal summation unit are connected in one-to-one correspondence in sequence. The L output ends of the signal summation unit obtain L time-interleaved sub-intensity modulation optical signals. The convolution kernel matrix control signal controls the L amplitude controllers and the L phase shifters to weight the L time-interleaved sub-intensity modulation optical signals according to the convolution kernel matrix coefficients respectively, obtaining L time-interleaved sub-weighted modulation optical signals. The signal summation unit realizes the coherent summation of the L time-interleaved sub-weighted modulation optical signals, obtaining a first-level coherent summation optical signal. Wherein, L = 2 Z , and Z is the number of levels of the tree structures of the signal distribution unit and the signal receiving unit.

2. The two-dimensional photon coherent convolution operation chip according to claim 1, wherein In the signal distribution unit of the coherent delay weighting unit, the lengths of the delay waveguides corresponding to different levels are Δlr = 2 (Z-z) cΔt / n w , where c is the speed of light in a vacuum, and n w is the effective refractive index of the waveguide delay line, Δt = 1 / S M is the duration of a single symbol of the signal to be convolved, and S M is the symbol rate of the signal to be convolved, z = 1, 2, 3, …, Z, corresponding to the levels of the tree structure of the signal distribution unit.

3. The two-dimensional photon coherent convolution operation chip according to claim 1, wherein The phase shifter is a thermally controlled phase shifter or an electrically controlled phase shifter; the amplitude controller is a Mach-Zehnder modulator, a micro-ring modulator, an optical amplifier or an optical attenuator.

4. A two-dimensional photon coherent convolution operation chip as claimed in claim 1, wherein The convolution kernel matrix control signal controls L amplitude controllers and L phase shifters to weight L time-interleaved sub-intensity modulated optical signals according to the convolution kernel matrix coefficients, specifically: The transmission characteristics of the amplitude controller and the phase of the phase shifter are determined according to the size and positive and negative signs of the convolution kernel matrix coefficients. The L amplitude controllers and the corresponding L phase shifters in each coherent delay weighted unit correspond to a row of coefficients in the convolution kernel matrix. The K×L amplitude controllers and the corresponding K×L phase shifters in the K coherent delay weighted units correspond to a two-dimensional convolution kernel matrix of size K×L.

5. A two-dimensional photon coherent convolution operation chip as described in claim 1, characterized in that, The length of the first-stage delay waveguide is ΔL = (P - L + 1)cΔt / n w , where P is the number of columns of the two-dimensional data matrix to be convolved.

6. The two-dimensional photon coherent convolution operation chip according to claim 1, characterized in that The chip is integrated based on III-V group material integration process or silicon-based integration process.

7. A two-dimensional photon coherent convolution operation chip as described in claim 1, wherein The intensity modulator is a Mach-Zehnder modulator or a micro-ring modulator.

8. The two-dimensional photon coherent convolution operation chip according to claim 1, wherein The signal to be convolved is a one-dimensional time series obtained after flattening the two-dimensional data to be convolved. The two-dimensional data to be convolved is obtained by matrix transformation of the original two-dimensional data. The specific transformation process is: Original two-dimensional data A Q×O Slide and divide it into H sub two-dimensional data B in the column direction with a step of P - L + 1 Q×P , and each sub two-dimensional data is a two-dimensional data to be convolved. Here, Q is the number of rows of the original two-dimensional data, O is the number of columns of the original two-dimensional data, and P is the number of columns of the two-dimensional data matrix to be convolved.

9. The two-dimensional photon coherent convolution operation chip according to claim 8, wherein, The original two-dimensional data is obtained by decomposing three-dimensional or multi-dimensional original data.

10. A convolution operation application system for a two-dimensional photon coherent convolution operation chip, characterized in that, include: A two-dimensional photonic coherent convolution operation chip, a light source, a signal source to be convolved, a convolution kernel control unit, an optical amplifier, a photodetector, and an acquisition and processing unit according to any one of claims 1-9; wherein, the output end of the signal source to be convolved is connected to the electrical input end of the two-dimensional photonic coherent convolution operation chip, the optical output end of the light source is connected to the optical input end of the two-dimensional photonic coherent convolution operation chip, and the convolution kernel control unit is used to respectively determine the transmission coefficient of the control amplitude controller and the phase of the phase shifter according to the magnitude and positive or negative sign of the convolution kernel matrix coefficient; the input end of the optical amplifier is connected to the optical output end of the two-dimensional photonic coherent convolution operation chip and is used to amplify the coherent summation signal; the optical input end of the photodetector is connected to the optical output end of the optical amplifier and is used to complete the photoelectric conversion of the coherent summation signal; the acquisition and processing unit is connected to the electrical output end of the photodetector and is used to reconstruct the electrical output signal to obtain the characteristic signal after the two-dimensional convolution operation of the signal to be convolved.

Images