Vector dot product method and device based on pattern interference

Abstract

The invention provides a vector dot product method and device based on mode interference, and relates to the technical field of photoelectric information processing, and the method comprises the steps: obtaining an input optical signal; the input optical signal comprises a vector group; the vector group comprises a first vector and a second vector; the input optical signal corresponding to the second vector in the vector group is coupled with the input optical signal corresponding to the first vector after phase shifting, and a first output and a second output are obtained; performing phase shift on the output of one vector group, and coupling the output with the output of the other vector group to obtain a combined output; and inputting the combined output into a balanced light detector to obtain a first output result. According to the invention, the pi / 2 phase difference is introduced into the two coherent output arms of the dot product unit, so that the two coherent output arms can be directly subjected to power superposition instead of amplitude superposition through the beam combiner, and two-dimensional vector multiplication under single wavelength is realized. According to the scheme provided by the invention, half of the balance light detectors can be saved, so that the area cost is reduced, the post-processing time of the architecture is shortened, and the overall power consumption is optimized.

Description

Technical Field

[0001] The present application relates to the field of optoelectronic information processing technology, and in particular to a vector dot product method and device based on pattern interference. Background Art

[0002] Artificial intelligence (AI) is increasingly being integrated into various fields, including autonomous vehicles, smart buildings, and smart factories, placing enormous demands on computing power. Optical components can be used to perform data computations, such as any real or complex matrix and vector operations. Therefore, the advantages of optics in speed and power consumption can be exploited to develop diverse optical computing architectures. As a result, photonic integrated circuits are becoming increasingly effective tools for accelerating these computations in deep learning.

[0003] In related technologies, singular value decomposition or microring resonator (MRR) arrays are usually used to implement matrix-vector multiplication. Among them, MRR realizes matrix calculation through wavelength selectivity and resonant coupling, but its computing power is heavily dependent on the number of wavelength channels of wavelength division multiplexing, and its parallel scale is limited by the number of wavelengths. In contrast, methods based on singular value decomposition are usually implemented using Mach-Zehnder interferometers (MZIs). This method requires pre-calculating the elements of the matrix and then encoding them into the phase of the MZI's phase shifter. The input vector and output vector correspond to the amplitude of the input light and the output light, respectively.

[0004] However, the main problem with the traditional computing architecture using singular value decomposition is the complexity of pre-computation. The calculation of singular matrices (M×N) requires multi-layer MZI implementation, which not only wastes resources but also increases area costs. Summary of the Invention

[0005] The embodiments of the present application provide a vector dot product method and device based on pattern interference to solve the problems of high cost and complex implementation process when performing vector multiplication in the architecture of the prior art.

[0006] In a first aspect, an embodiment of the present application provides a vector dot product method based on pattern interference, including:

[0007] Acquire an input optical signal; the input optical signal includes a first number of vector groups; the vector groups include a first vector and a second vector; the first number is a dimension of the first vector or the second vector; wherein the input optical signal has a second number of wavelengths; the second number is a positive integer;

[0008] Phase-shifting the input optical signal corresponding to the second vector in the vector group and coupling it with the input optical signal corresponding to the first vector to obtain a first output and a second output;

[0009] After phase shifting the first output of the 2mth vector group, the first output is coupled with the first output of the 2m-1th vector group to obtain a first combined output; and after phase shifting the second output of the 2mth vector group, the second output is coupled with the second output of the 2m-1th vector group to obtain a second combined output; wherein m≥1;

[0010] The first combined output and the second combined output are respectively input into a third number of balanced light detectors to obtain a first output result; the third number is a ratio of the first number to twice the second number.

[0011] In some feasible embodiments, when the first number is equal to 2 and the second number is equal to 1, m=1.

[0012] In some feasible embodiments, when the first number is equal to 4 and the second number is equal to 2, m=1.

[0013] In some feasible embodiments, when the first number is equal to 4 and the second number is equal to 1, m=1 or 2.

[0014] In some feasible embodiments, the step of shifting the phase of the input optical signal corresponding to the second vector in the vector group includes:

[0015] The waveguide length of the input optical signal corresponding to the second vector is adjusted so that the waveguide length difference between the second vector and the first vector satisfies: the phase difference between the first vector and the second vector is π / 2.

[0016] In some feasible embodiments, the step of phase-shifting the first output and the second output of the 2mth vector group includes:

[0017] Adjust the waveguide lengths of the input optical signals corresponding to the first vector and the second vector in the 2m-th vector group so that the phase difference between the first vector in the 2m-th vector group and the first vector in the 2m-1-th vector group is π / 2, and the phase difference between the second vector in the 2m-th vector group and the second vector in the 2m-1-th vector group is π / 2.

[0018] In a second aspect, the present application provides a vector dot product device based on pattern interference, the device comprising:

[0019] An input unit, configured to obtain an input optical signal; the input optical signal comprises a vector group; the vector group comprises a first vector and a second vector;

[0020] a coupling unit, configured to phase-shift an input optical signal corresponding to the second vector in the vector group and couple it with the input optical signal corresponding to the first vector to obtain a first output and a second output; and

[0021] Phase-shifting the first output of the 2mth vector group for coupling with the first output of the 2m-1th vector group; and phase-shifting the second output of the 2mth vector group for coupling with the second output of the 2m-1th vector group; wherein m≥1;

[0022] a beam combiner, configured to couple the phase-shifted first output of the 2mth vector group with the first output of the 2m-1th vector group to obtain a first combined output; and to couple the phase-shifted second output of the 2mth vector group with the second output of the 2m-1th vector group to obtain a second combined output;

[0023] The balanced optical detector is configured to obtain a first output result according to the first combined output and the second combined output.

[0024] In some feasible embodiments, the vector group has a first number; the first number is the dimension of the first vector or the second vector; wherein the input optical signal has a second number of wavelengths; the second number is a positive integer; the balanced optical detector has a third number, and the third number is the ratio of the first number to twice the second number.

[0025] In some feasible embodiments, when the first number is equal to 2 and the second number is equal to 1, the number of the coupling units and the combiners is equal to the first number, and m=1.

[0026] In some feasible embodiments, when the first number is equal to 4 and the second number is equal to 2, the number of the coupling units and the combiners is equal to half of the first number, and m=1.

[0027] This application proposes a method for expanding the dimensionality of input vectors at a given wavelength by introducing a π / 2 phase difference between the two coherent output arms of a conventional dot product unit. This allows for direct power addition rather than amplitude addition via a beam combiner, thereby achieving two-dimensional vector multiplication at a single wavelength. This proposed solution can save half the number of balanced photodetectors, thereby reducing area costs, shortening architecture post-processing time, and optimizing overall power consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A schematic diagram of the basic structure of a dot product unit provided in some embodiments of the present application;

[0029] Figure 2 A flowchart of a vector dot product method based on pattern interference provided in some embodiments of the present application;

[0030] Figure 3A schematic diagram of an embodiment of the method provided in this application. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0032] Deep learning is revolutionizing a wide range of scientific fields and industries, bringing with it heavy computational workloads. This has driven photonic integrated circuits (PICs) to become an efficient tool for accelerating matrix multiplication in deep learning. However, PICs typically face significant area overhead. Wavelength division multiplexing (WDM) is often used to improve computational efficiency or reduce area costs. However, the limited number of wavelengths available in conventional bands restricts the application of large-scale optical computing.

[0033] See also Figure 1 , which is a schematic diagram of the basic structure of the dot product unit provided in some embodiments of the present application.

[0034] Depend on Figure 1 As shown, in related art, the basic structure for implementing vector dot product calculations can include a directional coupler (DC) and a pair of balanced photodetectors (BPDs) connected in series. Wavelength division multiplexing technology can be used to implement dot product calculations of any two vectors, wherein the two balanced photodetectors are connected in series between a power supply voltage VDD and ground. The balanced photodetectors are used to calculate the photocurrent value (I0 = I1 + I2) based on the light intensity and phase corresponding to the first and second vectors input by the directional coupler, and determine the vector product result based on the current value.

[0035] In the embodiments of the present application, a directional coupler (DC) is a passive device widely used in radio frequency, microwave or optical communication systems. Its core function is to divide the power of the input signal into two paths: one part as a direct output (main path) and the other part as a coupled output (secondary path), thereby realizing signal monitoring, power distribution or reflection measurement. For example, in wireless communication base stations or network analyzers, it is often used to measure signal strength, isolate interference or implement feedback control, which helps to improve system performance and stability. Its design is usually based on coupled transmission lines or waveguide structures, which can achieve low insertion loss and high directivity, and is a key component of modern communication equipment.

[0036] A balanced photodetector (BPD) is a receiving device specifically designed for optical communications and photoelectric sensing. By simultaneously measuring the difference between the outputs of two input optical signals (i.e., the differential signal), it effectively suppresses common-mode noise (such as laser intensity noise or environmental interference), significantly improving detection sensitivity and signal-to-noise ratio. In practical applications, such as coherent optical communication systems or quantum key distribution, BPDs enable high-precision signal demodulation and capture of weak optical signals, supporting high-speed data transmission and precision measurement. Their structure typically combines a photodiode and a differential amplifier for fast response and low-distortion operation.

[0037] Specifically, in some examples, Figure 1 When the basic structure of the dot product unit is used for matrix multiplication and vector multiplication, taking the matrix multiplication X×Y=Z as an example, the element in the product matrix Z of matrix X and matrix Y is Each element of the matrix Z is a row vector x of the matrix X i and the column vector y of matrix Y j The dot product of matrices X and Y corresponds to the element x ik and y kj The phase and amplitude of the input light to the directional coupler can be encoded. A phase shifter PS is added to one of the input arms of the directional coupler to produce a -90° phase shift. The 2×2 directional coupler provides interference between the coherent light inputs of the two arms. For an ideal directional coupler, the transfer matrix of this structure can be expressed as:

[0038]

[0039] For the input data (x ik y kj ) T , the output light amplitude after the directional coupler can be expressed as:

[0040]

[0041] The photocurrent of the PD connected to the directional coupler is proportional to the received optical power. If cascaded PDs with the same response rate are used, the output current I out for:

[0042] I0∝|E1| 2 -|E2| 2 ∝|x ik +y kj | 2 -|x ik -y kj | 2 ∝x ik y kj ;

[0043] For wavelength division multiplexing signals, each wavelength corresponds to a (xik y kj ) T , and generates an output current in the BPD based on the dot product unit structure, the current is proportional to x ik y kj There is no interference between signals of different wavelengths, and the final output photocurrent is the sum of the photocurrents of each component, which is proportional to Therefore, using this structure, we can realize the vector x i with y j The dot product of , where the number of elements of the vector depends on the number of wavelengths used in wavelength division multiplexing.

[0044] In the basic structure of the above-mentioned dot product unit, since one of the balanced photodetectors BPD corresponds to an input light of one wavelength, as the number of wavelengths increases, the number of introduced balanced photodetectors will inevitably increase. Considering that too many wavelengths will bring higher costs and errors, the present application provides an improved architecture of the above-mentioned dot product unit, which can expand the scale of vectors that can be multiplied under a fixed number of wavelengths.

[0045] See also Figure 2 , which is a flow chart of a vector dot product method based on pattern interference provided in some embodiments of the present application;

[0046] Depend on Figure 2 It can be seen that the method provided in this application includes:

[0047] S100: Acquire an input optical signal; the input optical signal includes a first number of vector groups; the vector groups include a first vector and a second vector; the first number is a dimension of the first vector or the second vector; wherein the input optical signal has a second number of wavelengths; the second number is a positive integer;

[0048] In an embodiment of the present application, the input optical signal obtained can be an optical signal whose phase and amplitude are obtained by directly encoding the elements in the matrix to be multiplied; or it can be an optical signal first obtained by a grating coupler, and then the obtained optical signal is split and modulated to obtain each element in the matrix, and finally the corresponding optical signals are coupled to form an optical signal that can represent the vector group.

[0049] It should be noted that in some examples of the present application, the wavelength of the input optical signal may be one, that is, the second number is equal to 1. In this case, the method of the present application is used to perform vector multiplication operations using input optical signals of the same wavelength; in other examples, the wavelength of the input optical signal may be more than one, that is, the second number is greater than 1. In this case, the method of the present application is used to perform vector multiplication operations using input optical signals of multiple wavelengths.

[0050] At the same time, in some examples of the present application, when the dimension of the first vector or the second vector is both 2, it means the multiplication of two two-dimensional vectors, and in this case, the first quantity is equal to 2; in other examples, when the dimension of the first vector or the second vector is both 4, it means the multiplication of two four-dimensional vectors, and in this case, the first quantity is equal to 4.

[0051] As in the previous basic structure, after acquiring the input light signal, in order to make the two input lights interfere, the following steps need to be performed on each vector group separately:

[0052] S200: Phase-shifting an input optical signal corresponding to the second vector in the vector group and coupling it with the input optical signal corresponding to the first vector to obtain a first output and a second output;

[0053] Two sets of input with the same wavelength (x1 y1) T , (x2 y2) T For example, the corresponding two sets of outputs are Consider adding two sets of outputs with the same wavelength in the optical domain. If a beam combiner is used directly to combine the two, the amplitude after the combination is a linear superposition of the two, which cannot produce a correct output. Therefore, in this embodiment of the application, it is necessary to continue executing step S300 before combining.

[0054] S300: After phase shifting the first output of the 2mth vector group, coupling the output with the first output of the 2m-1th vector group to obtain a first combined output; and after phase shifting the second output of the 2mth vector group, coupling the output with the second output of the 2m-1th vector group to obtain a second combined output; wherein m≥1;

[0055] It should be noted that as the dimension of the vector increases, the process of step S300 can be performed in parallel on different vector groups by increasing the number of vector groups. At this time, multiple vector groups can be divided into odd groups and even groups. For each adjacent odd group and even group, step S300 can be performed once to obtain a group of optical signals for subsequent merging.

[0056] Still using the two sets of inputs (x1 y1) T , (x2 y2) T For example, at this time, the two sets of outputs based on the basic architecture of the dot product structure introduce a π / 2 phase difference, and then the output light amplitude obtained by merging in step S300 is

[0057] S400: Inputting the first combined output and the second combined output into a third number of balanced light detectors respectively to obtain a first output result; the third number is a ratio of the first number to twice the second number.

[0058] In an embodiment of the present application, the two outputs (the first combined output and the second combined output) obtained after steps S100-S300 can be used as two inputs of a balanced photodetector, and then by cascading the properties of the same response rate in the balanced photodetector, a first output result is finally output, wherein the first output result can represent the product of the vectors in the form of current.

[0059] Among them, the output current I obtained in the above example is out for:

[0060] I0∝|E1| 2 -|E2| 2 ∝|(x1+y1)+j(x2+y2)| 2 -|j(x1-y1)-(x2-y2)| 2 ∝x1y1+x2y2;

[0061] It can be seen from the above technical solution that, when the number of wavelengths is constant, the method of the present application is adopted to introduce a phase difference of π / 2 into the two sets of outputs of the input optical signal before inputting into the balanced light detector, so that the outputs can continue to be merged, and the original method of directly using the output results of the balanced light detector is replaced by adjusting the phase and combining the beam, thereby reducing the number of balanced light detectors. By adopting the method provided in the embodiment of the present application, passive devices (phase shifting, beam combining devices) can be used instead of active devices (balanced light detectors). Compared with the traditional architecture, half of the balanced light detectors can be saved, thereby greatly reducing the area cost and power consumption cost.

[0062] See also Figure 3 , is a schematic diagram of a method provided in this application under an embodiment;

[0063] based on Figure 3 , the following examples illustrate different scenarios in which the method of this application is applied:

[0064] Scenario 1: The wavelength of the input optical signal is 1, and the dimension of the vector is 2, that is, the first quantity is equal to 2 and the second quantity is equal to 1. At this time, based on Figure 3 , the input light signal corresponds to two vector groups (x1 y1) T , (x2 y2) T, where the elements of the first vector are x1 and x2, and the elements of the second vector are y1 and y2. According to the aforementioned method, a -90° phase shift operation is first performed on the second vector, and then coupled with the first vector respectively to obtain a first output and a second output. Then, based on the number of vector groups, m=1 is determined, and a -90° phase shift operation is first performed on the output corresponding to the second vector group, and then combined with the output corresponding to the first vector group through a combiner to finally obtain a first combined output and a second combined output. The first combined output and the second combined output are input into two input terminals of a balanced photodetector to obtain an output current representing the vector product.

[0065] Therefore, scenario 1 uses an input optical signal with 1 wavelength to perform 2D vector calculation, and the number of balanced optical detectors used is 2 / 1·2=1.

[0066] Scenario 2: The wavelength of the input optical signal is 2, and the dimension of the vector is 4, that is, the first number is equal to 4 and the second number is equal to 2. At this time, based on Figure 3 , the input optical signal can still correspond to two vector groups (x1 y1) T , (x2 y2) T Unlike scenario 1, the first vector's elements x1 and x2 can now include two elements of different wavelengths, indicating that the x1 vector can be derived by coupling optical signals from different wavelengths. Similarly, the second vector's elements y1 and y2 can also include two elements of different wavelengths. Based on this, the subsequent steps are the same as in scenario 1. Since the elements performing phase shifting and beam combining remain unchanged, the corresponding first and second combined outputs are still obtained.

[0067] Therefore, scenario 2 uses a 2-wavelength input optical signal to perform 4-dimensional vector calculation, and the number of balanced optical detectors used is 4 / 2·2=1.

[0068] In scenario three, the wavelength of the input optical signal is 1, and the dimension of the vector is 4. That is, when the first quantity is 4 and the second quantity is 1, the vector groups corresponding to the input optical signal change from two to four. For example, (x1 y1) T , (x2 y2) T , (x3 y3) T , (x4 y4) T , m = 2 is determined based on the vector group; similar to the method in scenario 1, when performing single-wavelength calculation of four-dimensional vectors, the process can be regarded as two parallel calculations of scenario 1, that is, the architecture at this time is equivalent to Figure 3 The same architecture is stacked below the architecture shown. At this time, the method in scenario 1 can be used to perform the operations on (x1 y1) and (x1 y1) respectively. T , (x2y2) TThe calculation process, as well as, perform the (x3 y3) T , (x4 y4) T The calculation results of the two calculation processes correspond to the input of a balanced light detector, that is, the number of balanced light detectors is 2.

[0069] Therefore, scenario three uses an input optical signal with 1 wavelength to perform 4-dimensional vector calculation, and the number of balanced optical detectors used is 4 / 1·2=2.

[0070] From the above three scenarios, it can be concluded that when performing vector multiplication using the method and architecture provided in this application, the relationship between the number of wavelengths n, the vector dimension N, and the number of balanced light detectors s is: s = N / 2n.

[0071] It should be understood that in the embodiments of the present application, the above three scenarios are only used as examples. When one of the parameters such as wavelength and vector dimension increases, any of the above three scenarios can also be used for explanation. In any new scenario that is expanded, the number of balanced light detectors satisfies the above relationship, which will not be repeated here.

[0072] Furthermore, in some embodiments, in the aforementioned step S200, the phase shifting operation performed on the second vector may be performed by a phase shifter, or may be performed by adjusting a waveguide length difference without using a phase shifter. Therefore, the step of phase shifting the input optical signal corresponding to the second vector in the vector group includes:

[0073] S210: Adjust the waveguide length of the input optical signal corresponding to the second vector so that the waveguide length difference between the second vector and the first vector satisfies: the phase difference between the first vector and the second vector is π / 2.

[0074] Furthermore, in some embodiments, in the aforementioned step S300, the step of phase-shifting the first output and the second output of the 2m-th vector group includes:

[0075] Adjust the waveguide lengths of the input optical signals corresponding to the first vector and the second vector in the 2m-th vector group so that the phase difference between the first vector in the 2m-th vector group and the first vector in the 2m-1-th vector group is π / 2, and the phase difference between the second vector in the 2m-th vector group and the second vector in the 2m-1-th vector group is π / 2.

[0076] In the embodiment of the present application, by adjusting the waveguide length to obtain the phase difference, the composition of the architecture can be further reduced and the area cost can be lowered.

[0077] As can be seen from the above technical solution, the vector dot product method based on pattern interference proposed in this application proposes a method for expanding the dimensionality of the input vector under the condition of a certain wavelength. By introducing a π / 2 phase difference between the two coherent output arms of the traditional dot product unit, it is possible to directly perform power addition rather than amplitude addition through a beam combiner, thereby realizing two-dimensional vector multiplication at a single wavelength. The solution provided by this application can save half the number of balanced photodetectors, thereby reducing area costs, shortening architecture post-processing time, and optimizing overall power consumption.

[0078] Corresponding to the method provided in any of the foregoing embodiments, the present application further provides a vector dot product device based on pattern interference, the device comprising:

[0079] An input unit, configured to obtain an input optical signal; the input optical signal comprises a vector group; the vector group comprises a first vector and a second vector;

[0080] a coupling unit, configured to phase-shift an input optical signal corresponding to the second vector in the vector group and couple it with the input optical signal corresponding to the first vector to obtain a first output and a second output; and

[0081] Phase-shifting the first output of the 2mth vector group for coupling with the first output of the 2m-1th vector group; and phase-shifting the second output of the 2mth vector group for coupling with the second output of the 2m-1th vector group; wherein m≥1;

[0082] a beam combiner, configured to couple the phase-shifted first output of the 2mth vector group with the first output of the 2m-1th vector group to obtain a first combined output; and to couple the phase-shifted second output of the 2mth vector group with the second output of the 2m-1th vector group to obtain a second combined output;

[0083] The balanced optical detector is configured to obtain a first output result according to the first combined output and the second combined output.

[0084] Furthermore, in some feasible embodiments, the vector group has a first number; the first number is the dimension of the first vector or the second vector; wherein the input optical signal has a second number of wavelengths; the second number is a positive integer; and the balanced optical detector has a third number, which is the ratio of the first number to twice the second number.

[0085] Furthermore, in some feasible embodiments, when the first number is equal to 2 and the second number is equal to 1, the number of the coupling units and the combiners is equal to the first number, and m=1.

[0086] Furthermore, in some feasible embodiments, when the first number is equal to 4 and the second number is equal to 2, the number of the coupling units and the combiners is equal to half of the first number, and m=1.

[0087] The technical effects of the device provided in the embodiments of the present application in actual applications can be found in the description of any of the aforementioned embodiments and will not be repeated here.

[0088] It is easy to understand that those skilled in the art can combine, split, reorganize, etc. the embodiments of the present application based on the several embodiments provided in the present application to obtain other embodiments, and these embodiments do not exceed the scope of protection of the present application.

[0089] The above specific implementation methods further explain in detail the purpose, technical solutions and beneficial effects of the embodiments of the present application. It should be understood that the above are only specific implementation methods of the embodiments of the present application and are not intended to limit the scope of protection of the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the scope of protection of the embodiments of the present application.

Claims

1. A vector dot product method based on pattern interference, characterized in that: include: Acquire an input optical signal; the input optical signal includes a first number of vector groups; the vector groups include a first vector and a second vector; the first number is a dimension of the first vector or the second vector; wherein the input optical signal has a second number of wavelengths; the second number is a positive integer; Phase-shifting the input optical signal corresponding to the second vector in the vector group and coupling it with the input optical signal corresponding to the first vector to obtain a first output and a second output; After phase shifting the first output of the 2mth vector group, the first output is coupled with the first output of the 2m-1th vector group to obtain a first combined output; and after phase shifting the second output of the 2mth vector group, the second output is coupled with the second output of the 2m-1th vector group to obtain a second combined output; wherein m≥1; The first combined output and the second combined output are respectively input into a third number of balanced light detectors to obtain a first output result; the third number is a ratio of the first number to twice the second number.

2. The vector dot product method based on pattern interference according to claim 1, characterized in that: When the first number is equal to 2 and the second number is equal to 1, m=1.

3. The vector dot product method based on pattern interference according to claim 1, characterized in that: When the first number is equal to 4 and the second number is equal to 2, m=1.

4. The vector dot product method based on pattern interference according to claim 1, characterized in that: When the first number is equal to 4 and the second number is equal to 1, m=1 or 2.

5. The vector dot product method based on pattern interference according to any one of claims 1 to 4, characterized in that: The step of phase-shifting the input optical signal corresponding to the second vector in the vector group comprises: The waveguide length of the input optical signal corresponding to the second vector is adjusted so that the waveguide length difference between the second vector and the first vector satisfies: the phase difference between the first vector and the second vector is π / 2.

6. The vector dot product method based on pattern interference according to any one of claims 1 to 4, characterized in that: The step of phase shifting the first output and the second output of the 2mth vector group comprises: Adjust the waveguide lengths of the input optical signals corresponding to the first vector and the second vector in the 2m-th vector group so that the phase difference between the first vector in the 2m-th vector group and the first vector in the 2m-1-th vector group is π / 2, and the phase difference between the second vector in the 2m-th vector group and the second vector in the 2m-1-th vector group is π / 2.

7. A vector dot product device based on pattern interference, characterized in that: The device comprises: An input unit, configured to obtain an input optical signal; the input optical signal comprises a vector group; the vector group comprises a first vector and a second vector; a coupling unit, configured to phase-shift an input optical signal corresponding to the second vector in the vector group and couple it with the input optical signal corresponding to the first vector to obtain a first output and a second output; and Phase-shifting the first output of the 2mth vector group for coupling with the first output of the 2m-1th vector group; and phase-shifting the second output of the 2mth vector group for coupling with the second output of the 2m-1th vector group; wherein m≥1; a beam combiner, configured to couple the phase-shifted first output of the 2mth vector group with the first output of the 2m-1th vector group to obtain a first combined output; and to couple the phase-shifted second output of the 2mth vector group with the second output of the 2m-1th vector group to obtain a second combined output; The balanced optical detector is configured to obtain a first output result according to the first combined output and the second combined output.

8. The vector dot product device based on pattern interference according to claim 7, characterized in that: The vector group has a first number; the first number is the dimension of the first vector or the second vector; wherein the input optical signal has a second number of wavelengths; the second number is a positive integer; the balanced optical detector has a third number, and the third number is the ratio of the first number to twice the second number.

9. The vector dot product device based on pattern interference according to claim 8, characterized in that: When the first number is equal to 2 and the second number is equal to 1, the number of the coupling units and the combiners is equal to the first number, and m=1.

10. The vector dot product device based on pattern interference according to claim 8, characterized in that: When the first number is equal to 4 and the second number is equal to 2, the number of the coupling units and the combiners is equal to half of the first number, and m=1.

Images